Abstract #3512

Test-retest repeatability of convolutional neural networks in detecting prostate cancer regions on diffusion weighted imaging in 112 patients

Amogh Hiremath¹, Rakesh Shiradkar¹, Harri Merisaari^1,2, Prateek Prasanna¹, Otta Ettala³, Pekka Taimen⁴, Hannu J Aronen⁵, Peter J Boström³, Ivan Jambor^2,6, and Anant Madabhushi¹

¹Department of Biomedical Engineering, Case Western Reserve University, Cleveland, OH, United States, ²Department of Diagnostic Radiology, University of Turku and Turku University Hospital, Turku, Finland, ³Department of Urology, University of Turku and Turku University hospital, Turku, Finland, ⁴Institute of Biomedicine, University of Turku and Department of Pathology, Turku University Hospital, Turku, Finland, ⁵Medical Imaging Centre of Southwest Finland, Turku University Hospital, Turku, Finland, ⁶Department of Radiology, Icahn School of Medicine at Mount Sinai, New York, NY, United States

We evaluated the short-term repeatability of convolutional neural networks (CNNs) in detecting prostate cancer (PCa) using DWI collected from patients who underwent same day test-retest MRI scans. DWI was post-processed using monoexponential fit (ADCm). Two models with similar architecture were trained on test-retest scans and short-term repeatability of network predictions in terms of intra-class correlation coefficient (ICC(3,1)) was evaluated. Although the observed ICC(3,1) was high for CNN when optimized for classification performance, our results suggest that network optimization with respect to classification performance might not yield the best repeatability. Higher repeatability was observed at lower learning rates.

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

Join Here