Abstract #0821

End-to-end Deep Learning Strategy To Segment Prostate Cancer From Multi-parametric MR Images

David Hoar¹, Peter Lee², Alessandro Guida³, Steven Patterson³, Chris Bowen^3,4, Jennifer Merrimen⁵, Cheng Wang⁵, Ricardo Rendon⁶, Steven Beyea^3,4, and Sharon Elizabeth Clarke^3,4

¹Department of Electrical and Computer Engineering, Dalhousie University, Halifax, NS, Canada, ²Faculty of Computer Science, Dalhousie University, Halifax, NS, Canada, ³Biomedical Translational Imaging Centre, Nova Scotia Health Authority and IWK Health Centre, Halifax, NS, Canada, ⁴Diagnostic Radiology, Dalhousie University, Halifax, NS, Canada, ⁵Pathology, Dalhousie University, Halifax, NS, Canada, ⁶Urology, Dalhousie University, Halifax, NS, Canada

The purpose of this study was to develop a convolutional neural network (CNN) for dense prediction of prostate cancer using mp-MRI datasets. Baseline CNN outperformed logistic regression and random forest models. Transfer learning and unsupervised pre-training did not significantly improve CNN performance; however, test-time augmentation resulted in significantly higher F1 scores over both baseline CNN and CNN plus either of transfer learning or unsupervised pre-training. The best performing model was CNN with transfer learning and test-time augmentation (F1 score of 0.59, AUPRC of 0.61 and AUROC of 0.93).

This abstract and the presentation materials are available to members only; a login is required.

Join Here