Professional Documents
Culture Documents
https://doi.org/10.1007/s11831-022-09858-w
REVIEW ARTICLE
Received: 17 March 2022 / Accepted: 19 November 2022 / Published online: 3 December 2022
© The Author(s) under exclusive licence to International Center for Numerical Methods in Engineering (CIMNE) 2022
Abstract
In the developing world, parasites are responsible for causing several serious health problems, with relatively high infec-
tions in human beings. The traditional manual light microscopy process of parasite recognition remains the golden standard
approach for the diagnosis of parasitic species, but this approach is time-consuming, highly tedious, and also difficult to main-
tain consistency but essential in parasitological classification for carrying out several experimental observations. Therefore,
it is meaningful to apply deep learning to address these challenges. Convolution Neural Network and digital slide scanning
show promising results that can revolutionize the clinical parasitology laboratory by automating the process of classifica-
tion and detection of parasites. Image analysis using deep learning methods have the potential to achieve high efficiency and
accuracy. For this review, we have conducted a thorough investigation in the field of image detection and classification of
various parasites based on deep learning. Online databases and digital libraries such as ACM, IEEE, ScienceDirect, Springer,
and Wiley Online Library were searched to identify sufficient related paper collections. After screening of 200 research
papers, 70 of them met our filtering criteria, which became a part of this study. This paper presents a comprehensive review
of existing parasite classification and detection methods and models in chronological order, from traditional machine learning
based techniques to deep learning based techniques. In this review, we also demonstrate the summary of machine learning
and deep learning methods along with dataset details, evaluation metrics, methods limitations, and future scope over the
one decade. The majority of the technical publications from 2012 to the present have been examined and summarized. In
addition, we have discussed the future directions and challenges of parasites classification and detection to help researchers
in understanding the existing research gaps. Further, this review provides support to researchers who require an effective and
comprehensive understanding of deep learning development techniques, research, and future trends in the field of parasites
detection and classification.
1 Introduction
13
Vol.:(0123456789)
2014 S. Kumar et al.
helminths parasites globally causes infection in more than This review provides a broad survey of the most impor-
800 million people [4]. In developing countries including tant aspects of parasite microscopic image analysis using
East Asia, South America, Saharan Africa etc. there are deep learning. This review depicted in Fig. 1 is organized
more than 100 human intestinal parasite species that produce as follows. Section 2 describes the fundamentals concepts
200,000 eggs per day and annually 41,500 human deaths are of deep learning, which include convolution layers compo-
reported due to parasitic infections [2, 5]. Infection caused nents, methods & models and evaluation metrics. In Sect. 3,
by these diseases display a wide range of clinical explana- we discuss methodology used to conduct this review includ-
tions ranging from malnutrition to asymptomatic anaemia ing research questions, paper retrieval and filtering process.
and even cause cancer [6, 7]. Conventional light microscopic In Sects. 4 and 5, detailed gist of each machine learning
examination remains the gold standard method for diagnosis and deep learning method and model elaborated in selected
of several parasites diseases, such as malaria [8]. research papers has been provided respectively. Summary
There are numerous shortcomings of these methods, of each research paper and their future scope is presented
which have serious potential to sway the results of clini- in the form of tables. In Sect. 6 various results have been
cal examinations, such as variable sensitivity, resource and discussed that will provide research directions in parasites
time-consumption. Another drawback of the traditional image detection and classifications. Finally, In Sect. 7 con-
clinical parasitology classification and detection is main- clusion and future scope is presented.
taining staff competency and engagement. The clinical
parasitology laboratory also suffers from two major aspects
firstly, educated technologists increasingly gravitate toward 2 Background
technology-driven and automated disciplines of laboratory
and secondly, shortage of adequately trained personnel [9] The following subsections provide a succinct explanation
to keep them staffed. Results based on diagnosis of parasites of various concepts and technologies associated with the
are often obtained from clinical signs and symptoms that work presented in this paper. Section 2.1 provides an over-
are susceptible to human errors which may lead to higher view of Deep Learning, Sect. 2.2 discusses about CNNs
mortality and buying of unnecessary drugs leading to eco- and Sect. 2.3 provides a gist of CNN models and parasite
nomic burden [10] and despondency. In consequence, alter- datasets.
nate methods are required which help to generate quality
diagnosis results. To date, to the best of our knowledge, there 2.1 Deep Learning Overview
have been no significant technological advancements for the
detection of protozoa in human stool specimens using per- In this section, we present the analysis of deep learning
manently stained slides (e.g., trichrome, modified acid-fast, fundamentals that addresses wide range of issues including
and modified safranin). medical images detection, segmentation and classification.
The objective of this review is to establish the vital aspect The section begins with an introduction to deep learning,
of deep learning in parasite microscopic image classifica- then on to cover various techniques, and concludes with
tion and detection to make it easy for researchers to have a transfer learning techniques and models built on CNNs to
clear image of deep learning for parasitology from this sur- increase the effectiveness of automatic parasite identification
vey paper. Further, this systematic review of deep learning and categorization. Deep learning is a subfield of machine
models for microscopic images helps people discover more learning inspired by the structure and function of the human
about recent growth in this field. In this review, we analysed brain neural network. In medical diagnosis, deep learning
deep learning techniques applied in parasitological image methods and models operate on large amounts of medical
application and in turn made the following contributions:
13
Advances Towards Automatic Detection and Classification of Parasites Microscopic Images… 2015
image data to transfer the given dataset to some specific vital role in the automatic diagnosis of novel coronavirus
labels (Fig. 2). [15]. DL methods and models include Deep Feedforward
A large number of artificial neural network layers are used Neural Network (DFNN), Convolutional Neural Network
in the development and design of deep learning, with each (CNN) [16], Recurrent Neural Network (RNN) and Auto-
layer carrying out a different evaluation based on the infor- encoder. The overview of prominent DL-model architectures
mation it receives. [11, 12]. Conventional machine learning is depicted in Fig. 3, which can be used to enhance the effi-
algorithms employ various steps to achieve the classifica- ciency of automatic parasite classification and detection.
tion tasks like pre- processing, feature extraction, feature Due to it high accuracy, speed and flexibility deep learning
selection and learning. Conversely, Deep learning techniques is being applied in microscopic examination of parasite spe-
automate several tasks of machine learning algorithms [12, cies. The end-results may include classification, detection
13], as shown in Fig. 1. Deep learning gained popularity in and segmentation. Classification task is applied for recog-
recent years due to exponential growth of data [14]. Deep nition of multi-parasite species where microscopic images
learning methods and models have the potential to enhance contain mixed infections. Detection is used to acquire the
human lives with accuracy in diagnosis including pathogen position of individual parasites or similar group of parasites.
detection and classification such as malaria, intestinal para- Segmentation task’s objective is grouping similar regions or
site and tuberculosis, etc. In recent times i.e. after the onset segments of a parasitic image under their respective class.
of COVID-19 pandemic, Deep learning (DL) is playing a Classification and detection of microscopic parasites are
13
2016 S. Kumar et al.
usually performed by CNN and RCNN series models [48, of the input image. It is a technique to maintain the size
70]. of image that reduces as a result of strides. Consider the
input parasitic image of size H×H×C, suppose F implies
2.2 Convolutional Neural Networks (CNN) number of filters, S implies stride, and P for padding.
The following formula can used to determine the output
CNN structure perform computations based on the structure image size:
of human visual cortex and is a popular choice for automatic
H − F + 2P
extraction of relevant features from large amounts of data Iout = +1 (2)
S
[17]. A CNN is designed with a sequence of convolution
layers, sampling layers followed by fully connected layer as The output size of parasitic image will be
shown in Fig. 3a. Input layers of CNN model is arranged in Iout × Iout × Cout
the 3-dimensions (height, width, and depth), m × m × r. The b. Pooling Layer: Pooling is another important component
depth (r) represents the number of channels [14]. In each in CNN architecture to reduce the dimensionality of the
convolutional layer, several filters are applied and these fil- feature map. This approach shrinks the large joint fea-
ters are also organized in 3D, (n × n × q), but values of both ture representation size into small valuable information.
n and q are smaller than m and r respectively. The convolu- The same process of convolution layer is applied in the
tional layer evaluates the product of inputs and weights, as pooling layer. There are different types of pooling meth-
shown in Eq. 1. ods that are applied to different pooling layers. These
are min pooling, max pooling, average pooling, global
(h)k = f W k ∗ x + bk (1)
( )
average pooling, global max pooling and tree pooling.
Three familiar pooling operations are shown in Fig. 4.
To accelerate the training process and to handle the over-
c. Activation Function: Activation functions are nonlinear
fitting of a model, the down-sampling is applied to layers to
functions and neural networks utilize different activation
extract features. Finally, mid and low-level features are fed to
functions to map the input values to output. The input to
fully connected layers to obtain the high-level of abstraction.
activation evaluate by summation of the dot product of
weights and inputs along with bias.
2.2.1 CNN Layers
d. Loss Function: In CNN architecture the final classifica-
tion is obtained from the output layer. CNN model uti-
This subsection discusses each layer of CNN basic architec-
lizes several loss functions in the output layer to evaluate
ture along with their functions to deal with parasitic image
the error predicted during training. Next, this error is
classification and detection as elaborated below.
applied to optimize the CNN learning process. Literally,
the loss function is calculated by two parameters. First is
a. Convolution Layer: Every CNN architecture includes an
the predicted output of the CNN model and the second
important component called convolution layer and each
parameter is the actual output
CNN layer is comprised of several filters (kernels). Over
N-dimensional metrics of an input image, convolution
operations are performed with these kernels to obtain
output features. Firstly, the filters are convolved over the
entire input image horizontally and vertically during the
forwarding pass, then, the product of input images and
the filters is determined. It generates a two dimensional
visual of image that shows the output of filters at spatial
location of the image. The entire process is repeated for
input images until there is no possibility of sliding.
