Scientific Talks

Computational Pathology for Predicting Prostate Cancer Recurrence After Surgery*


Peter Gann, MD
Professor Department of Pathology, University of Illinois at Chicago College of Medicine

Background: Prostate cancers (PCa) are distinguished by wide variability in aggressiveness. While the Gleason grading system has provided a durable foundation for determining prognosis for 40 years, improved outcome prediction would benefit patients by identifying those in need of adjuvant therapy or more intensive monitoring. We are developing an approach using machine-learning analysis of routine H&E stains to discriminate recurrent from non-recurrent patients after prostatectomy, based on architectural, color, and nuclear patterns.

Methods: We scanned TMAs containing quadruplicate cores from 176 recurrent cases of PCa and 154 controls, matched on age, Gleason, pTNM stage and race. Using Definiens Developer XD, we processed images at two spatial levels: a larger scale to recognize tissue compartments (epithelium, stroma, lumen and inflammation), and a smaller scale to recognize normal and atypical nuclei in each compartment. Multi-resolution segmentation identified homogeneous superpixels, which were then classified using machine learning into epithelium, stroma or ambiguous areas. We used relative darkness in a spatial neighborhood, size, and shape criteria to segment typical and atypical nuclei.

Relational features among objects within and between the tissue and nuclear levels were created. The resultant feature library contained 884 base features and 3,551 total variables. We used iterative L1 penalized logistic regression and 5-fold cross-validation to reduce variables and obtain AUC estimates.

Results: PSA, Gleason and CAPRA-S score (based on multiple clinical predictors) achieved AUCs <0.60 in this matched sample set. An L1 model containing 18 histometric features had cross-validation AUC=0.81 (95%CI: 0.76-0.85). A 74 feature model had AUC=0.95 (95%CI: 0.93-0.97). Pre-surgical PSA, Gleason grade or CAPRA-S did not change model performance. Notably, leading features were frequently derived from stromal or ambiguous regions and involved nuclear shape or texture. Their independence from tumor grading reflected subject matching and lack of explicit accounting for nuclear or stromal characteristics in the Gleason criteria.

Conclusion: A computational pathology approach, applied to discrimination of recurrent PCa in H&E stains, appears capable of adding to the predictive power of Gleason and CAPRA scoring. We are refining our models, conducting additional validation studies and ultimately plan to extend this work to biopsy samples.