Liver Cancer Detection using Machine Learning: A Review
Harshitha S
4VV20IS404
Dept of ISE,
VVCE, Mysuru
Email:
harshithanayak814@gmail.com
Abstract
Liver cancer is a life-threatening illness and one of the fastest-
growing cancer types in the world. Consequently, the early
detection of liver cancer leads to lower mortality rates. This work
aims to build a model that will help clinicians determine the type
of tumor when it occurs within the liver region by analyzing images
of tissue taken from a biopsy of this tumor. Working within this
stage requires effort, time, and accumulated experience that must
be possessed by a tissue expert to determine whether this tumor is
malignant and needs treatment. Thus, a histology expert can make
use of this model to obtain an initial diagnosis. This study aims to
propose a deep learning model using convolutional neural
networks (CNNs), which are able to transfer knowledge from pre-
trained global models and decant this knowledge into a single
model to help diagnose liver tumors from CT scans.
Keywords - Liver diagnosis, Machine learning, Expert System
Introduction
Cancer is a general term that includes a wide range of diseases that
can affect any part of the body. Cancer is considered a leading
cause of death worldwide, claiming nearly 10 million lives in 2020,
or approximately 1 in 6 deaths. One of the most common types of
cancer is liver cancer. The American Cancer Society estimates that
about 30,520 people (20,420 men and 10,100 women) will die of
liver cancer [1] at the end of 2022. Liver cancer causes more deaths
than other types of cancer, and the disease is usually diagnosed
when it is in its advanced stages. Therefore, diagnosing this disease
at an early stage leads to better treatment options. This paper
proposes an efficient method for the early detection of liver cancer.
Non-invasive procedures such as CT scans, MRI, and ultrasound
are used to identify liver cancer. The CT scan is a type of X-ray
that produces detailed images of your body. This scan can provide
information on the size, shape, and location of any type of liver or
abdominal tumor, as well as the blood arteries that surround them.
CT scans can also be used to precisely direct a biopsy needle into
a potentially cancerous tumor [2]. MRI scansproduce
comprehensive images of the soft
tissues of the body. In addition, they employ radio waves and
powerful magnets rather than X-rays. MRI scans can be very useful
in the investigation of liver tumors. They can occasionally
distinguish between benign and malignant tumors. They can also
be used to detect blockages in the blood arteries in and around the
liver, as well as to determine whether liver cancer has migrated to
other parts of the body [3,4]. Ultrasound
is usually used as the initial diagnosis when examining the liver. It
generates an image on a computer screen by using sound
waves. This test can detect liver tumors, which can then be
evaluated for cancer if necessary [5].
Related Work:
The liver is the second largest organ in the body, located on the
right side of the abdomen, and weighs about three pounds. The
liver has two lobes, right and left, and is in contact with the
gallbladder, pancreas, and intestines. Several organs are involved
with the liver. Cancer in the liver may be primary (originating from
various cells that compose the liver), secondary, or metastatic
(caused by cancerous cells from other organs). Malignant
hepatocellular carcinoma (HCC) is the most typical primary liver
disease among all liver cancers. The second most common global
disease is liver cancer. According to data from the World Health
Organization (WHO), it accounted for 8.8 million deaths in 2015,
out of which 788,000 deaths were caused by carcinoma [1]. The
American Cancer Society (ACS) predicted that around 20,710 new
cases in the USA would be diagnosed in the year (29,200 in men
and 11,510 in women). Out of these, 28,920 people (19,610 men
and 9310 women) died because of primary carcinoma and
intrahepatic epithelial canal cancer in 2017 [2] Researchers in the
past have investigated this topic. Masuda et al. [1] showed that by
enhancing contrast levels and using the expectation maximization
of the posterior marginal algorithm as well as a shape constrain,
liver tumors can be detected even in poor CT images. Pescia et al.
[2] proposed the use of advanced non-linear machine learning
techniques to determine the optimal features, as well as the
hyperplane that uses these features to separate tumor voxels from
healthy liver tissues. Bilello et al. [3] showed that liver. In [4], the
authors proposed a segmentation model based on triplane-views,
comprising axial, sagittal and coronal plane axes. These
interconnected planes form an input stream to feed the con-vNet
model. Multi-axis input streams resulted in the richness of input
information to assist segmentation results.
Automatic tumor segmentation methods do hold a certain degree
of dominance if compared to manual methods because of some
evident reasons. Traditional segmentation techniques are
prolonged with a time constraint [4]. These are tedious, and
computationally expensive as well in some cases. Another
drawback of these techniques is that sometimes these approaches
overlook an existing tumor and fail to detect it due to its
microscopic size. Also, traditional segmentation does not
efficiently differentiate between the tumor cell and normal cell due
to analogy [5]. There can be low contrast levels, equipment issues,
blurred edges, smudged cells, ambiguous boundaries, and other
spatial factors that impact handpicked features in traditional
segmentation [6]. This provides more room for error rates and
inaccurate results. Irregularity in the shape and size of the tumor
