Automatic detection and classification of abnormal human blood cells using computer vision and deep learning

This research proposal aims to enhance the performance and add more features to the automated microscope system that is being developed by Smart Labs ltd. This research goal is to increase the overall accuracy of the system while running in real-time. The current prototype has an accuracy of 91% and can process 17 frames per second. Moreover, it can only classify 2 types of cell abnormalities using traditional image processing techniques. This proposal aims to: 1) increase the accuracy from 91% to at least 95%, 2) increase the frame rate to be able to run at 40 frames per second, and 3) be able to classify a minimum of 10 abnormal cell types. The output of this research will enable Smart Labs to process the slides in real-time as they continue moving under the microscope’s objective lens, as well as being able to classify various blood cell abnormalities. This is crucial in having a commercial product that can be presented to labs in Canada and the international medical sector.

Faculty Supervisor:

Mohamed Shehata

Student:

Reece Walsh

Partner:

Smart Labs Ltd.

Discipline:

Computer science

Sector:

Professional, scientific and technical services

University:

Program: