Thesis Format
Integrated Article
Degree
Master of Science
Program
Medical Biophysics
Supervisor
Ward, Aaron D.
Abstract
Radical prostatectomy (RP) is a common treatment for prostate cancer. We used RP whole-mount tissue samples from 68 patients stained with haematoxylin and eosin to create cancer maps using four pretrained networks: AlexNet, NASNet, VGG16 and Xception, to classify regions of interest (ROIs) as cancer or non-cancer. Models were trained on either raw images or as tissue component maps (TCMs) containing nuclei, lumina and stroma/other components generated from a trained U-Net.
All models performed similarly; however, VGG16 trained on raw images performed with the highest area under the receiver operating characteristic curve (AUC) of 0.994 (95% confidence interval 0.992-0.996). Ensemble models using models trained on raw images performed with the lowest false positive rates. All models had high false negative rate for Gleason 5 cancer and those trained on raw images performed with higher AUC than models trained on TCMs.
Summary for Lay Audience
Prostate cancer is the second most frequently diagnosed cancer in men worldwide. Although this is a very treatable disease, survival rates of those with aggressive cancer is much lower than those with a less aggressive form. One treatment option available for those with gland localized cancer is radical prostatectomy, where the entire gland is surgically removed.
Expert contouring and grading of the aggressiveness of cancer on tissue slides could provide valuable information on patient prognosis, as well as guide treatment plans after the surgery. Unfortunately, detailed contours are incredibly time consuming and impractical to do in a clinical setting, so there is an unmet need for an automated cancer contouring algorithm. As a part of tissue processing, dyes are applied so the tissue is visible under a microscope. Application of the dye and the tissue properties itself can affect how intensely the dye appears, making it difficult for algorithms to be robust to the colour intensities.
In this thesis we use deep learning for cancer detection. Deep learning is a technique that teaches an algorithm to find patterns in a set of data and is a relatively young field with promising results in computer vision. Deep learning has been used to detect cancer in radical prostatectomy histology, but the papers we surveyed had a limited number of patients, or only tested one deep learning model. That is why we are comparing four deep learning models in cancer detection.
We used these four deep learning models to generate cancer maps from tissue slides and found that these models perform comparatively to each other. We produced accurate cancer maps but further research is needed to analyze potential impact on the clinical and research workflow.
Recommended Citation
Huang, Laurie, "Cancer Detection in Radical Prostatectomy Histology using Convolutional Neural Networks" (2020). Electronic Thesis and Dissertation Repository. 7388.
https://ir.lib.uwo.ca/etd/7388
Creative Commons License
This work is licensed under a Creative Commons Attribution-Noncommercial 4.0 License