Enhancing Histology Detection in MASH Cirrhosis for Artificial Intelligence Pathology Platform by Expert Pathologist Training

Abstract

This study addresses the need for precise histopathological assessment of liver biopsies in Metabolic dysfunction-Associated Steatohepatitis (MASH) cirrhosis, where assessing nuanced drug effects on fibrosis becomes pivotal. The study describes a framework for the development and validation of an Artificial Intelligence (AI) model, leveraging SHG/TPE microscopy along with insights from an expert hepatopathologist, to precisely annotate fibrous septa and nodules in liver biopsies in MASH cirrhosis. A total of 25 liver biopsies from the Belapectin trial (NCT04365868) were randomly selected for training, and an additional 10 for validation. Each biopsy underwent three sections: Smooth Muscle Actin (SMA) and Sirius Red (SR) staining for septa and nodule annotation by pathologists and an unstained section for SHG/TPE imaging and AI annotation using qSepta and qNodule algorithms. Re-training of qSepta and qNodule algorithms was executed based on pathologist annotations. Sensitivity and positive predictive value (PPV) were employed to evaluate concordance with pathologist annotations, both pre- and post-training and during validation. Post-re-training by pathologist annotations, the AI demonstrated improved sensitivity for qSepta annotations, achieving 91% post-training (versus 84% pre-training). Sensitivity for qSepta in the validation cohort also improved to 91%. Additionally, PPV significantly improved from 69% pre-training to 85% post-training and reached 94% during validation. For qNodule annotations, sensitivity increased from 82% post-training to 90% in the validation cohort, while the PPV was consistent at 95% across both training and validation cohorts.This study outlines a strategic framework for developing and validating an AI model tailored for precise histopathological assessment of MASH cirrhosis, using pathologist training and annotations. The outcomes emphasise the crucial role of disease-specific customisation of AI models, based on expert pathologist training, in improving accuracy and applicability in clinical trials, marking a step forward in understanding and addressing the histopathological evaluation of MASH cirrhosis.