Improving prostate cancer detection and grading using novel digital technologies

This thesis addresses the challenges in prostate cancer (PCa) diagnosis, focusing on Gleason grading in prostate needle biopsies, which plays a crucial role in risk stratification and subsequent treatment strategy. The main objectives were to evaluate the variation in Gleason grading in Dutch pathology laboratories in daily clinical practice, its impact on treatment strategies, and to explore interventions that might reduce interobserver variation. Additionally, the thesis discusses the potential of digital pathology and artificial intelligence (AI) in improving histological assessments of PCa and describes the first prospective clinical trials to assess the effect of AI-assisted pathology on daily clinical practice. The thesis highlights the systemic variation in Gleason grading across Dutch pathology laboratories. Analysis of pathology reports from over 35,000 PCa patients (2017-2019) revealed significant discrepancies between laboratories across all different ISUP Grades. This inconsistency in grading directly affects clinical decisions, with patients from higher-grading laboratories more likely to receive active treatments than those from lower-grading labs. The study suggests that the variability in grading can lead to differences in treatment, depending on which laboratory assesses the biopsy, and points to the potential risks in clinical studies that lack central pathology review. Subsequently, the effect of an ISUP Prostate Test B e-learning module on interobserver variability was assessed among 42 pathologists. Significant improvement in grading consistency was observed after e-learning (linearly weighted kappa 0.70 before versus 0.74 after, p=0.01). Among those with initially poor agreement, the most progress was observed. However, 4/10 slides still received all possible ISUP Grades, both before and after e-learning. We also examined the impact of laboratory-specific feedback reports on Gleason grading variation. Despite expectations, no significant improvement was seen in grading consistency after the feedback, likely due to a lack of awareness among pathologists about the feedback itself. This suggests that while feedback could be useful, it must be better integrated into laboratory workflows. This underlines the complexity of achieving uniform grading, even with educational interventions. The potential of AI in improving diagnostic accuracy and efficiency is then further explored. AI could enhance the reproducibility of diagnoses, reduce costs, and save time, though challenges remain, including algorithm imperfections, regulatory hurdles, and the need for digital infrastructure. AI's ability to assist in PCa grading, PCa detection, and workflow optimization is promising, but further prospective trials are needed to fully understand its benefits. Therefore, the CONFIDENT trials were designed to investigate how AI assistance can affect pathology workflow, as one of the first prospective implementations of AI in daily pathology practice, focusing on decreasing the need for costly immunohistochemistry and saving time. These trials showed that AI-assisted workflows could reduce the use of IHC without compromising diagnostic accuracy. However, the time-saving benefits of AI assistance in PCa detection were not as pronounced as expected, underscoring the need for better integration into existing pathology systems. Overall, the thesis emphasizes the need for more consistent PCa grading and explores the potential for AI and digital pathology to improve PCa diagnostics.