Abstract 2097: AI powered platform to identify primary prostate cancer patients with high risk of recurrence

Abstract

Abstract Background: To-date, there is no approved adjuvant therapy for primary prostate cancer (PCa). Long time of recurrence and a lack of a strong biomarker predicting recurrence after the first line of therapy is prohibitory to short clinical trials within a reasonable time. Artificial intelligence (AI) offers a unique opportunity for extracting critical morphological features from PCa tissues as prognostic marker that have evaded human eyes. We have built a AI-based platform to analyze H&E stained histological images of PCa. Our platform can accurately detect, grade and quantify PCa in patient tissue images. Here, we show how this platform can identify morphometric features that indicate biochemical recurrence within 3 years after radical prostatectomy (RP). Methods: Our dataset comprised of a cohort of 169 PCa patients who underwent RP. Whole slide H&E images (WSI) of their tissue sections were scanned at 40x magnification. corresponding pathologic parameters such as Gleason grade, tumor volume, TNM staging, margin status and post-treatment follow-up data were also obtained. Among the 169 patients, 89 patients recurred within 3 years of RP. Our platform technology involved two different deep CNN (Convolutional Neural Network) architectures. The platform first divided each WSI into multiple tiles. Each tile was analyzed using a CNN that graded the tile and generated a high dimensional vector to provide a mathematical equivalent of the morphology. The combination of high dimensional vectors across the WSI was then fed into a second CNN that generated a morphological score. The clinical parameters were combined with the morphologic score using a logistic classifier to predict outcome. Results: Thus far, we have obtained 84% AUC (Area under the Curve) for the ROC (Receiver operating characteristics) curve, using 10-fold cross-validation. In contrast, the most predictive clinical covariate was the T stage with 71% AUC. Conclusion: Our AI based platform can identify high risk PCa patients early for tumor progression post RP and may be selected for more cost effective and much shorter Clinical Trials necessary to develop adjuvant therapy. Citation Format: Wei Huang, Parag Jain, Raman Randhawa, Bijay Jaiswal, Samuel Hubbard, Hirak Basu, Rajat Roy. AI powered platform to identify primary prostate cancer patients with high risk of recurrence [abstract]. In: Proceedings of the Annual Meeting of the American Association for Cancer Research 2020; 2020 Apr 27-28 and Jun 22-24. Philadelphia (PA): AACR; Cancer Res 2020;80(16 Suppl):Abstract nr 2097.