High accurancy and effectiveness with deep neural networks and artificial intelligence in pathological diagnosis of prostate cancer: initial results

View Poster

INTRODUCTION

Most prostate cancers are adenocarcinomas and their pathologic diagnosis is highly associated with the prognosis. Big data based artificial intelligence (AI) now is going to become a hot topic in many areas of our daily life. AI and deep neural networks may play a major role in medical fields. In this study, we explored the application of AI in prostate pathology diagnosis, leading to a rapid, exact and high-efficiency technique.

METHODS

Histopathological whole mount (WM) sections of prostate after robot-assisted laparoscopic radical prostatectomy were used for machine learning. The pathology of prostate was evaluated according to 2014 International Society of Urological Pathology grading systems of prostate cancer. Since there were tumor grade heterogeneity within and between tumors, we identify the regions of different tumor grade levels at the WM sections (Fig.1). Grade 1 (Gleason Score 3+3) and 2 (Gleason Score 3+4) were defined as the low grade group, Grade 3 (Gleason Score 4+3) and 4 (Gleason Score =8) as the median grade group, Grade 5 (Gleason Score 9 or 10) as the high grade group. After deep machine learning with a certain amount of analyzed pathologic images, blank pathological sections were diagnosed with AI and compared with pathologist's results.

RESULTS

283 patients containing 918 pieces of WM sections were analyzed by AI for machine learning. 10 pieces of WM sections from 10 patients were tested to evaluate the AI's diagnosis effect. The accuracy rate is 99.38% to identify if the section contains cancer. Additionally, The pathologic diagnosis with the grade between AI and the pathologist were investigated. In our tested sections, AI's diagnosis was similar to the pathologist's (Fig.2).

CONCLUSION

According to our initial results, AI could replace pathologists to some extend in prostate cancer pathology diagnosis. AI will provided a more rapid, accurate and effective technique in medical area with the development of technology.

Funding: None