International Research Journal of Engineering and Technology (IRJET)
e-ISSN: 2395-0056
Volume: 12 Issue: 10 | Oct 2025
p-ISSN: 2395-0072
www.irjet.net
Multimodal deep‑learning system for Diagnosis of Autoimmune Hepatitis Using Clinical and Imaging Data Akshay Bankar1, Harshwardhani Kewate1, Sahil Tandekar1, Shejal Meshram1 1,2,3,4Dept. of CSE, Gurunanak Institute of Engineering and Technology, Nagpur, Maharashtra, India
---------------------------------------------------------------------***--------------------------------------------------------------------revised original and simplified AIH criteria provide Abstract - Autoimmune Hepatitis (AIH) is a chronic
structured diagnostic guidance, their practical use can be laborious, and histopathological interpretation remains inherently subjective, reliant on the pathologist’s expertise [4].
inflammatory disorder of the liver, and its diagnosis requires an intricate integration of clinical observations, serological findings, and histopathological evidence. The multifactorial nature of this process often poses significant challenges, demanding both time and specialized medical expertise. To mitigate these challenges, this study introduces a proof-of-concept multimodal web application developed to support hepatologists in diagnosing AIH. The proposed system seamlessly combines unstructured clinical text data with liver biopsy imagery and employs an advanced multimodal large language model (MLLM) for comprehensive analytical processing. Following data synthesis, the application produces two distinct outputs: a structured diagnostic support report for clinicians and an accessible educational handout for patients. The diagnostic report provides a probabilistic estimation of AIH, a modelderived confidence score, key clinical insights, differential diagnostic considerations, and preliminary management suggestions. In parallel, the patient handout translates complex medical information into comprehensible language to enhance health literacy and engagement. This research highlights the transformative potential of generative artificial intelligence in unifying heterogeneous clinical data sources, thereby improving diagnostic precision, optimizing clinical decision-making, and promoting patient-centered communication in the context of complex hepatic disorders.
The ongoing digitization of medical data and pathology slides has created new avenues for computational innovations that can complement clinical expertise. Recent progress in artificial intelligence, particularly the rise of large multimodal language models (MLLMs), has enabled systems capable of integrating and reasoning across diverse data modalities such as textual and visual inputs [5]. These models can correlate semantic content from clinical documentation with morphological patterns in medical images, effectively replicating certain aspects of clinical reasoning. In this context, the present study proposes a novel proofof-concept web-based system that utilizes an MLLM to assist in the diagnostic workflow of AIH. The platform accepts clinical information—either manually entered or extracted from uploaded records—alongside corresponding liver biopsy images. It performs a unified analysis of both data types to generate two outputs: a structured diagnostic support report for clinicians and a patient-oriented educational summary. Rather than substituting clinical judgment, the system aims to enhance it, offering a data-driven, multimodal synthesis of key diagnostic indicators to support more accurate, efficient, and informed decision-making in the evaluation of autoimmune liver disease.
Key Words: Autoimmune Hepatitis, Multimodal AI, Large Language Models, Diagnostic Support Systems, Medical Informatics, Histopathology
1.INTRODUCTION
2. Related Work
Autoimmune Hepatitis (AIH) is a chronic immunemediated liver disorder characterized by hepatocellular inflammation, interface hepatitis on histopathological examination, and the presence of circulating autoantibodies [1]. Diagnosing AIH poses substantial challenges due to its heterogeneous clinical manifestations and the frequent overlap of its features with other hepatic conditions, including viral hepatitis, drug-induced liver injury, and primary biliary cholangitis [2]. Accurate diagnosis requires a holistic evaluation that integrates clinical history, laboratory parameters—such as elevated aminotransferase and immunoglobulin G (IgG) levels— and, most critically, expert interpretation of liver biopsy specimens. Although established scoring systems like the
© 2025, IRJET
|
Impact Factor value: 8.315
The integration of artificial intelligence into hepatology and pathology has advanced rapidly in recent years, with early research primarily centered on developing machine learning algorithms capable of predicting liver fibrosis stages from clinical variables or classifying histopathological patterns from whole-slide images (WSIs) [6]. Convolutional Neural Networks (CNNs), in particular, have achieved near-expert accuracy in identifying histological features such as steatosis, inflammation, and hepatocellular ballooning within biopsy specimens [7]. Despite these successes, most existing systems remain unimodal, focusing exclusively on either structured clinical data or imaging data, thereby limiting
|
ISO 9001:2008 Certified Journal
|
Page 763