Developing The Landscape For Artificial Intelligence In Liver Pathology: A Review And Analysis Of Interobserver Variation In Ishak And Knodell Scoring For Viral Hepatitis

Introduction Digitisation and increasing workload presents an opportunity for artificial intelligence (AI) tools in histopathology. In the UK, approximately 580,000 people live infected with hepatitis B or C and an estimated 30,000 liver biopsies are performed annually in the US. We performed a literature review and analysis to determine understanding of interobserver variation in viral hepatitis grading and staging as a foundation for developing a novel AI tool. Methods A literature search for papers examining viral hepatitis, interobserver variation and Ishak/Knodell scoring returned 24 papers. Abstracts were reviewed independently with inclusion and exclusion criteria by two consultants and a registrar, and consensus discussion determined the inclusion of eight papers in the final analysis. Average Cohen’s kappa coefficient scores of interobserver variation for necro-inflammatory activity (NIA) and fibrosis were gathered and these were used to give a range, mean and weighted mean kappa scores. Results Results are summarised in the table 1 below. There is poor interobserver variation amongst pathologists, particularly for grading (NIA) of viral hepatitis with mean kappa score 0.33 and weighted kappa score 0.30. Kappa scores for fibrosis showed moderate to substantial agreement (mean kappa score was 0.66 and weighted mean kappa score 0.63. Additional papers noted success with image analysis to improve observer agreement for fibrosis. Discussion To our knowledge, this is the first review examining interobserver variation of the Knodell and Ishak systems. The results demonstrate where AI can be used to improve agreement between pathologists and therefore provide more consistent pathological assessment for patients.