Deep learning helps discriminate between autoimmune hepatitis and primary biliary cholangitis

Area of Science:

• Hepatology and digital pathology.
• Application of artificial intelligence in medical diagnostics.

Background:

• Autoimmune hepatitis (AIH) and primary biliary cholangitis (PBC) frequently present with overlapping biliary and interface hepatitis features.
• Misdiagnosis can lead to incorrect treatment and adverse patient outcomes.
• Accurate differential diagnosis is critical for effective patient management.

Purpose of the Study:

• To investigate the efficacy of a deep learning (DL) pipeline for the differential diagnosis of AIH and PBC.
• To develop and validate a DL model for analyzing digitized liver biopsy slides.
• To reduce diagnostic errors and improve patient care through AI-assisted pathology.

Main Methods:

• A multicenter study utilized a library of digitized liver biopsy slides from six European centers.
• A novel DL model, the autoimmune liver neural estimator (ALNE), was trained on whole-slide images (WSIs) without human annotations.
• The ALNE model was validated on an external dataset and compared against clinico-pathological diagnoses and interobserver variability.

Main Results:

• The ALNE model achieved an area under the receiver operating characteristic curve of 0.81 in external validation for differentiating AIH from PBC.
• Attention heatmaps indicated ALNE focused on inflammatory areas, associating them with AIH.
• The model's performance highlighted discrepancies with general pathologists' assessments, showing significant interobserver variability (Fleiss's kappa = 0.09).

Conclusions:

• The ALNE model represents a pioneering system for quantitative and accurate differential diagnosis between AIH and PBC.
• This DL approach demonstrates significant potential in distinguishing these liver conditions using digitized histology.
• The findings support the use of AI in pathology to enhance diagnostic accuracy and consistency.