PV26 Speakers

Subject to change.

 

 

image

Athittiyakorn Phonhong

M.Sc. Student, Khon Kaen University

Athittiyakorn Phonhong is a Master's student in Pathology with a background in Medical Educational Technology. Her research focuses on the intersection of digital pathology and artificial intelligence, with particular interests in computational pathology and liver pathology. She has contributed to research projects applying machine learning to diagnostic pathology and is interested in exploring how emerging technologies can enhance pathology practice and education.

 

 

SESSIONS

AI-Powered Histopathology imaging Diagnosis of Autoimmune Hepatitis
   Sun, Oct 18
   1:10 PM - 1:30 PM PT
  Seaport G

Introduction/Background: Autoimmune hepatitis (AIH) is a chronic immune-mediated liver disease with histopathological features that overlap with other liver disease, making accurate diagnosis challenging. Liver biopsy remains the diagnostic gold standard. However, interobserver variability and the subjective nature of histological interpretation remain key limitations. Artificial intelligence (AI), particularly deep learning using convolutional neural networks (CNNs), offers a promising approach to improving diagnostic precision in digital pathology.

Methods: This retrospective study utilizes archival FFPE liver biopsy specimens from patients diagnosed with AIH, viral hepatitis, and metabolic dysfunction-associated steatotic liver disease (MASLD)/MASH at Srinagarind Hospital, Khon Kaen University (2018-2025). A minimum of 230 cases (>=115 AIH; >=115 non-AIH) are included. Whole-slide images (WSIs) are digitized. Two EfficientNet-B0-based CNN models are developed: a classification model to differentiate AIH from non-AIH and normal liver, and a fibrosis prediction model to stage liver fibrosis (F0-F4) from H&E stained WSIs using Masson's trichrome-guided pathologist assessment as the reference standard.

Results: Preliminary evaluation of the classification model demonstrated promising performance, with an accuracy of 92%, sensitivity of 90%, and specificity of 94% in differentiating AIH from non-AIH and normal liver on H&E-stained WSIs. Evaluation of the fibrosis prediction model is ongoing.

Conclusion/Discussion: This study aims to demonstrate that a deep learning framework can reliably differentiate AIH from histologically similar liver diseases and predict fibrosis stage directly from H&E-stained WSIs. If validated, this approach may serve as a reproducible diagnostic support tool to reduce interobserver variability and enhance clinical decision-making in hepatopathology.

Learning Objectives:

  1. Discuss potential applications and future directions of this approach in real-world settings
  2. Assess the clinical applicability of AI in improving early detection, prognosis in liver diseases
  3. Differentiate between conventional diagnostic approaches and AI-assisted decision-making in pathology
Chat bot