Artificial intelligence for streamlined immunofluorescence-based biomarker discovery in prostate cancer.

279 Background: Immunofluorescence (IF) performed on tissue microarrays (TMA) is used for biomarker discovery but is limited by the arduous and subjective human visual assessment with an IF microscope. We aim to implement deep learning-based artificial intelligence (AI) models to automate and speed up the analysis of numerous biomarkers and generate prediction models of recurrence and metastasis after surgery. Methods: A TMA was constructed consisting of 648 samples (424 tumors, 224 normal tissue) generated from prostatectomy specimens. IF staining was performed on the TMA using anti Ki-67, ERG antibodies and analyzed for differential expression using “gold standard” manual microscopy and using an AI algorithm. Analysis was done blinded to any clinicopathological data. For manual microscopy, relative mean fluorescence intensity of cancerous versus normal tissue was determined. The AI algorithm was generated using a training cohort of digitized images. To do so the Otsu method thresholding algorithm combined with mean shift clustering was employed to find cell centers, followed by a level-set algorithm, to compute cell boundaries.These predictions were then combined with pixel predictions of a fully convolutional deep model to refine the regions of overlapping epithelium, stroma, and artifact. The algorithm was then validated using a separate cohort. Results from the algorithm were then compared to the data from manual microscopy. Results: Ki-67 and ERG expression levels generated by the algorithm showed only a 5% variance compared to the manually generated results. The algorithm was able to pick out which tumor were positive for ERG with 100% accuracy in spite of variance from artifacts. The algorithm also had the ability to improve its accuracy after each iteration of modifications and feedback through the training cohort. Conclusions: The AI algorithm produced similar outcomes than manual quantification with high accuracy but with more efficiency, cost effectiveness and objectivity. We are now developing more complex algorithms that will include the differential pattern of expression of PTEN, MYC and others with the objectives of streamlining biomarker discovery.