Automated detection and classification of Leukemia using deep learning

Lee, Kye Fung (2022) Automated detection and classification of Leukemia using deep learning. Final Year Project, UTAR.

Preview

PDF Download (5Mb) | Preview

Abstract

Leukemia is a type of blood cancer that has been affecting the lives of many. The main procedure to diagnose and classify leukemia is through microscopic examination of blood smears, which can be costly, time-consuming, and labour-intensive. Hence, thisproject aims to produce an efficient way to detect and classify leukemia by using deep learning. In this project, transfer learning is implemented on three pre-trained deep learning models, namely Inception-V3, ResNeXt, and SENet models. They were trained to tackle two main tasks: binary classification between ALL and healthy cells, and 5-class classification between ALL, AML, CLL, CML, and healthy cells. The microscopic image samples of these classes are retrieved from two sources, including the Acute Lymphoblastic Leukemia Image Database 1 (ALL-IDB1) and American Society of Hematology (ASH) ImageBank. It is observed that the SENet model performed the best out of the three, hence it is selected to undergo further fine-tuning to improve its performance. With a slow converging feature selection process added with the dropout regularization technique, the SENet model can achieve an average testing accuracy of 99.84% and 84.48% in binary and 5-class classification problems.

Actions (login required)