Stride is a component of convolutional neural net-
works that is streamlined for image data compression.
Stride is a CNN filter parameter that regulates the
amount of movement all over the image. For instance,
if the stride of a neural network is set to 1, the filter will
move one pixel at a time. Since the filter size influences
the encoded output volume, stride is frequently set to a
whole integer rather than a fraction or decimal. Padding
is also applied to determine the border size information
Fig. 4 Pooling layers operations
13
Advances Towards Automatic Detection and Classification of Parasites Microscopic Images… 2017
13
2018 S. Kumar et al.
AlexNet Used Dropout and ReLU 8 ImageNet 60 16.4 227 × 227 × 3 [17]
VGG Increased depth, small 19 ImageNet 138 7.3 224 × 224 × 3 Simonyan and Zisserman
filter size [18]
GoogLeNet Increased depth, different 22 ImageNet 4 6.7 224 × 224 × 3 Szegedy et al. [19]
filter size, concatenation
Inception-V3 Utilizes small filter size, 48 ImageNet 23.6 3.5 229 × 229 × 3 Szegedy et al. [20]
better feature represen-
tation
Inception-V4 Divided transform & 70 ImageNet 35 3.08 229 × 229 × 3 Szegedy et al. [21]
integration concepts
ResNet Robust for over-fitting 152 ImageNet 25.6 3.57 224 × 224 × 3 He et al. [22]
Inception-ResNet-v2 Use residual links 164 ImageNet 55.8 3.52 229 × 229 × 3 Szegedy et al. [21]
Xception Convolution followed by a 71 ImageNet 22.8 0.05 229 × 229 × 3 Chollet [23]
point wise convolution
DenseNet Used blocks of layers 201 ImageNet, 25 3.46, 5.54 224 × 224 × 3 Huang et al. [24]
CIFAR-
100,
MobileNet-v2 Inverted residual structure 53 ImageNet 1.677 – 224 × 224 × 3 Sandler et al. [25]
HRNetV2 High-resolution represen- – ImageNet 45.2 5.4 224 × 224 × 3 Ke Sun et al. [26]
tation
and outline prominent challenges in parasites detection research questions are shown in Table 3. In the end, this sys-
and classification. We reviewed 70 related technical papers tematic review paper also responds to these research ques-
from 2012 to the present. This work will act as a guide to tions posed here.
researchers for a comprehensive understanding of the pre-
sent state-of-the-art, future challenges and trends in the 3.2 Paper Retrieval and Filtering Process
research area of parasites detection and classification.
By applying deep learning, we have conducted a potential
3.1 Research Questions investigation in the field of image detection and classifica-
tion of various parasites. As shown in Fig. 7, online data-
This review aims to provide insights into cutting-edge bases and digital libraries were searched in order to collect
machine learning and deep learning methods for parasite sufficient relevant or related paper collections. The focus
detection and classification based on microscopic images, was to collect papers from the most reputed publishers, such
as well as an analysis and summarization of previous work as ACM, IEEE, ScienceDirect, Springer, and Wiley Online
on deep computer vision. From that perspective, the authors Library.
develop the research questions to be addressed through the The search and filtering process of papers involved in this
research methodologies implemented by researchers. These review is shown in Fig. 8. In the papers retrieving stage, we
13
Advances Towards Automatic Detection and Classification of Parasites Microscopic Images… 2019
13
2020 S. Kumar et al.
13
Advances Towards Automatic Detection and Classification of Parasites Microscopic Images… 2021
In the year 2001, Yang et al. [31] proposed a framework Widmer et al. [34] developed an ANN-based model
based on an artificial neural network classifier and digital for automatic identifications of Cryptosporidium oocyst
image processing techniques for automatic detection of and Giardia cyst digital images. The digital images were
helminth eggs of humans on microscopic faecal speci- captured using a camera at ×400 magnification and con-
mens. Digital image processing techniques were applied to verted into a binary array. The ANN for Cryptosporidium
extract the morphometric characteristics of eggs of human oocyst was trained with 1,586 images whereas the ANN for
parasites in faecal specimens from microscopic images. The Giardia cyst was trained with 2,431 images. After training
dataset contained 82 microscopic images of seven common these networks were validated with unseen 500 images (250
human helminth eggs, which were used to train the pro- positives, 250 negatives) of Cryptosporidium oocyst and 282
posed model. The proposed ANN model performed identi- (232 positives, 50 negatives) images of Giardia cysts. Exper-
fication of human helminth eggs in two stages. In stage first, imental result of proposed model shows that the ANNs cor-
ANN-1 isolated eggs from confusing artefacts and in stage rectly identify the Cryptosporidium oocyst and Giardia cyst
two, ANN-2 classified eggs by species, as shown in Fig. 9. images with an accuracy of 91.8% and 99.6% respectively.
In the same year, Tchinda et al. [32] presented a machine Chen et al. [35] proposed a model based on machine
learning technique to recognize intestinal parasite cysts from learning that performed classification and counting of bacte-
microscopic images. Probabilistic neural network approach rial colonies from Petri dish images. This model recognized
trained by using image pixels feature was employed. It is an both achromatic and chromatic images effectively. Support
effective machine learning approach for classification prob- Vector Machine approach was used for classification based
lems. In this approach parasites separated from microscopic on morphological features. Two types of Petri-dish were
images and then resized to 12 × 12 pixels images. Principal used for experiment. The proposed model predicted com-
component analysis basis projection is used to reduce the parable performance to accomplish automation of the bacte-
dimensionality. The proposed model was trained on 540 rial colony. The proposed machine learning model achieved
human parasite cyst images. The trained model successfully 96% accuracy level, For 75 achromatic images 97% accuracy
classified intestinal parasites into 9 different kinds. obtained and for 25 chromatic images 95%.
In year 2002, K. W. Widmer et al. [33] proposed a ANN An automatic and rapid detection model based on neural
based system for Cryptosporidium parvum oocysts detection network was developed by Kumar et al. [36] for pathogens
to reduce the analysis time and achieve the high accuracy of in foods. The proposed model involved identification in two
diagnostic process. A total of 525 images of labelled oocysts, stages. In stage first, background correction is applied to dis-
fluorescent microspheres, and other miscellaneous nonoo- tinguish the treated image from the image background using
cyst images were collected and employed in the training of better approaches. In stage second, collecting the images of
the ANN. Each type of digital images were separated into the local region. Thereafter textural, optical and geometrical
20% for training datasets and 80% for test datasets. Result features of processed pathogen images are collected. Finally,
shows that the correct identification of authentic oocyst the proposed model based on Probabilistic Neural Network
images ranged from 80 to 97%, and the correct identification applied to classify the microorganisms from collected.
of nonoocyst images ranged from 77 to 82%, on test dataset. Osman et al. [37] developed a model based on image pro-
cessing and genetic neural network techniques for automatic
13
2022 S. Kumar et al.
detectioin of Mycobacterioum tuberculosis in tissues. The characteristics (shape, shell smoothness and size). In stage
proposed model is divided into two steps: step one invoved third, filtration with Steady Determinations Thresholds Sys-
K-means clustering methods for image segmentation and tem method was used to identify and classify the types of
step second involved GANN method for feature selection, parasites based on features values. The final result predicted
classification and feature extraction purposes. After applying success rates of Ascarislumbricoides and Trichuristrichiura
a genetic algorithm to select features, a multilayer percep- almost 93% and 94% respectively.
tron was trained for the final classification of bacteria (true In [40] developed a model based on machine learning
TB and Possible TB). Dataset collection includes 960 total for automatic classification and segmentation of human
object images of which 360 for true TB and 600 for possible intestinal parasites. The proposed model classifies into two
TB. The proposed model was trained on 400 images and the stages, in stage first segmentation process was performed
rest of 280 images used for testing purposes. Experimental using image transformation, quantization border enhance-
results demonstrated that proposed approach able to produce ment and ellipse matching, as seen in Fig. 11. In the second
84.9% accuracy with fewer input features. stage, classification was performed with different ML algo-
Hiremath et al. [38] presented identification and clas- rithms like ANN-MLP cum Adaboost, optimum path forest
sification of a bacterial cell of cocci data collection using classifier, SVM and SVM cum Adaboost. After investigating
machine learning technique. The proposed model was devel- the performance of different ML algorithms, the optimum
oped using 3α, KNN classifiers and a selected neural net- path forest classifier prediceted a good result for classifica-
work to recognize the pattern of cocci bacterial cells. The tion images. The experiment result shows that the proposed
data is processed by applying a neural network pipeline that
includes inputs layers, output layers, gradient descent and
backpropagation function. In this experiment 500 different
types (sarcinae, streptococci, diplococci, cocci and tetrad)
of digital bacterial images were used. Using the proposed
model up to 94% level accuracy was achieved based on the
3α classifier. With KNN classifier where k = 1, 75% to 100%
accuracy achieved and up to 100% accuracy achieved with
Neural Network classifier.
A ML based approach was implemented by Ghazali et al.
[39] for human fecal parasite detection based on comput-
erized image analysis. The presented model contains three Fig. 11 Overview segmentation pipeline include operation like Quan-
stages as shown in Fig. 10. In stage first, the pre-processing tization for colored conversion, Border enhancement for quantized
image border with Sobel Gradient operator, and Ellipse matching:
techniques were applied to enhance features. In stage sec- objects having higher degree of ellipse matching considered as para-
ond, a features extraction mechanism was used with three site candidate
Fig. 10 Overview of three different Pre-processing methods: (Pre-method 1) Content noise reduction with median filter, (Pre-method 2) Content
edge detection using canny filter and (Pre-method 2) Content edge enhancement using unsharp filter
13
Advances Towards Automatic Detection and Classification of Parasites Microscopic Images… 2023
ANN based model classify 155 images from the test dataset spreading infection in cucumber plant leaves. The proposed
with 98.22% accuracy. scheme contained total 300 images of healthy and infected
Nugroho et al. [41] developed a model based on image plants. Among these, 250 images were used for training
processing techniques to recognize three phases of malaria ANN model and the remaining were used for inspection. In
parasite cell host in microscopic images i.e. schizont, tropho- this experiment, the training dataset indicated good fit. The
zoite and gametocyte plasmodium falciparum. The proposed relationship between the outputs and the inputs for validation
model develops in two phases. In phase first image pre-pro- and test dataset was 0.9.
cessing was implemented with median filter and contrast Liu et al. [46] focused on KNN classifier to classify the
stretching. In phase second k-means method was applied for morpho type bacterial species based on morphological fea-
image segmentation. Finally, a multilayer perception back- tures. Total 1937 digital images were collected for the pro-
propagation technique was employed for classification. The posed system. Among these, 1271 cells images were applied
data collection contained 60 images of trophozoite, game- to train the classifier which exhibits 96% accuracy and 466
tocyte and schizont. The proposed model achieved accuracy test cell images exhibit 97% accuracy.
level of 87% with specificity 90% for detection. Inayah et al. [47] worked on a Randomly Wired Neural
In the year 2016, Seo et al. [42] developed a machine Network for recognize the parasites from red blood cells.
learning model for classification of staphylococcus species. Secondary data set is used in this model collected from the
In this experiment, the authors used five different types of National Library of Medicine (NLM). In this model total
species of staphylococcus bacteria namely aureus, haemo- of 27,558 images of red blood cells were used as data. The
lyticus, hyicus, sciuri and simulans. Mahalanobis distance proposed model worked on feedforward and backpropaga-
method was applied to eliminate the outliers, after that wave- tion techniques. Using this model, average accuracy 95.08%
length selection performed using correlation coefficient. The was achieved in fivefold cross-validation.
proposed model classified the staphylococcus bacteria spe-
cies using Partial least square discriminant analysis and sup- 4.2 Summary
port vector machine. With the proposed model 89.8% accu-
racy was achieved using Support Vector Machine and 97.8% In the past two decades, traditional ML methods for parasite
accuracy using Partial least square discriminant analysis. detection and classification have been updated from time to
In [43] purposed an approach that utilized the multi-scale time. Table 4 displays the related research works, including
wavelet counter detection to detect the parasites. Jointly references, publication date, methods, objectives, species
active contours and Hough Transform were used to perform categories, and data details evaluation metrics, limitations,
detection and segmentation of parasites images. The pro- and future scope.
posed model involved Principal Component Analysis and
Probabilistic Neural Network. Principal Component Analy-
sis was used to extract and reduce the features acquired from 5 DL Based Methods and Models
parasites images pixels and a probabilistic neural network
model was used for the classification task. This model is In recent years, high-stake applications have been imple-
tested on 15 intestinal parasites species with 900 microsco- mented using deep learning methods and models for micro-
pies images. Using this approach, the correct rate of clas- scopic image diagnosis. An extensive literature review and
sification obtained was 100%. study is carried out for the proposed topic of convolutional
Nkamgang et al. [44] has purposed neuro-fuzzy approach neural network techniques based on deep learning for detec-
to automatic detection and classification of human intestinal tion and interpretation using clinical intestinal protozoa
parasites. This model is based on segmentation and training microscopic images. In this section, an overall reviewed
of classifier. In this approach, parasites were localized using summary of methods and models of related subjects is
circular Hough transform and after that distance regularized prepared.
level was initialized for segmentation. Finally, classification
was performed by applying a trained neuro-fuzzy classifier. 5.1 DL Related Works in Proposed Field
The proposed model has been applied for identification and
classificatoin of 20 type’s human intestinal parasites. For Hung et al. [48] presented the task of detection of individual
every 20 classes of intestinal parasites satisfactory classi- cells and their respective classes using Faster Region-based
fication result was obtained and 100% recognized rate was Convolutional Neural Network (Faster R-CNN). The pro-
achieved. posed model contains two sub-module, as seen in Fig. 12.
Vakilian et al. [45] developed a model based on image Sub-module first apply Faster R-CNN to detect individual
processing techniques and an artificial neural network for cells from the image by generating a bounding box around
recognizing two types of fungi that are responsible for cells like red blood cells or non-red blood cell and in
13
Table 4 Summary of reviewed research works on traditional machine learning methods and models as applied for detection and classifications used objective, dataset details which include class
2024
(C), total images (T.I), training (Tr.), test (Ts.) and evaluation metrics which include accuracy (Acc.), precision (Pre.), recall (Rec.), sensitivity (Sn.), specificity (Sp.)
References Date Method Type of features Objective Parasites (species) Dataset details Evaluation metrics Limitations Future scope
13
[31] 2001 ANN Pixel intensity Automatic detection Helminth eggs C=2 Acc. = 86.1% Small dataset An enhanced model
of human helminth T.I = 82 proposed to classify
eggs helminth eggs
[32] 2001 ANN Pixel intensity Automatic Identifica- Helminth Eggs C=7 Det. = 86.1% Small dataset To improve results
tion of human T.I = 82 large dataset may
helminth eggs be used
[34] 2005 ANN Shape feature Classification of Giardia cyst and C=2 Acc.(GC) = 91.8 Insufficient dataset More tests from food
Giardia cyst (GC) Cryptosporidium Tr. = 2431(GC), Acc.(CO) = 99.6 matrices need to be
and Cryptosporid- oocyst 1586(CO) continued to validate
ium oocyst(CO) Ts. = 100 & confirm HMI
methods
[35] 2009 SVM Shape feature automated bacterial Bacterial colony C=2 Acc = 96% Clustered colonies of Model may improve
colony counting T.I = 100 Pre. = 0.97 bacteria’s are not for better classifica-
and classification Rec. = 0.96 distinguishes tion
F1-Score = 0.96
[36] 2010 ANN Shape feature Rapid detection of Microorganisms C=5 Acc. = 100% Small dataset
microorganisms TI = 1
[37] 2010 GA-NN Shape feature GA-neural network mycobacterium C=2 Acc. = 89.64% In sufficient Evalua- Improving the
approach for myco- tuberculosis T.I. = 960 tion Criteria performance of the
bacterium tubercu- Tr. = 400 bacteria colony clas-
losis detection Ts. = 280 sification
[38] 2011 K-NN, ANN Geometric features Identification and cocci bacterial cells C=6 Acc. = 99% Over lapped cells Improving the limita-
classification of T.I = 350 not considered in tion of proposed
cocci bacterial purposed model model
cells
[39] 2013 F-SDTS Shape feature Automated system Ascaris lumbri- C=2 Acc.(ALO) = 93% limited to the basic An enhanced model
for diagnosis intes- coides (ALO), T.I (ALO) = 100 Acc.(TTO) = 94% diagnosis parasitic proposed to diagno-
tinal parasites by Trichuris trichiura T.I (TTO) = 100 worms sis parasite worms
ova (TTO)
[40] 2013 ANN,SVM Shape feature Classification of Human intestinal C = 16 Sn. = 90.38% Insufficient dataset Improve conventi onal
human intestinal parasites TI = 5763 Sp. = 98.22% slides impurities, by
parasites Tr. = 2881 applying parasite-
Te. = 2 ological techniques
[42] 2016 SVM Shape feature Identification of Staphylococcus C=5 Acc. = 97% Insufficient dataset Size of dataset may
Staphylococcus species be increased for
species validation
[41] 2015 ANN Shape feature Classification for Trophozoite, schiz- C=3 Acc. = 8708% Insufficient dataset Increase
detection malaria ont, and gameto- T.I = 180 Sn. = 81.7% classification with
parasites cyte Sp. = 90.8% more feature extrac-
tion methods
for identification
S. Kumar et al.
Advances Towards Automatic Detection and Classification of Parasites Microscopic Images… 2025
working efficiency
improved to detect
types of parasites
DL model signifi-
Model may be 1300 images and after pre-processing these images contain
Model may be
Future scope
sites too
accuracy of 98% for the proposed model.
M. Górriz et al. [49] used U-Net, a deep convolution
neural network to classify and segment the Leishmanio-
sis Parasite, which causes thousands of deaths in some
Insufficient dataset
Imbalance dataset
Rec. R. = 100%
Rec. = 96.44%
Pre. = 93.62%
C.R = 100%
T.I = 1800
Ts. = 900
Tr. = 600
Tr. = 900
C = 15
Classification of fal-
for intestinal para-
Automated medical
and classification
diagnosis system
ciparum parasite
divided this model into two parts, first to detect the parasites
entities Iterative Global Minimum Screening technique used
parasites
sitosis
AUC of 98.39%.
