Improvement of the Quality of Prostate Cancer Diagnosis Using AI In Digital Pathology Short Heading: Quality Improvement by Using AI In Pathology

Abstract

Automated AI-based tools are becoming increasingly important in modern medicine, including pathology, bysignificantly supporting pathologists’ diagnoses and reducing human biases. Histopathological evaluation of prostate biopsies plays a crucial role in diagnosing prostate cancer. Pathologists assess tumor type, grade (Gleason Grade), and tumor extension to determine the management plan. Diagnosis accuracy, particularly in tumor grading, can be affected by inter- and intraobserver variability among pathologists. Due to the increased incidence of prostate cancer and subsequent workload on pathologists, an AI-based tool like Ibex Prostate, can potentially reduce pathologists’ workflow and enhance diagnostic accuracy.3 This study aimed to retrospectively compare histologically diagnosed prostate cancer by pathologists to the AIbased algorithm, Ibex Prostate. The study evaluates the algorithm’simpact on laboratory workflowand diagnostic accuracy. Methods: The study was conducted at the Laboratory of Pathology East Netherlands (Lab PON, Hengelo, The Netherlands), using hematoxylin and eosin-stained (H&E) Whole Slide Images (WSI) from 2021. A total of 169 randomly selected and de-identified prostate biopsy cases, consisting of 809 slides and 701 parts, were used. Slides were digitized using a Philips Ultrafast Scanner (UFS). Of these, 674 parts from 168 cases were used for the study, while 33 slides were excluded: 16 slides lacked a definitive diagnosis from the original report, and 17 slides were out-of-focus. According to pathologists’ diagnoses, 391 parts (58%) were benign, and 283 (42%) contained carcinoma. Ibex Prostate, a validated and CE-marked AI tool developed using advanced machine learning techniques, particularly convolutional neural networks (CNNs), assessed slide-level scores for cancer probability, Gleason grading, and perineural invasion. The algorithm’s performance was evaluated using the area under the receiver operating characteristic curve (AUC).

Ähnliche Arbeiten