BioMedical Sciences Research Forum 2023-24
Wednesday, Oct. 11, 12-2 p.m.
5M101, Faculty of Medicine; online
Title: The Future of Breast Cancer screening
Presenter: Professor Edward Kendall PhD.
Biomedical Sciences Division, Faculty of Medicine
The situation: Breast cancer prevalence is about 0.13% of the adult female population and about 0.4% of women aged 50 and older. Early detection is key to positive outcome and this led to screening programs for higher risk women (aged 50 and older). The programs operate with a sensitivity near 87% and a specificity near 89%. This means that there are a considerable number of false negatives and false positives. Our aim is to eliminate these.
please contact email@example.com for Webex details
The analysis: The vast number of screening program errors arise from misinterpretation. That is, the evidence exists in the mammogram but the reader incorrectly assigns the case. Distractions, lack of conspicuity and overlap with normal variants all contribute to the errors. So far, tinkering with the existing system to reduce these errors has enjoyed limited success.
The intervention: We propose to replace the screening radiologist function with an artificial intelligence based binary classification scheme. Our current offering uses a convolutional neural network with connected layers, the so called deep-learning option. You will see that this combination significantly outperforms the conventional program… in the laboratory.
This lecture will describe the screening modality, examine the type of data it produces and highlight why interpretation is so challenging. We will also look at why simply examining intensity based images may not be the best option for detection and we will explore some alternatives. Finally we will look at how CNNs use features to achieve highly accurate classifications and what are its limitations.
There will be a pizza lunch provided for attendees.
Presented by Division of Biomedical Sciences