Shape feature
sifier
[43]
[47]
13
2026 S. Kumar et al.
images. Traditionally, the ova-and-parasite (O&P) inspec- hundred twenty-seven slides of 11 categories of protozoa
tion method is used to manually microscopic evaluation to train the model. During the model development, various
of stool. It is a resource and time-consuming method. The training steps were executed and analysis of model perfor-
purpose of this research is to develop a novel CNN model mance was evaluated with the resulting metrics. All images
along with scan high-resolution digital side images to rec- were resized to 250 by 250-pixel images, of which 10% of
ognize intestinal parasites from stool, as shown in Fig. 14. label images of all classes were used for validation of CNN
The whole task is divided into three parts. Part I, collect the after training. The proposed intestinal detection model archi-
digital scan microscopic images of intestinal protozoa con- tecture is RGB CNN based on the SSD Inception V2 trans-
taining target classes. Part II, input the collected microscopic fer learning model. The based model pertained with COCO
images into CNN model for training so as to detect defined image dataset. The trained model was shown the collection
classes. Part III, perform the validation and prediction of of 250 by 250-pixel images scenes to recognize the parasites.
the trained model. The author collected and prepared one The model created a label image box to detect the parasite.
Fig. 12 Detail of how proposed two stage deep learning model for objects and label them as RBC or other. b The labelled objects as
recognition and classification is applied to images during test phase. other are sent to AlexNet model to undergo more fine-grained clas-
a An original image is fed into Faster R-CNN model to recognize sification
13
Advances Towards Automatic Detection and Classification of Parasites Microscopic Images… 2027
The precision recall plot were used to view the model per- construction. The dataset was prepared by collecting 5000
formance on the basis of per labelled parasite image box. bacterial cells images of five foodborne bacterial cultures, of
The slide-level agreement is used to calculate the accuracy which 72% are used for training, 18% are used for validation
of the model, the positive agreement achieved was 98% and and10% are used for the test dataset. Results show that the
the negative was 98.11%. proposed model yields classification accuracy up to 98.4%.
Baek et al. [54] developed a model with Fast Regional Luo et al. [56] report a deep learning-based model for
Convolutional Network (Fast R-CNN) based on DL to predicting Cryptosporidium and Giardia in drinking water.
quantify and classify five cyanobacteria. The proposed Fast The proposed system merged imaging flow cytometry with
R-CNN model includes two stages, in stage first classify MCellnet an efficient neural network. Figure 15 shows the
cyanobacteria species taken from microscopic images using architecture of MCellNet. The dataset collection included
the fast R-CNN method and in step second CNN technique millions of raw images of which 80,146 images were
is used to quantify the cyanobacteria cells. The dataset col- selected for the final image database and each image from
lection covered 200 images of five species of cyanobacteria the database was patched to 120 × 120 pixels. 13 classes of
(Microcystiswesenbergii, Microcystis aeruginosa, Dolichos- Cryptosporidium (2082 images), Giardia (3569 images)
permum, Aphanizomenon and Oscillatoria). Experiment were included in the dataset. The image data set is randomly
results show that the Fast R-CNN based model was able to split into a training data set (38,469 images), a validation
achieve a reasonable accuracy for classification and yielding data set (9618 images) and a test data set (32,059 images)
average precision (AP) values of 0.929, 0.973, 0.829, 0.890, that contained 48%, 12%, and 40% images respectively. The
and 0.890 for respective species. proposed model for multiclass and binary classifications
Kang et al. [55] purposed an expensive deep learning net- achieves accuracies of 99.69% and 99.7%, respectively.
work which utilized 1D CNN, Long-Short Term Memory Nakasi et al. [57] evaluate the performance of the
Network (LSTM) and Deep Residual Network (ResNet). The AlexNet and GoogleNet model based on transfer learning
proposed hybrid deep learning model define as Fusion-Net for the diagnosis of scan digital intestinal parasites stool
perform the classification of foodborne bacteria at a sin- microscopic images. The authors compared these two mod-
gle-cell level. The Fusion-Net formation was performed in els with trained Convolution Neural Network for the same
three parts comprising hyper parameter optimization, mul- work. Models are evaluated on the system having low speci-
tiple deep learning architecture selections, and Fusion-Net fication which shows that models can be deployed to tackle
13
2028 S. Kumar et al.
real word diagnostics problems. A total of 6500 (10.9% posi- network. The trained model was validated with validation
tive) image patches were applied in AlexNet, 6461 image images. Figure 17 depict the architecture of the proposed
patches (11% positive) applied in GoogleNet and 2071 Convolution auto-encoder model. The dataset collection
image patches (30.5% positive) applied in Custom CNN. included 644 nematode microscopic images that were used
From proposed models, the AlexNet attained accuracy ROC to investigate soybean cyst nematode eggs. 80% of images
AUC of 1.00 and GoogleNet attained ROC AUC of 99. were used to train the model and the rest of the images were
Lee et al. [58] developed a Helminth Egg Analysis Plat- used for validation. Experimental results predicted accuracy
form (HEAP) that help medical technicians to diagnosis of 94.33% for the proposed model.
parasite infections. The authors integrated the various deep Tahir et al. [61] develop a Convolution Neural Network
learning techniques (SSD, Single Shot MultiBox Detector, based on deep learning for the detection and classification of
U-net, and Faster R-CNN, Faster Region-based Convolu- five different types of fungus spores and dirt. Around 40,800
tional Neural Network) to recognize the helminth egg speci- annotated RGB images of 6 classes were developed for fun-
mens as shown in Fig. 16. The proposed model also includes gus detection and classification. The model was trained on
pixel level based methods image binning and egg-in-edge 30,000 fungus images and each class contain 5000 images.
algorithm to improve performance. HEAP exhibit effective The test set comprises 10,800 fungus images with 800
performance in counting and recognizing the helminth eggs images per class. The accuracy achieved by the purposed
from digital images. model is 94.8%.
Litjens et al. [59] investigated a deep convolution neural Oomman et al. [62] developed an automatic approach
network to improve the efficiency and accuracy of cancer based on deep convolution neural network for the detection
diagnosis in H and E images. This model is used to perform of Tuberculosis bacilli from microscopic images. The pro-
two different tasks, first prostate cancer recognition from posed model was developed in two stages, stage first done
biopsy samples, and second breast cancer recognition from image binarisation with Otsu threshold algorithm and in sec-
sentinel lymph nodes. Pre-processing pipeline of the model ond stage classification of detected regions done using a con-
consists of 4 convolution layers for features extraction, 3 volutional neural network. The dataset collection included
max-pooling layers to reduce the dimensionality and a dense 120 images along with ground truth, each image in the data-
layer for classification. Dataset for prostate cancer includes set has 2816 × 2112 pixel resolution. For CNN training and
225 glass slides of which 100 are selected for training, 75 testing the images were cropped to 900 negative patches and
for testing and remaining for validation. Dataset collection 900 positive patches. From the total of 1800 patches, 80%
for breast cancer sentinel lymph nodes for the experiment were used for training purposes and 20% used for testing.
includes 271 slides of which 98 for training, 33 for validation Experimental results show that the proposed model achieved
and remaining for testing the model. Optimal percentiles a recall of 97.13%, a precision of 78.4% and an F-score of
were obtained using the validation set for both ROC curves 86.76%.
and the highest specificity that was the median and 90th Treebupachatsaku et al. [63] purposed a method based
percentile for both. on deep learning for the detection of a genus of a bacte-
Panicker et al. [60] work on end to end selective auto- rium from microscopic bacterial images using the Tensor-
encoder approach based on convolution neural network to flow framework. More than 800 sample images of S. aureus
recognize complex soybean cyst nematode eggs from micro- and L. delbruekii datasets were collected. Eighty percent of
scopic images. The soybean cyst nematode eggs training images from both datasets were applied to train the proposed
patches were used to train the proposed convolutional neural
Fig. 16 Functionality of HEAP-assisted parasite egg investigation. In the images. HEAP carried out the egg recognition and egg counting
stage first, all the specimen preparation measures are applied. Then, using cloud computing. Finally, medical expert verify the model pre-
an automatic microscope image system was used to digitalize the diction result using computer client on internet
specimen slides. Multiple focusing planes were required to gather all
13
Advances Towards Automatic Detection and Classification of Parasites Microscopic Images… 2029
Fig. 17 Convolutional auto-encoder architecture to recognize complex soybean cyst nematode eggs from microscopic images
model and the remaining images were used for testing. The classification using transfer learning technique ResNet. The
proposed model achieved 96% of validation accuracy. dataset collection was prepared from UTMC.V1.DB and
Pedraza et al. [64] worked on deep learning-based neu- UTMC.V2.DB databases containing 703 images from 55
ral to check the diatom detection from water. The authors different classes and 1303 images from 97 different classes
determined the diatom detection with two popular transfer from respectively. The experimental result exhibits that the
learning techniques i.e. RCNNs (Region-based Convolu- former method acquired an accuracy of 80.81% and 84.81%
tional Neural Networks), which applies convolution opera- on both datasets. The secondary method acquired accuracy
tion on candidate region and YOLO (You Only Look Once) of 90.24% and 85.96% on both datasets.
which applies a neural network over the whole image. These Zhou et al.[66] implemented a model by applying a
two methods were trained on 11,000 microscopic images of transfer learning approach based on a convolution neural
diatom from 10 species. Diatom detection results of RCNN network to automatic analyses diatom from digital whole-
and YOLO are depicted in Fig. 18. The experimental result slide images. The proposed model applied the GoogLeNet
shows that the YOLO model performs better with 84% Inception-V3 transfer learning technique for training to rec-
F-Measure than RCNN. ognize the diatom, as shown in Fig. 19. The dataset collec-
Sajedi et al. [65] proposed a model to recognize bacte- tion comprised 53 digital whole-slide images of which 43
rial species from solid culture plates. Two methods based slides were selected for training and 10 slides for validation.
on deep learning were applied to detect action bacterial Experimental results show that the transfer learning model
strains. In the first method, a two-level wavelet transform using the augmented database achieved accuracy of 97.67%
was utilized on action bacterial strains images. In the sec- and AUC: 99.51%.
