Background: Prostate cancer diagnosis heavily relies on histopathological evaluation, which is subject to variability. While immunohistochemical staining (IHC) assists in distinguishing benign from malignant tissue, it increases workload, costs, and leads to diagnostic delays. Artificial intelligence (AI) presents a promising solution to reduce reliance on IHC by accurately classifying atypical glands and borderline morphologies in hematoxylin and eosin (H&E) stained tissue sections.
Methods: In this study, we evaluated an AI model's ability to minimize IHC use without compromising diagnostic accuracy. We retrospectively analyzed prostate core needle biopsies from routine diagnostics at three different pathology sites. These cohorts consisted exclusively of diagnostically challenging cases where pathologists had required IHC to finalize the diagnosis.
Results: We show that the AI model achieves high performance, with area under the curve values ranging from 0.951 to 0.993 for detecting cancer in routine H&E-stained slides. When applying sensitivity-prioritized diagnostic thresholds, the model reduces the need for IHC staining by 44.4%, 42.0%, and 20.7% across the three cohorts, without a single false negative prediction. Among slides with a benign ground truth label, IHC use is reduced by up to 80.6%.
Conclusions: This AI model shows promise for reducing unnecessary IHC use in difficult prostate biopsy cases while maintaining diagnostic safety. Its integration into clinical workflows could streamline decision-making in prostate pathology and alleviate resource burdens.
Diagnosing prostate cancer typically involves examining tissue samples under a microscope. In challenging cases, doctors often use a special test called immunohistochemistry (IHC) to help confirm whether cancer is present. However, IHC adds time, cost, and extra work to the diagnostic process. In this study, we tested an artificial intelligence (AI) tool to see if it could accurately identify prostate cancer using only standard tissue images—without needing IHC. We analyzed especially difficult biopsy cases from three different hospitals, where pathologists had originally needed IHC to make a diagnosis. The AI tool was highly accurate and, when using a safety-first approach, it could reduce the use of IHC by 20% to 44% depending on the site without missing any cancers. Importantly, when focusing only on slides that were ultimately benign, the AI could reduce IHC use by up to 80.6%. This suggests AI could help pathologists make faster and more efficient decisions while maintaining diagnostic safety.
© 2025. The Author(s).