An international multi-institutional validation study of the algorithm for prostate cancer detection and Gleason grading

Abstract

Pathologic examination of prostate biopsies is time consuming due to the large number of slides per case. In this retrospective study, we validate a deep learning-based classifier for prostate cancer (PCA) detection and Gleason grading (AI tool) in biopsy samples. Five external cohorts of patients with multifocal prostate biopsy were analyzed from high-volume pathology institutes. A total of 5922 H&E sections representing 7473 biopsy cores from 423 patient cases (digitized using three scanners) were assessed concerning tumor detection. Two tumor-bearing datasets (core n = 227 and 159) were graded by an international group of pathologists including expert urologic pathologists (n = 11) to validate the Gleason grading classifier. The sensitivity, specificity, and NPV for the detection of tumor-bearing biopsies was in a range of 0.971–1.000, 0.875–0.976, and 0.988–1.000, respectively, across the different test cohorts. In several biopsy slides tumor tissue was correctly detected by the AI tool that was initially missed by pathologists. Most false positive misclassifications represented lesions suspicious for carcinoma or cancer mimickers. The quadratically weighted kappa levels for Gleason grading agreement for single pathologists was 0.62–0.80 (0.77 for AI tool) and 0.64–0.76 (0.72 for AI tool) for the two grading datasets, respectively. In cases where consensus for grading was reached among pathologists, kappa levels for AI tool were 0.903 and 0.855. The PCA detection classifier showed high accuracy for PCA detection in biopsy cases during external validation, independent of the institute and scanner used. High levels of agreement for Gleason grading were indistinguishable between experienced genitourinary pathologists and the AI tool.

Harnessing artificial intelligence for prostate cancer management

“… 21 detection of prostate cancer and perineural invasion and automated Gleason grading CNN, supervised learning biopsy training: 549 slides external ‡ : 100 consecutive cases, 1,627 slides detection: AUC 0.991 GS 6 or ASAP vs. GS 7–10: AUC 0.941 GP3–4 or ASAP vs. GP5: AUC 0.971 perineural invasion: AUC 0.957 Tolkach et al. 22 detection and automated GG grading CNN, supervised learning RP training: TCGA-PRAD dataset external 1 22 : 2 cohorts (592 and 279 patients) external 2 36 : 7473 cores, 423 patients from five centers external 1: detection: AUC 0.9919 and 0.9918 GG grading † : κ 0.51–0.66 external 2: detection: sensitivity 0.971–1.000; specificity 0.875–0.976 grading: κ quad 0.72–0.77 Ström et al. 37 detection, measurement of tumor length and automated grading DNN, supervised learning biopsy training: 1,069 patients, 6,953 slides external: Imagebase and Karolinska dataset (330 cores from 73 cases) detection: AUC 0.986 GG grading: for Imagebase dataset, mean pairwise κ 0.62.…”

“…In Tolkach’s study, pathologists recognized false-positive alerts from AI as useful warnings, as these highlighted areas required additional attention and immunostaining for further evaluation. 36 However, there is a risk that pathologists may overly rely on AI predictions without critically evaluating its predictions. Meyer’s small-scale study showed that pathologists were willing to trust AI regardless of its accuracy.…”

Breaking Barriers: AI’s Influence on Pathology and Oncology in Resource-Scarce Medical Systems

“…HALO Prostate AI, developed by Indica Labs is a DL-based screening tool currently CE-marked under IVDD, and its validation study was published in August 2023. The model identified biopsies that contained tumors with sensitivity, specificity, and negative predictive values ranging from 0.971 to 1.000, 0.875 to 0.976, and 0.988 to 1.000, respectively, across multiple test cohorts [59]. Histotype Px, by DoMore Diagnostics (CE-marked under IVDD), uses deep learning to provide risk stratification for patients with CRC using WSI.…”

Detection of Muscularis propria Invasion in Urothelial Carcinoma Using Artificial Intelligence

“…Artificial intelligence (AI) has the potential to be an accurate and time efficient technology that can aid pathologists in diagnosing muscularis propria invasion. AI has already shown promising results in several areas of pathology, such as Gleason grading for prostate adenocarcinoma, 7 – 11 melanoma scoring, 12 and more. In previous studies we have developed and applied a new algorithmic approach called hierarchical contextual analysis (HCA) for the detection of perineural invasion in pancreatic adenocarcinoma, 13 and also for the detection of ganglion cells, a process of significant importance for improving the accuracy of Hirschsprung's disease diagnosis.…”