ond method, two operations are performed i.e. data augmen- Qian et al. [67] demonstrated a novel multi-target deep
tation for blurring, cropping, and horizontal rotation and learning framework developed with Faster R-CNN for algae
13
2030 S. Kumar et al.
detection and classification. The proposed extensive model model and the remaining were used for testing and validation
was trained on a large-scale coloured microscopic algae of the model. Train model showed 97% accuracy without
dataset, as depicted in Fig. 20. The dataset collection was any over-fitting. The model achieved the same test accuracy
prepared with 1859 images of 37 algae as well as anno- on low-resolution images among 15 views. Evaluation repre-
tations of genera and classes. In this experiment, 80% of sented using confusion matrices also detect similarities and
images of genera were collected in the training set to train classification based on relevant image features.
the model and the remaining were collected for the testing Peixinho et al. [70] proposed a deep learning-based
set. The experimental result show that the successful identi- approach ConvNet that recognize the image features effec-
fication rate achieved at genus level by the proposed model tively for human intestinal parasite images. Random kernels
was 74.64% and at class level it was 81.1% are defined for hyper-parameters optimization CNN archi-
Salido et al. [68] focus on mitigating the Diatom detection tecture. For the experiments, the dataset contained 16,437
for specimen counting and sample classification challenge objects including the 15 most common species of human
with the YOYO and SegNet network based on deep learn- intestinal parasites. Using the proposed approach effective
ing. The dataset prepared collects microscopic images of accuracy was obtained for the classification of human intes-
80 species of diatoms and each species contain hundreds of tinal protozoa’s and eggs.
images. Detection of diatoms for faster diatom counting by López et al. [71] implement a model based on CNN to
using YOLO for on-time inferences with an average sensitiv- detect the mycobacterium tuberculosis (MT). Dataset consist
ity of 84,6%, specificity of 96.2%, and precision of 72.7%. of 9770 positive and negative smears patches prepared from
Holmström et al.[69] focuses on deep learning methods 492 digital images. Three types of patches (grey-scale, RGB,
for accurate and fast detection of helminths and Schistosoma R-G) were used to train the proposed CNN based model.
haematobium. The author used 8,342,769 echocardiograph The proposed model includes three layers to perform the
images of 276 patients. 80 images were used to train the
Fig. 20 Model architecture depicts 3 branches simultaneously out- predict the genus of algae. Branch-2 is used for algal detection and
puts the genus, bounding box, and a biological class of the algae and localization. Branch-3 is used to predict the class of algae
orange component extended classification branch. Branch-1 is used to
13
Advances Towards Automatic Detection and Classification of Parasites Microscopic Images… 2031
classification of patches into two classes i.e. positive or neg- to increase the size of the dataset. From this dataset, 70%
ative MT. The proposed model accomplished 96% accuracy of images were used to train the model and 30% were used
level. for validation. The proposed model obtained experimental
Zieliski et al. [72]developed a hybrid deep learning model results with 88.59% accuracy.
for the classification of bacterial species from digital images. In [77] presented a deep learning-based approach to clas-
In the proposed model deep convolutional neural network is sify bacterial colonies. The authors used Deep Convolutional
used to recognize image descriptors, subsequently features neural network (CNN) to obtain image descriptors and sup-
vectors are generated with the pooling encoder method and port vector machine (SVM) and Random forest for classifi-
finally Support Vector Machine was used for classification. cation, as shown in Fig. 21. Features based on shapes namely
The dataset used in this model includes 33 bacteria spe- spiral, cylindrical and spherical were extracted. The dataset
cies with 20 digital images of each bacterium. The proposed consisted of 660 images with 33 different genera and species
model used 50% data of the dataset for training purposes of bacteria. The experimental results predicted the accuracy
and 50% for testing. The results exhibit that the proposed of recognition to be 97.24%.
DL based model is 96.82% accurate in the classification of Hay et al. [30] presented a convolutional neural network-
bacterial species from digital images. based tool for differentiating bacteria images from non-
Wahid et al. [73] implement a model by using a transfer bacterial images using three-dimensional microscopy data
learning technique to automate the classification and rec- of gut bacteria found in larval Zebrafish. The authors used
ognition task using a deep convolution neural network. The TensorFlow framework to implement 3D convolutional neu-
proposed inspection DCNN model was trained using 500 ral network and compared the performance of the model
digital microscopic images of five bacterial species. The with support vector machine and random forest classifiers.
dataset was split into the training (80% images) and testing The proposed model performed better with 89.3% accuracy
(20% images) parts. The detection and classification rate of whereas random forest classifier and support vector machine
the proposed model is 85% classifiers achieved accuracy of around 78.5% and 83.1%
Ahmed et al. [74] Implemented a hybrid approach to respectively.
classify microscopic bacterial images using the SVM and In [78] developed a CNN based model using transfer
Inception-V3 model. In the proposed model image pro- learning approaches for automatic classification of parasites
cessing techniques like image cropping, converting images with low quality microscopic scanned images, as shown in
from grayscale to RGB, image flipping, image translation
and feature extraction by Inception-V3 Deep CNN method
were used. The SVM was used to classify the microscopic
bacterial images into defined classes. The authors used 800
bacterial images to train the proposed model and 200 images
for testing. Based on the proposed model 75% accuracy in
the classification of bacterial species from digital images
was achieved.
In [75] performed an experiment on datasets of In Situ
plankton images using deep learning techniques. Proposed
model extract features from various planktonic images data-
sets i.e. Imaging Flow Cytobot (IFCB), Scripps Plankton
Camera System (SPC) and Situ Icthyoplankton imaging sys-
tem (ISIIS). The authors train CNN model by using IFCB
and ISIIS plankton images datasets. The dataset SPC was
small in size and so it was used for testing purposes. To train
the proposed CNN model images of plankton were resized
to 256 by 256. Experimental results of the proposed model
work well on the feature extraction from planktonic images
using CNN.
In [76] determined the classes of microalgae using convo-
lution neural network technique of deep learning. FlowCam
practical analyser used to extract microalgae images from
water extracted from South Atlantic Ocean. The dataset con-
tained 29,449 microalgae images that are further classified Fig. 21 Flowchart of deep learning-based approach to classify bacte-
into 19 classes. The data augmentation technique was used rial colonies
13
2032 S. Kumar et al.
Fig. 22. The patches based technique was used to search patch size set to 100 × 100 pixels. The data augmentation
the location of eggs from images. The dataset collection technique was applied to increase the size of the dataset,
contained ×10 magnification microscopic images of four approximately 10,000 patches per egg type. In order to all
different types of parasites i.e. Ascarislumbricoides (67 implement the proposed framework, a transfer learning
images), Hymenolepisdiminuta (27 images), Fasciolopsis- technique was employed with fine-tuning pertained models.
buski (32images) and Taeniaspp. (36 images). Before apply- These models have been trained on large dataset of images
ing data augmentation and the patch overlapping, grayscale collected from different specific applications. The last two
conversion and contrast enhancement are performed on the layers of these models were replaced with a fully connected
parasite egg image collection. The grayscale conversion layer and a softmax layer to classify images into five classes.
operation decreases the depth of the input parasite image For object detection, AlexNet is a cutting-edge model that
from three channels of RBG to one channel of grayscale. improved CNN execution performance, whereas ResNet50
Further, the visualisation of low magnification of images is is a more sophisticated architecture that performs better
enhanced with contrast enhancement. Each parasitic image for image classification tasks. The dataset of parasite egg
was split into small patches, which allowed the model to images was split into two parts, part first contained 60% of
extract features from the image by examining the local areas. the images for training purposes, and the second contained
In order to encapsulate the mentioned parasites, patch size 40% of the images for testing purposes. Based on the pro-
set to 100 × 100 pixels. The data augmentation technique posed framework, experiment results represented state-of-
was applied to increase the size of the dataset, approxi- the-art parasitic egg detection and classification task. Based
mately 10,000 patches per egg type. In order to implement on the proposed framework, experiment results represented
the proposed framework, a transfer learning technique was state-of-the-art parasitic egg detection and classification.
employed with fine-tuning pretrained models. These models Quinn et al. [27] authors purposed a framework based
have been trained on large dataset of images collected from on Deep CNN to evaluate the performance with differ-
different specific applications. The last two layers of these ent microscopy tasks i.e. intestinal parasite eggs in stool
models were replaced with a fully connected layer and a samples, tuberculosis in sputum samples, and diagnosis of
softmax layer to classify images into five classes. Grayscale malaria in thick blood smears. The experts mark bounding
conversion operation decreases the depth of the input para- boxes around each interested entity in all images. Finally,
site image from three channels of RBG to one channel of prepared, plasmodium was annotated (7245 entities in 1182
grayscale. Further, the visualisation of low magnification of images) in thick blood smear images, tuberculosis bacilli
images is enhanced with contrast enhancement. Each para- were annotated (3734 entities in 928 images); in sputum
sitic image was split into small patches, which allowed the samples, and, the eggs of hookworm, Taenia and Hymeno-
model to extract features from the image by examining the lepsis nana were annotated (162 entities in 1217 images)
local areas. In order to encapsulate the mentioned parasites, in stool samples. The proposed model is trained on the
13
Advances Towards Automatic Detection and Classification of Parasites Microscopic Images… 2033
collected images. After training the resulting model was potential model of machine learning used for microscopic
applied to the test set: plasmodium detection set, which con- parasite image classification and detection is the Support
tains 261,345 test patches, tuberculosis set contain 315,142 Vector Machine. SVM is a supervised algorithm used for
test patches and hookworm set contains 253,503 patches. In both linear and non-linear data. It defines the classification
all cases, experiment results show that accuracy was higher by constructing the set of hyperplanes in feature space [79].