also varies from person to person. In low-level images, it becomes
a challenging task to detect and segment out tumors effectively. On
the other hand, the modality of automatic methods is speedy,
precise, and efficient. Automatic methods have highly impacted
the procedure of liver tumor detection and segmentation for being
quick. Therefore, research has taken a shift from traditional
segmentation methods to automatic segmentation methods. The
1|Page
optimal design of automatic liver tumor segmentation is of much
more use today rather than a well-built design of traditional liver
segmentation.
CNN based Feature Extraction In this implementation,
Convolutional Neural Networks (CNN) based feature extraction
method is adopted to obtain the more accurate features. These
features are like diameter, length, width, area and perimeter of the
leaves. Flavia data set is applied as the input training images for
the CNN. The performance of CNN based approach is far better
than the hand-crafted features of the previous approaches. The
CNN architecture is shown in the figure
This CNN algorithm was implemented on falvia dataset used as the
training images with 1800 leaves to extract the features from 32
different species. Here we have used 4 layer CNN model. The CNN
architecture is shown in figure 3.1 the network layers are given
below. i) Input Layer The input data set vector (p) is fed to this
layer. Its dimensions are 5x1. ii) Radial Basis Layer In this layer,
the vector distance (n) is calculated between the vector p and the
weight vector ω. It is implemented as the dot product. ni = || ωi −
pi ||·∗pi (1) radial basis (ni) = exp (-ni 2 ) (2) Where i is the row
number of the vector. iii) Some characteristics of Radial Basis
Layer ni=1, if pi is identical with ωi In this case, its output weights
in the competitive layer will pass their values to the competitive
function. iv) Competitive Layer The output vector d = n.*Mi . (3)
The competitive function C is calculated from d. C equals to 1 at
the largest element of d and 0 elsewhere. It is used as the index to
look for the scientific name of this liver. ii. Machine Learning
based classification Classification is the final step in the automatic
liver recognition process. This classification process can be
implemented using the various machine learning algorithms like
Artificial Neural Network (ANN), Support Vector Machines
(SVM), K-Nearest Neighbour (KNN) and Naive Bayes (NB).
SVM [7] classification is characterized by an isolating hyper plane.
The concept of SVM classification can be understand easily from
the following training data set which is described into two different
classes, D = {(x1 , y 1 ), (x2 , y 2 ) .. (x k , y k ), xi ϵ Rm, yi ϵ [+1,
-1] (4) + b = 0 (5) The set of vectors is said to be optimally
separated by the hyper plane (Eq. (5)) if the separation error is ‘0’
and the distance function is maximal. Now a day’s ANN [8] is the
most widely used research method in the machine learning process.
This concept of ANN has the similarity with the human brain
neurons system. Performance of ANN depends on the neuron count
and hidden layers. In this implementation, feed forward neural
network with a single hidden layer was used. In training phase, the
system uses the default Scaled Conjugate Gradient function. The
minimum gradient is used for stopping criteria. Also, KNN method
can be used for the liver species recognition. KNN algorithm [9-
10] is based on distance measure metrics in feature space. KNN
classifier at first determines k nearest neighbors and then
determines labels for the sample based on neighbor weight.
Bayesian classifier is a statistical approach. It performs the
classification based on the concept of probability. Accuracy of
Bayesian classifier method is less due to its conditional
independence of Bayesian probability. 4. SIMULATION
RESULTS The experiment is carried out on the 10 different liver
species [16] as shown in table 1. The table 2 shows the
classification report (Precision, Recall, F1-Score & Support) [17]
obtained using the proposed algorithm. The following are the
metrics that are used in the classification process. i. Precision is the
ratio of correctly predicted positive observations to the total
predicted positive observations. Precision (P) = TP/ (TP+FP)
TPTrue Positives FPFalse Positives ii. Recall is the fraction of
positives that are correctly identified. Recall(R) = TP/ (TP+FN)
FN False Negatives iii. F1-Score is created by finding the
harmonic mean of precision and recall. F1-Score = 2*(R * P)/(R +
P) iv. Support is the number of occurrences in each class.
Proposed system
Proposed system is an automation Liver Cancer prediction using
deep learning techniques. The motive of this method is to initiate
deep learning based framework with CNN architecture and execute
the investigation for image classification of Liver Cancer. This
efficient deep learning process will provide us some insights into
the neuronal pathology of autistic children, and also help the
diagnosis of the early stage of Liver Cancer in children.
We have taken some initial results on detecting Liver Cancer from
Brain X-Rays using a deep-learning model. We neded
demonstrated significant improvement in performance over Liver
Cancer -Net, the only publicly maintained tool for classification of
LIVER CANCER X-rays, on the same Brain-xray-dataset. The
results look promising, though the size of the publicly available
dataset is small. We plan to further validate our approach using
larger LIVER CANCER-ray image datasets and clinical trials.
2|Page
 Methodologies: Step2 : ReLU Layer
In this layer, we remove every negative value from the filtered
images and replaces them with zeros.It is happening to avoid the
values from adding up to zero.

