MorphoQuant-NASH : Fully Automated and Explainable Software to Support Pathologist’s Reading in NASH Clinical Trials

By 23 July 2021September 8th, 2021Paris Nash 2021

Cindy Serdjebi1, Karine Bertotti1, Damien Barbes1, Bastien Lepoivre1, Yvon Julé1

1. Biocellvia, 10 rue Grignan, Marseille

Objectives:

Most of the commercialized programs developed to support pathologist’s review in NAFLD/NASH clinical trials rely on artificial intelligence (AI), potentially translating human bias to the algorithms through the training process and black-boxing decision-making. Our objective was to develop a fully automated and explainable software suite to quantify key NAFLD/NASH endpoints in clinical trials.

Methods:

Specific histology was developed using a combination of different immunohistological labels in conjunction with standard stains to allow the identification and quantification of steatosis, inflammation, ballooning and fibrosis. In particular, a picrosirius-red stain combined with CK19 IHC dual stain was developed, and the Sonic Hedgehog signaling pathway was chosen as a reliable marker for ballooned hepatocytes. Algorithms using morphometry, topology, morphological and vector operations were used to develop MorphoQuant-NASH® and applied to quantify steatosis, inflammation, fibrosis and ballooning on liver biopsies from a prospective multicenter clinical trial.

Results:

MorphoQuant-NASH is a fully automated and 100% reproducible software. All readouts are quantitative. Steatosis- and inflammation-related parameters are quantified from H&E sections, fibrosis-related endpoints from PSR-CK19 labelled sections, and ballooning from Shh-labelled sections (Table 1). Beside accurate and objective quantification of the four major NAFLD/NASH endpoints, other readouts are provided to bring more consistency and insights to the information extracted from liver biopsies: tissue density, lipid macrovesicle size, number of immune cells, proportion of portal/periductular collagen, ductular reaction, Sonic hedgehog area are also

computed. Native and mapped images of the sections are also provided as supporting elements for the pathologists’ reading (Figure1).

Conclusion:

We have developed a non-AI program suite to assess key endpoints of NAFLD/NASH. MorphoQuant-

MorphoQuant-NASH: Fully Automated and Explainable Software to Support Pathologist's Reading