and better than traditional medical imaging techniques. In SVM, kernel functions are used to perform the transfor-
Butpoly et al. [75] proposed DL based method for clas- mation in the hyperplane. The most commonly used kernel
sification of Ascaris lumbeicodes parasites images. The pro- functions are linear, polynomial, sigmoid, and radial basis
posed model recognize three types of eggs of Ascaris lumbe- functions. A better classification is achieved with the hyper-
icodes with effective approach of deep learning. The dataset plane maximum distance to the nearby training data point of
collection included training and testing data. Both training classes [80]. However, this model is not effective for multi-
and testing dataset included three types Ascaris lumbeicodes ple species classification and detection.
eggs namely infertile eggs, fertile eggs and decorticate eggs. The deep learning methods and models surveyed in this
For this experiment the training dataset consist of 200 of paper are playing a major role in parasitological research.
each type (total images 600). Experimental results predicted Furthermore, it supports tackling similar problems within
93.33% classification accuracy of the parasites eggs. several other subdomains. Deep learning models handle
Avic et al. [76] implemented methodology based on more complex tasks than traditional machine learning, such
multi-class support vector machine for classification of as object detection, image segmentation, image recognition,
human parasites eggs from digital microscopic images. The and classification. Figure 24 shows CNN-based transfer
proposed model consists of four steps. These steps are pre- learning models used in parasite classification and detec-
pre-processing feature extraction, classification and testing. tion. Popular classical deep learning models, such as CNN
In pre-processing step, the image processing methods, such and its offspring methods and models, are constantly used in
as contrast enhancement, thresholding, noise reduction are the task of parasite classification and detection, such as CNN
applied. In second step i.e. feature extraction, the invariant mentioned in [60, 61, 70], AlexNet and GoogleNet men-
moments of parasites images obtained in step first are evalu- tioned in [57], AlexNet and ResNet50 mentioned in [78],
ated. In classification step, the multi-class support vector Faster R-CNN mentioned in [50], and R-CNN mentioned
machine was applied to classified feature collected in feature in [54]. According to the survey on deep learning methods
extraction step. The proposed model was tested with test and models for parasites, the most popular detection and
data. The proposed approach achieved average accuracy of classification model is Faster R-CNN. In [50, 51] Faster
97.70% for classification of human parasites. R-CNN is mentioned. In comparison to R-CNN, Faster
R-CNN implemented with the Regional Proposal Network
5.2 Summary (RPN) technique performs better because in Faster R-CNN,
RPN improves the performance and accuracy of detection.
In the past two decades, traditional deep learning methods Moreover, it also exhibits end-to-end detection [81]. The
for parasite detection and classification have been updated main limitations of Faster R-CNN are that it cannot detect
from time to time. Table 5 displays the related research objects in real time and that it performs large amounts of
works, including references, publication date, methods, computations due to the RPN extraction method.
objectives, species categories, data-details evaluation met-
rics, limitations, and future scope.
7 Conclusion and Research Directions
13
Table 5 Summary of reviewed research works on deep learning methods and models as applied for detection and classifications used objective, dataset details which include class (C), total
2034
images (T.I), training (Tr.), validation (V), test (Ts.) and Evaluation metrics which include accuracy (Acc.), precision (Pre.), recall (Rec.), area under the ROC curve (AUC), mean average preci-
sion (mAP), sensitivity (Sn.) and specificity (Sp.)
13
References Date Types of features Method Objective Parasites (species) Dataset details Evaluation Limitations Future scope
[48] 2015 Deep features Faster R-CNN Object detection on Plasmodium vivax –- Acc. = 98% Validation of our model Intend to test the
malaria images jane needs to be done model on more
reliable ground
truth and test for
robustness by
testing on samples
prepared in a dif-
ferent lab
[70] 2016 Deep features CNN Diagnosis of human Intestinal parasites C = 15 Acc. = 0.9649 ± 0.0043 Clinical laboratory Intend to further
intestinal parasites T.I = 16,437 validation was not extend the work to
employed diagnosis of para-
sites in animals
[71] 2017 Deep features CNN Classification of Mycobacterium tuber- C=2 AUC = 96% Insufficient data The proposed model
Mycobacterium culosis T.I = 9770 can be used in
tuberculosis sliding for detect-
ing the parasite in
full smear micros-
copy images
[75] 2017 Deep features AlexNet Deep feature extraction Planktonic images T.I = 30,336(ISIIS),53,239 Qualitative Small set of classes Include more
for planktonic images (IFCB) image data from
plankton domains
to improve classi-
fication accuracy
[60] 2018 Deep features CNN Automatic detection of Tuberculosis bacilli C=2 Rec. = 97.13% Insufficient dataset An enhanced model
Tuberculosis bacilli T.I = 1800 (900 + ve & 900 –ve Pre. = 78.4% proposed to
patches) F-score = 86.76% diagnosis
tuberculosis bacilli
[49] 2018 Deep features U-net model Leishmaniasis parasite Promastigote C=3 Promastigote (J > 0.54) Imbalance dataset Results that could
segmentation and Adhered T.I = 45 Adhered(J > 0.82) be improved using
classification g Amastigote Amastigote(J > 0.88) larger databases
[61] 2018 Deep features CNN Fungus detection Fungus species C=5 Acc. = 94.8% Evaluation details are explore regional
T.I = 40,800 not described convolutional
Tr. = 10,800 neural network
Ts. = 5000 and transfer learn-
ing for the fungus
detection
[63] 2019 Deep features LeNet CNN Bacteria classification S. aureus and L. C=2 Acc. = 96% Small dataset limited to Apply another CNN
using DL delbruekii T.I = 800 only two species of methodology
bacteria for comparison
such as ResNET,
AlexNET
[50] 2019 Deep features Faster R-CNN Parasite worm egg Parasite species C=8 mAP = 97.67% Less dataset details Advanced tech-
automatic detection mentioned niques
applied for
better results
[53] 2020 Deep features D-CNN Detection of intestinal Intestinal Protozoas (10 T.I = 910 Rec. = 98% Clinical laboratory Collection of addi-
protozoa Catg.) validation was not tional slides in
employed order to enrich the
data set and refine
the model
S. Kumar et al.
Table 5 (continued)
References Date Types of features Method Objective Parasites (species) Dataset details Evaluation Limitations Future scope
[54] 2020 Deep features R-CNN Identification and Cyanobacteria species C=5 AP = 0.929, 0.973, Less dataset details Research recom-
enumeration of 0.829, 0.890, and mentioned mended to
cyanobacteria species 0.890 for five classes improve the cell-
count accuracy
[55] 2020 Deep features Fusion-Net Single-cell classifica- Pathogens C=5 Acc. = 98.4% Insufficient dataset Collection of images
tion of foodborne in order to enrich
pathogens the data set and
refine the model
[57] 2021 Deep features AlexNet & GoogleNet Intestinal parasite Intestinal parasite – AlexNet AUC = 1.00 No details about quan- Transfer learning
detection in GoogleNet = 0.99 tity of database could be used to
analyze the data
further to under-
stand why
[51] 2019 Deep features Faster-RCNN Parasite detection in Parasites T.I = 2967 Detection rates on Evaluation details Focus on reducing
thick blood smears image and patient are not described false positives
based level = 96.84% &
96.81%, respectively
[69] 2017 Detection of soil- Helminths and Schisto-
transmitted helminths soma haematobium
and Schistosoma
haematobium
[56] 2021 Deep features MCellNet Deep learning- Cryptosporidium and C=2 Acc. = 99.0% Insufficient dataset High-quality
(CNN) enabled predicting Giardia Tr. = 38,469 Pre. = 98.66% training data,
Cryptosporidium and Ts. = 32,059 F1-Score = 99.09% MCellNet could
Giardia in drinking Val. = 9618 be extended to
water detect other types
of bioparticles in
high speed
[58] 2021 Deep features Single Shot Detector Helminth egg Helminth egg C = 17 Qualitative Evaluation details Model improved
(SSD), U-net and identification and are not described by integrating
Faster R-CNN quantification based increasing number
Advances Towards Automatic Detection and Classification of Parasites Microscopic Images…
13
2035
2036 S. Kumar et al.
13
Advances Towards Automatic Detection and Classification of Parasites Microscopic Images… 2037
2. Hotez PJ, Brown AS (2009) Neglected tropical disease vaccines. 21. Szegedy C, Vanhoucke V, Ioffe S, Shlens J, Wojna Z (2016) Pro-
Biologicals 37(3):160–164. https://doi.org/10.1016/j.biologicals. ceedings of the IEEE conference on computer vision and pattern
2009.02.008 recognition (CVPR). pp 2818–2826
3. CDC - DPDx - Parasites A-Z Index. https://www.cdc.gov/dpdx/ 22. He K, Zhang X, Ren S, Sun J (2015) Deep residual learning for
az.html. Accessed 20 Nov 2021 image recognition. arXiv. https://doi.org/10.48550/arXiv.1512.
4. Schistosomiasis and soil-transmitted helminthiases: numbers of 03385
people treated in 2017. https://www.who.int/publications/i/item/ 23. Chollet F (2016) Xception: deep learning with depthwise sepa-
who-wer9350. Accessed 20 Nov 2021 rable convolutions. arXiv. https://doi.org/10.48550/arXiv.1610.
