Machine Learning Tools for Analyzing Medical Imaging Data

Abstract

Cancer is one of the most fatal disease around the globe. As per estimate around 10 million people have died of cancer in 2020. Cancer is a kind of disease that involves abnormal growth of cells that can spread to other nearby tissues. Cancer cells can affect the other parts of the body by reaching them through the blood flow and lymph system. Cancer is composed of several tissue types and hence their detection and classification can play a significant role in increasing the survival rate. In this research we aim to provide a novel approaches of different machine learning and deep learning algorithms for the detection, and classification of cancer phenotypes. The study has focused mainly on two types of cancer including colorectal and thyroid. Colorectal cancer dataset is composed of histology images which contains seven and nine different classes in two different datasets. Thyroid dataset contains cytology images. Conventional visual inspection is very time consuming and the process can undergo inaccuracies because of the subject-level assessment. In computational pathology, automatic tissue phenotyping in whole slide images (WSIs) of colorectal cancer (CRC) assists pathologists for better cancer grading and prognostication. For colorectal cancer tissue phenotyping we have proposed two different models. One model utilizes the graph based classification by using hypergraph neural network to discriminate between multiple tissue types. While second model is developed by blending convent feature extractor with the broad learning system for CRC classification. For thyroid cancer we first developed the cell level dataset which is annotated by an experienced pathologist. On that dataset transfer learning approach has been applied using multiple pre-trained model. Results have shown that our models have performed very well in identifying and classifying the different labels in the dataset.

Date of Award	May 2021
Original language	American English