AI-supported Primary Diagnosis of Gastric Biopsies: Multi-Site and Multi-Reader Study
Background: Gastric cancer has poor prognosis and, therefore, timely diagnosis and prevention of diagnostic mistakes is of high importance. Multiple pathologies are detected and reported upon review of gastric biopsies, such as carcinoma, high-grade dysplasia, lymphoma, Helicobacter pylori, neuroendocrine lesions and more. Artificial Intelligence (AI) based solutions for computer-assisted review of gastric biopsies may assist improving the accuracy and efficiency of diagnosis.
This study aimed to clinically validate the use of an artificial intelligence (AI)-based solution by pathologists for evaluating and reporting gastric biopsies.
Methods: A two-arm prospective reader study comparing the performance of pathologists supported by AI with pathologists reviewing digital slides was performed at multiple sites (with varied staining protocols and different scanners). Both arms were compared to ground truth (GT) established by a consensus of two gastric pathologists. Rates of major discrepancies between each arm and GT, as determined by an adjudicating pathologist, were compared.
Results: Six pathologists reported on 235 cases (426 H&E slides), each case being reported twice, once in each study arm. Pathologists first reviewed only H&E slides and IHC were provided upon request, while the AI diagnoses were rendered on H&E-stained slides only. The AI solution demonstrated high performance for the detection of gastric neoplasia (Carcinoma/HGD/ HG Lymphoma): AUC of 0.98 (95%CI: 0.967,0.994), sensitivity 96%, specificity 90%, NPV 100% and PPV 94%. High performance was demonstrated for H.Pylori detection: AUC of 0.93 (95%CI: 0.88,0.97) sensitivity 91%, specificity 80%. Pathologists’ feedback showed the AI solution is user friendly (92.5%), adds confidence to the reviewed cases (80%) and in a large proportion (83%) pathologists would favor continuing to work with the AI system.
Conclusions: This multi-site multi-reader study reported high accuracy for the detection of gastric neoplasia by the AI solution. The AI solution performed accurately and equally well with slides issued from different staining platforms and scanners. Thus, AI solutions have the potential to be significant support tools for pathologists in various clinical decision-making in routine pathology practice, enhancing the quality and reproducibility of diagnoses.
Objective:
- Increasing awareness of Artificial Intelligence (AI) based solutions for computer-assisted review of gastric biopsies, that may assist in improving the accuracy and efficiency of diagnosis.
Presented by:
Adam Booth, MD
Assistant Professor of Pathology
Northwestern University Feinberg School Of Medicine
Dr. Booth is an Assistant Professor of Pathology at Northwestern University Feinberg School of Medicine specializing in gastrointestinal, pancreatic, and hepatobiliary pathology. He completed his anatomic and clinical pathology residency at the University of Texas Medical Branch (UTMB), in Galveston, TX. Followed by a subspecialty fellowship in gastrointestinal, pancreatic, and hepatobiliary pathology at Beth Israel Deaconess Medical Center/Harvard Medical School in Boston, MA. He serves in leadership positions within the College of American Pathologists, Digital Pathology Association, Rodger C. Haggitt Gastrointestinal Pathology Society, Hans Popper Hepatopathology Society, and Intersociety Council on Pathology Information. His research interests include digital pathology, serrated polyps of the colon, and gastrointestinal post-transplant lymphoproliferative disorders.