5. Pullan RL, Smith JL, Jasrasaria R, Brooker SJ (2014) Global num- 02357
bers of infection and disease burden of soil transmitted helminth 24. Huang B, Thorne PW et al (2017) Extended Reconstructed Sea
infections in 2010. Parasites Vectors. https://doi.org/10.1186/ Surface Temperature version 5 (ERSSTv5), Upgrades, valida-
1756-3305-7-37 tions, and intercomparisons. J Climate https://doi.org/10.1175/
6. Marcilla A et al (2012) Extracellular vesicles from parasitic JCLI-D-16-0836.1
helminths contain specific excretory/secretory proteins and are 25. Sandler M, Howard A, Zhu M, Zhmoginov A, Chen L (2018)
internalized in intestinal host cells. PLoS ONE. https://doi.org/ MobileNetV2: inverted residuals and linear bottlenecks. arXiv.
10.1371/JOURNAL.PONE.0045974 https://doi.org/10.48550/arXiv.1801.04381
7. Dematei A, Fernandes R, Soares R, Alves H, Richter J, Botelho 26. Sun K, Zhao Y, Jiang B, Cheng T, Xiao B, Liu D, Mu Y, Wang
MC (2017) Angiogenesis in Schistosoma haematobium-associated X, Liu W, Wang J (2019) High-Resolution representations for
urinary bladder cancer. APMIS 125(12):1056–1062. https://doi. labeling pixels and regions. arXiv. https://d oi.o rg/1 0.4 8550/a rXiv.
org/10.1111/APM.12756 1904.04514
8. Hart BL, Hart LA (2018) How mammals stay healthy in nature: 27. Quinn JA, Nakasi R, Mugagga PK, Byanyima P, Lubega W,
the evolution of behaviours to avoid parasites and pathogens. Andama A (2016) Deep convolutional neural networks for micros-
Philos Trans R Soc B Biol Sci. https://doi.org/10.1098/RSTB. copy-based point of care diagnostics, pp 1–12, [Online]. Available
2017.0205 from http://arxiv.org/abs/1608.02989
9. Bennett A et al (2014) Building a laboratory workforce to meet the 28. Li S, Du Z, Meng X, Zhang Y (2021) Multi-stage malaria parasite
FutureASCP task force on the laboratory professionals workforce. recognition by deep learning. Gigascience 10(6):1–11. https://d oi.
Am J Clin Pathol 141(2):154–167. https://doi.org/10.1309/AJCPI org/10.1093/gigascience/giab040
V2OG8TEGHHZ 29. Li S, Li A, Molina Lara DA, Gómez Marín JE, Juhas M, Zhang
10. Petti CA, Polage CR, Quinn TC, Ronald AR, Sande MA (2006) Y (2020) Transfer learning for Toxoplasma gondii recognition.
Laboratory medicine in Africa: a barrier to effective health care. mSystems 5(1):1–12. https://d oi.o rg/1 0.1 128/m
syste ms.0 0445-1 9
Clin Infect Dis 42(3):377–382. https://doi.org/10.1086/499363 30. Parasitic Egg Detection and Classification in Microscopic Images
11. Zhang Z, Cui P, Zhu W (2020) Deep learning on graphs: a survey. | IEEE DataPort. https://ieee-dataport.org/competitions/parasitic-
IEEE Trans Knowl Data Eng 14(8):1–1. https://doi.org/10.1109/ egg-detection-and-classification-microscopic-images. Accessed
tkde.2020.2981333 19 May 2022
12. Lecun Y, Bengio Y, Hinton G (2015) Deep learning. Nature 31. Yang YS, Park DK, Kim HC, Choi MH, Chai JY (2001) Auto-
521(7553):436–444. https://doi.org/10.1038/nature14539 matic identification of human helminth eggs on microscopic fecal
13. Shrestha A (2019) Review of deep learning algorithms and archi- specimens using digital image processing and an artificial neural
tectures. IEEE Access 7:53040–53065. https://doi.org/10.1109/ network. IEEE Trans Biomed Eng 48(6):718–730. https://d oi.o rg/
ACCESS.2019.2912200 10.1109/10.923789
14. Alzubaidi L et al (2021) Review of deep learning : concepts, CNN 32. Saha B, Tchiotsop D, Tchinda R, Wolf D, Noubom M (2015)
architectures, challenges, applications, future directions. Springer, Automatic recognition of human parasite cysts on microscopic
Cham stools images using principal component analysis and probabilis-
15. Jamshidi M et al (2020) Artificial Intelligence and COVID-19: tic neural network. Int J Adv Res Artif Intell 4(9):26–33. https://
deep learning approaches for diagnosis and treatment. IEEE doi.org/10.14569/ijarai.2015.040906
Access 8:109581–109595. https://d oi.o rg/1 0.1 109/A CCESS.2 020. 33. Widmer KW, Oshima KH, Pillai SD (2002) Identification of
3001973 Cryptosporidium parvum oocysts by an artificial neural network
16. Prashar N, Sangal AL (2022) Plant disease detection using deep approach. Appl Environ Microbiol 68(3):1115–1121. https://doi.
learning (convolutional neural networks). Lect Notes Netw Syst org/10.1128/AEM.68.3.1115-1121.2002
300 LNNS:635–649. https://d oi.o rg/1 0.1 007/9 78-3-0 30-8 4760-9_ 34. Widmer KW, Srikumar D, Pillai SD (2005) Use of artificial
54 neural networks to accurately identify Cryptosporidium oocyst
17. Krizhevsky BA, Sutskever I, Hinton GE (2012) Cnn实际训练的. and Giardia cyst images. Appl Environ Microbiol 71(1):80–84.
Commun ACM 60(6):84–90 https://doi.org/10.1128/AEM.71.1.80-84.2005
18. Simonyan K, Zisserman A (2015) Very deep convolutional net- 35. Chen WB, Zhang C (2009) An automated bacterial colony count-
works for large-scale image recognition. The 3rd international ing and classification system. Inf Syst Front 11(4):349–368.
conference on learning representations (ICLR2015). https://arxiv. https://doi.org/10.1007/S10796-009-9149-0
org/abs/1409.1556 36. Kumar S, Mittal GS (2010) Rapid detection of microorganisms
19. Szegedy C, Vanhoucke V, Ioffe S, Shlens J, Wojna Z (2016) using image processing parameters and neural network. Food
Rethinking the inception architecture for computer vision. Proc Bioprocess Technol 3(5):741–751. https://d oi.o rg/1 0.1 007/
IEEE Comput Soc Conf Comput Vis Pattern Recognit 2016:2818– s11947-008-0122-6
2826. https://doi.org/10.1109/CVPR.2016.308 37. Osman MK, Ahmad F, Saad Z, Mashor MY, Jaafar H (2010)
20. Szegedy C, Vanhoucke V, Ioffe S, Shlens J, Wojna Z (2016) A genetic algorithm-neural network approach for mycobacte-
Rethinking the inception architecture for computer vision. In: rium tuberculosis detection in Ziehl-Neelsen stained tissue slide
Proceedings of the IEEE conference on computer vision and pat- images. In: 2010 10th international conference on intelligent sys-
tern recognition. pp 2818–2826 tems design and applications, pp 1229–1234. https://doi.org/10.
1109/ISDA.2010.5687018.
13
2038 S. Kumar et al.
38. Hiremath PS, Bannigidad P (2011) Identification and classifica- neural network. J Clin Microbiol 58(6):1–13. https://doi.org/10.
tion of cocci bacterial cells in digital microscopic images. Int 1128/JCM.02053-19
J Comput Biol Drug Des 4(3):262–273. https://doi.org/10.1504/ 54. Baek SS et al (2020) Identification and enumeration of cyanobac-
IJCBDD.2011.041414 teria species using a deep neural network. Ecol Indic 115:106395.
39. Ghazali KH, Hadi RS, Mohamed Z (2013) Automated system https://doi.org/10.1016/J.ECOLIND.2020.106395
for diagnosis intestinal parasites by computerized image analysis. 55. Kang R, Park B, Eady M, Ouyang Q, Chen K (2020) Single-
Mod Appl Sci 7(5):98–114. https://doi.org/10.5539/mas.v7n5p98 cell classification of foodborne pathogens using hyperspectral
40. Suzuki CTN, Gomes JF, Falcão AX, Papa JP, Hoshino-Shimizu microscope imaging coupled with deep learning frameworks.
S (2013) Automatic segmentation and classification of human Sens Actuators B Chem 309(February):127789. https://doi.org/
intestinal parasites from microscopy images. IEEE Trans Biomed 10.1016/j.snb.2020.127789
Eng 60(3):803–812. https://d oi.o rg/1 0.1 109/T
BME.2 012.2 18720 4 56. Luo S et al (2021) Deep learning-enabled imaging flow cytometry
41. Nugroho HA, Akbar SA, Murhandarwati EE (2016) Feature for high-speed Cryptosporidium and Giardia detection. Cytom A.
extraction and classification for detection malaria parasites in thin https://doi.org/10.1002/cyto.a.24321
blood smear. In: 2015 2nd international conference on information 57. Nakasi R, Aliija ER, Nakatumba J (2021) A poster on intestinal
technology, computer, and electrical engineering (ICITACEE), parasite detection in stool sample using AlexNet and GoogleNet
1(c):197–201. https://doi.org/10.1109/ICITACEE.2015.7437798 architectures. In: ACM SIGCAS conference on computing and
42. Seo Y, Park B, Hinton A, Yoon SC, Lawrence KC (2016) Iden- sustainable societies, pp 389–395, https://doi.org/10.1145/34601
tification of Staphylococcus species with hyperspectral micro- 12.3472309.
scope imaging and classification algorithms. J Food Meas Char- 58. Lee CC et al (2021) Helminth egg analysis platform (HEAP): an
act 10(2):253–263. https://doi.org/10.1007/S11694-015-9301-0/ opened platform for microscopic helminth egg identification and
TABLES/3 quantification based on the integration of deep learning architec-
43. Tchinda BS, Noubom M, Tchiotsop D, Louis-Dorr V, Wolf D tures. J Microbiol Immunol Infect. https://d oi.o rg/1 0.1 016/J.J MII.
