AI-Empowered Digital Workflow for Prostate Pathology in Clinical Routine: A Reader Study for Prostate Biopsies
Background: Prostate core needle biopsies (PCNBs) remain an important element of population-based screening and surveillance programs for prostate cancer. Successful implementations of computer-assisted diagnostic support solutions have started to demonstrate benefits for routine diagnosis, encouraging adoption.
We developed a novel multi-feature AI solution embedded within a digital workflow system to support pathologists review of PCNBs. This study evaluates the impact of the AI solution on review efficiency and utility in a large pathology laboratory.
Methods: A prospective two-arm reader study with retrospectively collected cases compared a digital read only arm with a digital + AI arm. Three pathologists reported on 4,366 H&E slides from 180 PCNB cases. Each case was reported twice, once in each study arm and randomized between pathologists. Case review times were compared between study arms. To assess the effect on the accuracy of reporting, discrepant reports were adjudicated and reviewed by a blinded uropathologist. The study endpoints were pathologists' reporting efficiency, the AI accuracy, and usability and productivity as reported from the reader pathologists’ feedback survey.
Results: The study demonstrated that reporting with the AI solution led to a 58% decrease in case review time (p-value < 0.01), regardless of the pathologist or the diagnosis, and a significant decrease in turnaround time (37%). Cancer detection by the AI solution was highly accurate when compared to pathologists' diagnosis, with a Negative Predictive Value of 99.9%, Positive Predictive Value of 100%, as well as a high PNI accuracy with an AUC of 0.95 (95%CI: 0.93,0.97). For Gleason grading, pathologists with AI showed high agreement with the adjudicator (ICC = 0.701 [95%CI: 0.64, 0.75]). Reader pathologists' satisfaction was very high, and all 3 pathologists were motivated to implement the solution in their daily practice.
Conclusions: We report the successful evaluation of a multi-feature AI solution embedded within a digital workflow system. This solution improved efficiency and assisted pathologists with accurate cancer detection and Gleason grading. Thus, the system contributed to diagnostic quality, utility, and productivity. The solution could be used as a significant ancillary tool for pathologists in clinical decision-making within routine pathology practice.
Objective:
- Increasing awareness of Artificial Intelligence (AI) empowered digital workflow solutions for review of prostate biopsies, that may assist in improving the accuracy and efficiency of diagnosis
Presented by:
Piotr Borkowski, MD
Managing and Medical Director
Quest Diagnostics/Ameripath
Dr. Piotr Borkowski attended and graduated from the Medical University in Gdansk, Poland. He has over 25 years of experience in diagnostic tissue pathology, research, molecular diagnostics, and computerized image analysis. He was a Resident in Anatomic and Clinical Pathology at Mount Sinai Medical Center, Florida. Dr. Borkowski worked in a hospital based private practice at St. Joseph Hospital and then Parkway Regional Medical Center, Florida, before joining Quest Diagnostics as a Pathologist specializing in Uropathology, Gastrointestinal pathology, Women's Health and Clinical Informatics. He currently serves as a Managing and Medical Director for Quest/Ameripath Tampa/Central Florida and the Director of the Center of Excellence for Digital and AI-Empowered Pathology of Quest Diagnostics. He also serves on the board of Florida Society of Pathologists and the CAP Digital and Computational Pathology Committee. Dr. Borkowski is board certified in Anatomic Pathology, Clinical Pathology and Clinical Informatics by the American Board of Pathology.