Rectified Linear unit(ReLU) transform functions only activates a

node if the input is above a certain quantity. While the data is below
zero, the output is zero, but when the information rises above a
threshold. It has a linear relationship with the dependent variable.

Step3: Pooling Layer

In the layer, we shrink the image stack into a smaller size. Pooling
is done after passing by the activation layer. We do by
implementing the following 4 steps:

o Pick a window size (often 2 or 3)

o Pick a stride (usually 2)
o Walk your Window across your filtered images
o From each Window, take the maximum value
Collection datasets:
• We are going to collect datasets for the prediction from
the kaggle.com
• The data sets consists of many Classess Step4: Fully Connected Layer

Data Pre Processing:

The last layer in the network is fully connected, meaning that
• In data pre-processing we are going to perform some neurons of preceding layers are connected to every neuron in
image pre-processing techniques on the selected data subsequent layers.

• Image Resize This mimics high-level reasoning where all possible pathways
from the input to output are considered.
• And Splitting data into train and test
Then, take the shrunk image and put into the single list, so we have
Data Modelling: got after passing through two layers of convolution relo and
pooling and then converting it into a single file or a vector.
• The splitted train data are passed as input to the CNN
algorithm, which helps in training.
• The trained liver image data evaluated by passing test
data to the algorithm
• Accuracy is calculated

Build Model:
• Once the data is trained and if it showing the accuracy
rate as high, then we need to build model file

CNN:
Convolutional Neural Networks have the following layers:

o Convolutional
o ReLU Layer Convolutional Neural Networks have the following layers:
o Pooling
o Fully Connected Layer o Convolutional
o ReLU Layer
o Pooling
Step1: Convolution Layer: o Fully Connected Layer
• Convolutional neural networks apply a filter to an input to create a
feature map that summarizes the presence of detected features in
the input.

3|Page
 Then, take the shrunk image and put into the single list, so we have
got after passing through two layers of convolution relu and
pooling and then converting it into a single file or a vector.

Step1: Convolution Layer:

• Convolutional neural networks apply a filter to an input to create a
feature map that summarizes the presence of detected features in
the input.
Step2 : ReLU Layer

In this layer, we remove every negative value from the filtered

images and replaces them with zeros.It is happening to avoid the
values from adding up to zero.

Rectified Linear unit(ReLU) transform functions only activates a

node if the input is above a certain quantity. While the data is below
zero, the output is zero, but when the information rises above a
threshold. It has a linear relationship with the dependent variable.

Step3: Pooling Layer

In the layer, we shrink the image stack into a smaller size. Pooling
is done after passing by the activation layer. We do by
implementing the following 4 steps:

o Pick a window size (often 2 or 3)

o Pick a stride (usually 2)
o Walk your Window across your filtered images
o From each Window, take the maximum value

Step4: Fully Connected Layer

The last layer in the network is fully connected, meaning that

neurons of preceding layers are connected to every neuron in
subsequent layers.

This mimics high-level reasoning where all possible pathways

from the input to output are considered.

