Impact of alternative antiandrogen therapy for Japanese prostate cancer patients in the era of new hormonal therapy

Pilot study of an artificial intelligence-based deep learning algorithm to predict time to castration-resistant prostate cancer for metastatic hormone-naïve prostate cancer

The object in this study is to develop an artificial intelligence-based deep learning algorithm for prediction of time to castration-resistant prostate cancer by combined androgen blockade therapy in metastatic hormone-naïve prostate cancer. We included 180 metastatic hormone-naïve prostate cancer patients who initially received combined androgen blockade. We first evaluated whether time to castration-resistant prostate cancer was a significant prognostic factor. Then, using the patients' needle-biopsy specimen images, we developed and validated our deep learning algorithm. The results are shown below. First, we confirmed that time to castration-resistant prostate cancer correlated with overall survival (P < 0.001). Next, we selected two groups by time to castration-resistant prostate cancer of >24 months (n = 18) and <6 months (n = 6) and developed a deep learning algorithm by artificial intelligence-based machine deep learning. In 16 other metastatic hormone-naïve prostate cancer patients used as an external validation set, we confirmed the prediction accuracy remained significant (P < 0.05). In conclusion, our obtained deep learning algorithm has high predictive ability for the effectiveness of combined androgen blockade.