You are on page 1of 15

2014 ISSNIP-IEEE Biosignals and Biorobotics Conference

Biosignals and Robotics for Better and Safer Living (BRC)

A Telediagnostic System for Automatic Detection


of Lesions in Digital Mammograms

Luis Claudio de Oliveira Silva, Allan Kardec Barros, Ewaldo Eder


Carvalho Santana

UNIVERSIDADE FEDERAL Programa de Pós-Graduação em Engenharia de Processamento da Informação


DO MARANHÃO - UFMA Eletricidade - UFMA Biológica
Summary
● Introduction
● Materials and methods
● Results and discussion
● System screenshots
● Conclusion
Introduction
● Breast cancer is the second type of cancer that more affects women
in the world
● Mammography is the most accurate exam for diagnostic
● The Limitation of human visual perception together with work
overload may cause misinterpretation
● This work proposes the modeling and implementation of a Web based
telediagnostic system for automatic detection of lesions in
mammograms using ICA and SVM
Materials and Methods
● The software system is a telediagnostic system, able to analyse
mammograms images remotely
● The pre-diagnosis of the mammogram is generated automatically and
could be confirmed by a specialist

Figure 1: General vision of the system proposed


Materials and Methods
System funcionalities
● MySQL DMBS for storage
the patient anonimous data
● J2EE 6.0 with Apache
Tomcat for the JSP/HTML
web pages
● C++ programs for
pre-processing,
segmentation, classification
of the images
Figure 2: UML use case diagram

5
Materials and Methods

Figure 3: UML classes diagram


6
Operations of the mammography
image processing module

Figure 4: UML Activity diagram of


processing module for mammographic
images.
Results of pre-processing,
segmentation, and classification
Training and test algorithm
● Patches of 50x50 pixels were extracted from mini-MIAS database
● A set of basis function was obtained from the extracted patches using
FastICA algorithm (MATLAB)
● The projection coefficients of each sample of mammography image
were used in the SVM
● The test was conducted using Leave One Out Cross Validation
(LOOCV) and 10 Fold Cross Validation (k-Fold)

Suspicious
regions
samples
ICA
Normal
regions
samples
Results and Discussion
● Performance results for suspicious images detection methods

Method Sensitivity
Lin 90.68%
Campos 88.30%
Martins 86.00%
Shanthi 88.59%
Our method 92.17%

● The system prototype was installed on a machine with Intel Core i5


processor containing two cores of 2.75 GHz and 4 Gigabytes of RAM.
In this machine was installed the Apache Tomcat Web, simulating a
production Web server.
System Screenshots

Login
System Screenshots

Submit image for analysis


System Screenshots

Result of processing (and confirmation by specialist)


Conclusion
● This paper presents a Web system for remote diagnosis of
mammographic images based on independent component analysis
and support vector machine.
● We obtained 92.17% of correct classification rate in the system's
prototype
● Future works:
– Use others database of mammographic images
– Use others techiniques of feature extraction and classifications
machines (LDA, ANN, BNN, etc.)
– Enhance the detection algorithm to be used with dense breasts
– Improve the graphical user interface for the specialist
– Classify the type of lesion
That's all!!!

Thanks!!!

15

You might also like