4|Page
Table1.1 Summary of the papers reviewed

Paper datatype result limitation

An Efficient Liver Xray image datasets They perform clustering on our collected It is evident that the accuracy
Tumor Detection using dataset and extract LBP features as well as of SVM is 85% and the
Machine Learning HOG features from these clusters. Ensemble is 49%. Both of
Furthermore, we perform classification these values are low as
which is based on these extracted features accuracy.
using KNN. Furthermore, they have
compared our results with two classifiers
namely SVM and Ensemble to achieve a
better understanding. Our proposed
technique outperformed existing
techniques and showed encouraging
results when compared to other methods.
Accuracy :0.85

Liver Cancer Detection CT images of the liver between method includes the successful extraction The datasets used for this
using Hybridized Fully liver metastases of colorectal of features from Inception combined with project are very less in number
Convolutional Neural cancer and benign cysts has been residual and pre-trained weights. Feature and feature extraction results
Network based on analyzed. maps have been consistent with the are not that much accuracte.
Deep learning original image voxel features, and The
framework importance of features seemed to
represent the most relevant imaging
criteria for every class.
Accuracy:0.92

Deep Learning magnetic resonance imaging This study proposed a more efficient Despite the Res-UNet
Approach for Liver (MRI), computer tomography method for segmenting liver and tumors producing very promising
and Tumor (CT) from CT image volumes using a hybrid results, there are a few
Segmentation in CT ResUNet model, combining the ResNet limitations. We may get
Images Using and UNet models to address this gap. around these limitations by
ResUNet planning for more epochs,
using more data, using
different datasets, or using
different preprocessing
strategies.

5|Page
Liver Tumor Detection 3D computed tomography (CT) The mean VO of tumor segmentation is The distance between f(x)
and Segmentation images 67.15%. The improvement over and the label indicates the
using Kernel-based traditional ELM in mean VO is 5%. class type. The oneclass
Extreme Learning Compared with the performance degrades ELM shows that it is
Machine slightly possible to detect tumors
even there is no tumor data
for training. We have also
compared the performance
of one class tumor detection
and two-class ELM using
cross validation detection.

Liver Tumor CT scan Image datasets proposed method hasimproved the Test size f data is less
Segmentation using segmentation rate by providing
Resnet based Mask-R- segmentation accuracy with average dice
CNN coefficient, JC, VOE and RVD of 0.95,
0.41, 0.12, and 0.15, respectively and
lesser training time due to converted CT
image slices instead of training on CT
volumes.

Automatic Detection CT scans The proposed model achieved an accuracy Activate function can be
of Liver Cancer Using of 99.5% on both datasets, which is better used according to data.
Hybrid Pre-Trained than the results of other models used in
Models this field

