Applications of Machine Learning Techniques in Disease Classification From Medical Images: Recent Practices and Future Challenges

Applications of Machine Learning Techniques in Disease Classification From Medical Images: Recent Practices and Future Challenges

Bikesh Kumar Singh (National Institute of Technology Raipur, India) and Satya Eswari Jujjavarapu (National Institute of Technology Raipur, India)
Copyright: © 2018 |Pages: 20
DOI: 10.4018/978-1-5225-4969-7.ch016
OnDemand PDF Download:
List Price: $37.50
10% Discount:-$3.75


Machine learning techniques such as artificial neural network (ANN), support vector machine (SVM), radial basis function network (RBFN), random forest (RF), naive Bayes classifier, etc. have gained much attention in recent years due to their widespread applications in diverse fields. This chapter is focused on providing a comprehensive insight of various techniques employed for key areas of medical image processing and analysis. Different applications covered in this chapter include feature extraction, feature selection, and cancer classification in medical images. The authors present current practices and evaluation measures used for objective evaluation of different machine learning methods in context to above-mentioned applications. Various factors associated with acceptance/rejection of such automated systems by medical research community are discussed. The authors also discuss how the interaction between automated analysis systems and medical professionals can be improved for its acceptance in clinical practice. They conclude the chapter by presenting research gaps and future challenges.
Chapter Preview

Cancer Detection And Classification Based On Medical Images: Introduction To Computer Aided Diagnosis (Cad) System

In order to reduce the number of unnecessary biopsies by automatic detection and diagnosis of cancer, several researchers have investigated computer aided diagnosis (CAD) systems. CAD systems can be an effective tool for diagnosis of cancer due to following reasons:

  • 1.

    It can assist the physicians in decreasing observational oversights (i.e. decreasing the false negative rate), in the detection and classification of breast cancer in early stages and thus reducing death rates due to cancer.

  • 2.

    Interpretation of medical images varies from expert to expert and is subject to expert’s variability. CAD systems can improve inter-reader and intra-reader variability by providing more objective evidences for diagnosing cancer.

  • 3.

    It can provide a second opinion to the physicians by detecting and classifying masses automatically.

  • 4.

    It is less time consuming and less expensive compared to conventional methods used to classify cancer.

  • 5.

    It can be used to extract and select most appropriate features suitable for classification of cancer.

Figure 1 shows the general layout of methodology used in CAD systems. Various stages of the CAD system are summarized below:

Data Collection

The first step in development of CAD systems for medical image analysis is to collect reliable and authentic image data in consultation of expert medical professionals. Subjects involved in the experiments must be made aware of the experimental protocol and side effects of the data acquisition process. Informed consent of the participants is usually obtained before acquiring data. Approval of institute ethical committee/international review board should be obtained in prior. To develop a reliable CAD system the images collected should be accompanied by patient’s demographics, criterions adopted for inclusion/rejection of the participants, ground truth validations and pathology reports. Ground truth can be generated by experts in case of segmentation problems wherein the radiologists can manually plot the contours of the lesion. The manual delineations thus provided by the expert radiologists can be used as reference for performance evaluation of segmentation techniques. On the other hand, for classification problems, ground truth is required to confirm whether the detected lesion is benign (low risk) or malignant (high risk). This information can be obtained from expert radiologists, biopsy reports and other pathology reports.

Figure 1.

A general layout of CAD system


Some of the application areas of CAD systems which can be applied to different types of medical images are: classification of lesions into malignant or benign in medical imaging modalities such as breast mammography, breast ultrasound, lung computed tomography images, chest radiographs, brain computed tomography images, brain magnetic resonance imaging, detecting calcifications in mammogram, hairline fractures in x-ray images etc.

Complete Chapter List

Search this Book: