A Comprehensive Study On The Applications Of Machine Learning For Diagnosis Of Cancer

TL;DR

This paper develops and evaluates machine learning algorithms for early diagnosis of breast cancer, melanoma, and lung cancer, achieving high accuracy and low false rates, with a mobile app and an online collaborative platform.

Contribution

It introduces new ML-based diagnostic algorithms for multiple cancers, integrating image processing and mobile technology for early detection.

Findings

Breast cancer diagnosis accuracy: 91-95%

Melanoma diagnosis accuracy: 93%

Lung cancer diagnosis accuracy: 94%

Abstract

Collectively, lung cancer, breast cancer and melanoma was diagnosed in over 535,340 people out of which, 209,400 deaths were reported [13]. It is estimated that over 600,000 people will be diagnosed with these forms of cancer in 2015. Most of the deaths from lung cancer, breast cancer and melanoma result due to late detection. All of these cancers, if detected early, are 100% curable. In this study, we develop and evaluate algorithms to diagnose Breast cancer, Melanoma, and Lung cancer. In the first part of the study, we employed a normalised Gradient Descent and an Artificial Neural Network to diagnose breast cancer with an overall accuracy of 91% and 95% respectively. In the second part of the study, an artificial neural network coupled with image processing and analysis algorithms was employed to achieve an overall accuracy of 93% A naive mobile based application that allowed people to take diagnostic tests on their phones was developed. Finally, a Support Vector Machine algorithm incorporating image processing and image analysis algorithms was developed to diagnose lung cancer with an accuracy of 94%. All of the aforementioned systems had very low false positive and false negative rates. We are developing an online network that incorporates all of these systems and allows people to collaborate globally.