Pd45-01 Highlighting Regions of Interest in Prostate Cancer AIded by Synthetically Generated Tissue Biopsies

You have accessJournal of UrologyProstate Cancer: Staging II (PD45)1 May 2024PD45-01 HIGHLIGHTING REGIONS OF INTEREST IN PROSTATE CANCER AIDED BY SYNTHETICALLY GENERATED TISSUE BIOPSIES Derek Van Booven, Cheng-Bang Chen, Sheetal Malpani, Oleksandr Kryvenko, and Himanshu Arora Derek Van BoovenDerek Van Booven , Cheng-Bang ChenCheng-Bang Chen , Sheetal MalpaniSheetal Malpani , Oleksandr KryvenkoOleksandr Kryvenko , and Himanshu AroraHimanshu Arora View All Author Informationhttps://doi.org/10.1097/01.JU.0001008792.09108.b4.01AboutPDF ToolsAdd to favoritesDownload CitationsTrack CitationsPermissionsReprints ShareFacebookLinked InTwitterEmail Abstract INTRODUCTION AND OBJECTIVE: Prostate cancer (PCa) poses significant challenges for timely diagnosis and prognosis, leading to high mortality rates and elevated treatment costs. While recent machine learning advancements in digital imagery offer potential for the creation of automated and objective assessment pipelines that can reduce human capital and resource costs, the scarcity and bias of training data remain challenging. The data is often biased, lacking diversity, and not readily available. To address this limitation, we employed generative adversarial network (GAN) models to produce high-quality synthetic images of different PCa grades (radical prostatectomy (RP) and needle biopsies), which were customized to account for the heterogeneity associated with each Gleason grade and required a significantly smaller training data sample size. METHODS: PCa biopsies from 500 individuals were sourced from the cancer genome atlas (TCGA), and 3,949 needle biopsies were obtained from the PANDA Kaggle competition (original source was the Radbound University Medical Center and Karolinska Institute). Following multiple Quality Control optimization and standard color normalization to image data, an AI training database of tumor areas was created based on pathologist-annotated regions of interest according to Gleason pattern (n=108 RP sections, n=500 needle biopsies). A deep convolutional GAN was created from scratch to generate 5000 RP section patches and 2000 needle biopsy patches for each Gleason pattern. These synthetic patches underwent evaluation using a convolutional neural network for automated classification of tumor/normal areas highlighting regions of interest. RESULTS: CNN's accuracy improved in GS3 from 0.53 to 0.67 (p=0.0010), in GS4 from 0.55 to 0.63 (p=0.0274), and in GS5 from 0.57 to 0.75 (p<0.0001) when trained with the combination of original and synthetic image patches compared to just being trained with original image patches. Moreover, the comparative analysis revealed a notable enhancement in accuracy and the receiver operating characteristic curve (ROC) for the combined (original and synthetic) dataset relative to the original (p=0.0381). In case of needle biopsies the enhancement was consistent across both benign and malignant samples, with an overall accuracy increase from 91% when using the original training images to an accuracy of 95% when using original and synthetic images combined (p=0.0402). The original and synthetic combined training database yielded a sensitivity of 0.81, and a specificity of 0.92. CONCLUSIONS: This approach eases the burden of the need for extensive clinical data for training any AI model in the domain of digital imagery thus ultimately leading to cost and time-effective diagnosis and prognosis. Source of Funding: Sylvester Cancer Center © 2024 by American Urological Association Education and Research, Inc.FiguresReferencesRelatedDetails Volume 211Issue 5SMay 2024Page: e967 Advertisement Copyright & Permissions© 2024 by American Urological Association Education and Research, Inc.Metrics Author Information Derek Van Booven More articles by this author Cheng-Bang Chen More articles by this author Sheetal Malpani More articles by this author Oleksandr Kryvenko More articles by this author Himanshu Arora More articles by this author Expand All Advertisement PDF downloadLoading ...