(2018) Towards an automated medical diagnosis system for intes- 2021.07.014
tinal parasitosis. Inform Med Unlocked 13(September):101–111. 59. Litjens G et al (2017) A survey on deep learning in medical image
https://doi.org/10.1016/j.imu.2018.09.004 analysis. Med Image Anal 42:60–88. https://doi.org/10.1016/j.
44. Nkamgang OT, Tchiotsop D, Tchinda BS, Fotsin HB (2018) A media.2017.07.005
neuro-fuzzy system for automated detection and classification of 60. Panicker RO, Kalmady KS, Rajan J, Sabu MK (2018) Automatic
human intestinal parasites. Inform Med Unlocked 13(June):81–91. detection of tuberculosis bacilli from microscopic sputum smear
https://doi.org/10.1016/j.imu.2018.10.007 images using deep learning methods. Biocybern Biomed Eng
45. Asefpour Vakilian K, Massah J (2013) An artificial neural network 38(3):691–699. https://doi.org/10.1016/J.BBE.2018.05.007
approach to identify fungal diseases of cucumber (Cucumis sati- 61. Tahir MW, Zaidi NA, Rao AA, Blank R, Vellekoop MJ, Lang
vus L.) plants using digital image processing. Arch Phytopathol W (2018) A fungus spores dataset and a convolutional neural
Plant Prot 46(13):1580–1588. https://doi.org/10.1080/03235408. networks based approach for fungus detection. IEEE Trans Nano-
2013.772321 biosci 20(20):1–10. https://doi.org/10.1109/TNB.2018.2839585
46. Liu J, Dazzo FB, Glagoleva O, Yu B, Jain AK (2001) CMEIAS: 62. Oomman R, Kalmady KS, Rajan J, Sabu MK (2018) Automatic
a computer-aided system for the image analysis of bacterial mor- detection of Tuberculosis bacilli from microscopic sputum smear
photypes in microbial communities. Microb Ecol 41(3):173–194. images using deep learning methods. Integr Med Res 38(3):691–
https://doi.org/10.1007/S002480000004 699. https://doi.org/10.1016/j.bbe.2018.05.007
47. Inayah N, Liebenlito M, Fitriyati N, Monardo K (2020) Clas- 63. Treebupachatsakul T, Poomrittigul S (2019) Bacteria classifica-
sification of falciparum parasite in human red blood cells using tion using image processing and deep learning. 34th Int. Tech.
randomly wired neural network. In: 2018 international conference Conf. Circuits/Systems, Comput. Commun. ITC-CSCC 2019, pp.
on computer, information and telecommunication systems (CITS), 2–4. https://doi.org/10.1109/ITC-CSCC.2019.8793320
pp 2018–2021. https://d oi.o rg/1 0.1 109/C ITSM5 0537.2 020.9 2688 64. Pedraza A, et al (2018) Identification. https://doi.org/10.1117/12.
06 2309488
48. Hung J, et al, Applying faster R-CNN for object detection on 65. Sajedi H, Mohammadipanah F, Rahimi SAH (2019) Actinobac-
malaria images deepali ravel * Agency for Science & Technol- terial strains recognition by machine learning methods. Mul-
ogy ( A * STAR ) Campinas, pp 1–7 timed Tools Appl 78(14):20285–20307. https://doi.org/10.1007/
49. Górriz M, Aparicio A, Raventós B, Vilaplana V, Sayrol E, López- s11042-019-7379-9
Codina D (2018) Leishmaniasis parasite segmentation and clas- 66. Zhou Y et al (2019) Digital whole-slide image analysis for auto-
sification using deep learning. Lect Notes Comput Sci 10945 mated diatom test in forensic cases of drowning using a convo-
LNCS:53–62. https://doi.org/10.1007/978-3-319-94544-6_6 lutional neural network algorithm. Forensic Sci Int 302:109922.
50. Viet NQ (2019) Parasite worm egg automatic detection in micros- https://doi.org/10.1016/j.forsciint.2019.109922
copy stool image based on Faster R-CNN, pp 197–202 67. Qian P et al (2020) Multi-target deep learning for algal detection
51. Yang F, Yu H, Silamut K, Maude RJ, Jaeger S, Antani S (2019) and classification. In: 2020 42nd annual international conference
Parasite detection in thick blood smears based on customized of the IEEE engineering in medicine & biology society (EMBC),
faster-RCNN on smartphones. In: 2019 IEEE Applied Imagery pp 1954–1957. https://doi.org/10.1109/EMBC44109.2020.91762
Pattern Recognition Workshop (AIPR), vol. 2019-Octob, pp 0–3. 04
https://doi.org/10.1109/AIPR47015.2019.9174565 68. Salido J, Sánchez C, Ruiz-Santaquiteria J, Cristóbal G, Blanco S,
52. Ravendran A, de Silva KR, Senanayake R (2016) Moment invari- Bueno G (2020) A low-cost automated digital microscopy plat-
ant features for automatic identification of critical malaria para- form for automatic identification of diatoms. Appl Sci. https://d oi.
sites. In: 2015 IEEE 10th international conference on industrial org/10.3390/app10176033
and information systems (ICIIS), pp 474–479. https://doi.org/10. 69. Holmström O et al (2017) Point-of-care mobile digital microscopy
1109/ICIINFS.2015.7399058 and deep learning for the detection of soil-transmitted helminths
53. Mathison BA et al (2020) Detection of intestinal protozoa in and Schistosoma haematobium. Glob Health Action. https://doi.
trichrome-stained stool specimens by use of a deep convolutional org/10.1080/16549716.2017.1337325
13
Advances Towards Automatic Detection and Classification of Parasites Microscopic Images… 2039
70. Peixinho AZ, Martins SB, Vargas JE, Falcao AX, Gomes JF, 76. Correa I, Drews P, Botelho S, de Souza MS, Tavano VM (2017)
Suzuki CT (2016) Diagnosis of human intestinal parasites by deep Deep learning for microalgae classification. In: 2017 16th IEEE
learning. In: Computational Vision and Medical Image Process- International Conference on Machine Learning and Applications
ing V: Proceedings of the 5th Eccomas Thematic Conference on (ICMLA), pp 20–25. https://doi.org/10.1109/ICMLA.2017.0-183
Computational Vision and Medical Image Processing. VipIM- 77. Zieliński B, Plichta A, Misztal K, Spurek P, Brzychczy-Włoch M,
AGE 2015, Tenerife, Spain, pp 107–112. https://doi.org/10.1201/ Ochońska D (2017) Deep learning approach to bacterial colony
b19241-19 classification. PLoS ONE. https://doi.org/10.1371/journal.pone.
71. López YP, Costa Filho CF, Aguilera LM, Costa MG (2017) Auto- 0184554
matic classification of light field smear microscopy patches using 78. Suwannaphong T, Chavana S, Tongsom S, Palasuwan D, Chalid-
convolutional neural networks for identifying Mycobacterium abhongse TH, Anantrasirichai N, Parasitic egg detection and clas-
tuberculosis. In: 2017 CHILEAN conference on electrical, elec- sification in low-cost microscopic images using transfer learning.
tronics engineering, information and communication technologies 79. Hashem EM, Mabrouk M (2014) A study of support vector
(CHILECON), pp 1–5. https://d oi.o rg/1 0.1 109/C HILEC
ON.2 017. machine algorithm for liver disease diagnosis
8229512 80. Suykens JAK, Vandewalle J (1999) Least squares support vector
72. Zieliski B, Sroka-Oleksiak A, Rymarczyk D, Piekarczyk A, machine classifiers. Neural Process Lett 9(3):293–300. https://d oi.
Brzychczy-Woch M (2020) Deep learning approach to describe org/10.1023/A:1018628609742
and classify fungi microscopic images. PLoS ONE 15(6):1–16. 81. Ren S, He K, Girshick R, Sun J (2017) Faster R-CNN: towards
https://doi.org/10.1371/journal.pone.0234806 real-time object detection with region proposal networks. IEEE
73. Wahid MF, Ahmed T, Habib MA (2019) Classification of micro- Trans Pattern Anal Mach Intell 39(6):1137–1149. https://doi.org/
scopic images of bacteria using deep convolutional neural net- 10.1109/TPAMI.2016.2577031
work. In: 2018 10th international conference on electrical and
computer engineering (ICECE), pp 217–220. https://doi.org/10. Publisher's Note Springer Nature remains neutral with regard to
1109/ICECE.2018.8636750 jurisdictional claims in published maps and institutional affiliations.
74. Ahmed T, Wahid MF, Hasan MJ (2019) Combining deep convo-
lutional neural network with support vector machine to classify Springer Nature or its licensor (e.g. a society or other partner) holds
microscopic bacteria images. In: 2019 international conference on exclusive rights to this article under a publishing agreement with the
electrical, computer and communication engineering (ECCE), pp author(s) or other rightsholder(s); author self-archiving of the accepted
1–5. https://doi.org/10.1109/ECACE.2019.8679397 manuscript version of this article is solely governed by the terms of
75. Orenstein EC, Beijbom O (2017) Transfer learning & deep feature such publishing agreement and applicable law.
extraction for planktonic image data sets. In: 2017 IEEE Winter
Conference on Applications of Computer Vision (WACV), pp
1082–1088. https://doi.org/10.1109/WACV.2017.125
13