6|Page
Conclusion
we used CNN and deep learning methods for the feature extraction
and classification of liver diseasespecies. The different machine
learning algorithms are ANN, SVM, KNN and Naive Bayes (NB).
We have considered the Flavia dataset in this method. This data
sets used for both and training and testing purpose. It has been
achieved an accuracy of 98%. All the performance metrics like
precision, recall, F1-score and support are calculated. Also the
achieved training and validation accuracies are nearly equal.
Images used in the training purpose are small and gray scale
images. As a future work it is possible to implement the colour
image classification for liver diseaserecognition.
This paper presents the convolutional neural network (CNN)
method for liver cancer and lesion identification and segmentation.
Various layers in the neural network are utilized to extract features
of medical images to improve the accuracy of the detection of
medical images. 2D feature maps are combined with several slices
in the featureextraction process. The algorithm showed very
accurate liver volume measurements of 97.22%. The study showed
a high accuracy of the segmentation method had an average Dice
coefficient of 0.92. The results show that the FCN produces the
best results with data changes, adjacent slices, and appropriate
class weights. Note that a limited dataset and testing have been
done 3 fold cross-validation. A Convolutional Neural Network is
trained to identify tumors and healthy voxels on all base-line liver
maskings. CNN is used to generate the follow-up segmentation of
tumors as a voxel classifier. The segmentation leaks are then
eliminated in the resulting segmentation so that the final results can
be obtained. The proposed
HFCNN method has high accuracy in terms of identifying liver
tumors.
References
1. Guo, L. H., Wang, D., Qian, Y. Y., Zheng, X., Zhao, C. K., Li,
X. L., ... & Xu, H. X. (2018). A two-stage multi-view learning
framework based on computeraided diagnosis of liver tumors with
contrast-enhanced ultrasound images. Clinical hemorheology and
microcirculation, 69(3), 343-354.
2. Vorontsov, E., Cerny, M., Régnier, P., Di Jorio, L., Pal, C. J.,
Lapointe, R., ... & Tang, A. (2019). Deep learning for automated
segmentation of liver lesions at CT in patients with colorectal
cancer liver metastases. Radiology: Artificial Intelligence, 1(2),
180014.
3. Lakhani, P., & Sundaram, B. (2017). Deep learning at chest
radiography: automated classification of pulmonary tuberculosis
by using convolutional neural networks. Radiology, 284(2), 574-
582.
4. Yan, K., Wang, X., Lu, L., & Summers, R. M. (2018).
DeepLesion: automated mining of large-scale lesion annotations
and universal lesion detection with deep learning. Journal of
medical imaging, 5(3), 036501.
5. Christ, P. F., Elshaer, M. E. A., Ettlinger, F., Tatavarty, S.,
Bickel, M., Bilic, P., ... & Sommer, W. H. (2016, October).
Automatic liver and lesion segmentation in CT using cascaded
fully convolutional neural networks and 3D conditional random
fields. In International Conference on Medical Image Computing
and Computer-Assisted Intervention (pp. 415-423). Springer,
Cham.
6. Hu, P., Wu, F., Peng, J., Liang, P., & Kong, D. (2016).
Automatic 3D liver segmentation based on deep learning and
globally optimized surface evolution. Physics in Medicine &
Biology, 61(24), 8676.
7. Men, K., Chen, X., Zhang, Y., Zhang, T., Dai, J., Yi, J., & Li,
Y. (2017). Deep deconvolutional neural network for target
segmentation of nasopharyngeal cancer in planning computed
tomography images. Frontiers in oncology, 7, 315.
8. Wu, K., Chen, X., & Ding, M. (2014). Deep learning based
classification of focal liver lesions with contrast-enhanced
ultrasound. Optik, 125(15), 4057-4063.
9. Trivizakis, E., Manikis, G. C., Nikiforaki, K., Drevelegas, K.,
Constantinides, M., Drevelegas, A., & Marias, K. (2018).
Extending 2-D Convolutional Neural Networks to 3-D for
Advancing Deep Learning Cancer Classification With Application
to MRI Liver Tumor Differentiation. IEEE journal of biomedical
and health informatics, 23(3), 923-930.
10. Chlebus, G., Schenk, A., Moltz, J. H., van Ginneken, B., Hahn,
H. K., & Meine, H. (2018). Deep learning based automatic liver
tumor segmentation in CT with shape-based post-processing.
11.https://www.selleckchem.com/blog/Magnetic-resonance-
imaging-visible amonafide-eluting-alginate-microspheres-novel-
drug-carrier-for-targeted arterial-infusion-chemotherapy-for-liver-
tumors.html
12. Ben-Cohen, A., Klang, E., Kerpel, A., Konen, E., Amitai, M.
M., & Greenspan, H. (2018). Fully convolutional network and
sparsity-based dictionary learning for liver lesion detection in CT
examinations. Neurocomputing, 275, 1585-1594.
13. Tsuboi, A., Oka, S., Aoyama, K., Saito, H., Aoki, T., Yamada,
A., ... & Koike, K. (2020). Artificial intelligence using a
convolutional neural network for automatic detection of small‐
bowel angioectasia in capsule endoscopy images. Digestive
Endoscopy, 32(3), 382-390.
14. Vivanti, R., Joskowicz, L., Lev-Cohain, N., Ephrat, A., &
Sosna, J. (2018). Patient-specific and global convolutional neural
networks for robust automatic liver tumor delineation in follow-up
CT studies. Medical & biological engineering & computing, 56(9),
1699-1713.
15. Ben-Cohen, A., Klang, E., Raskin, S. P., Amitai, M. M., &
Greenspan, H. (2017, September). Virtual PET images from CT
data using deep convolutional networks: initial results. In
International Workshop on Simulation and Synthesis in Medical
Imaging (pp. 49-57). Springer, Cham.
16. Hassanzadeh, T., Essam, D., & Sarker, R. (2020, March).
EvoU-Net: an evolutionary deep fully convolutional neural
network for medical image segmentation. In Proceedings of the
35th Annual ACM Symposium on Applied Computing.
7|Page