1. Description and objectives

Recent breakthroughs in Large Language Models (LLMs) and Artificial Intelligence (AI) are profoundly transforming the healthcare sector. These technologies enable advanced clinical applications such as medical text summarization, clinical decision support, intelligent patient interaction, and personalized healthcare delivery. However, challenges remain in terms of explainability, regulatory compliance, integration with existing systems, and bias mitigation.

The HealthLLM workshop aims to provide an international forum for researchers, developers, and practitioners to present and discuss recent advances, current challenges, and future directions at the intersection of AI, natural language processing, knowledge engineering, and digital health.

This workshop addresses the following technical issues:

  • Effective integration of LLMs in clinical workflows
  • Medical language model development and fine-tuning
  • Hybrid AI approaches combining symbolic reasoning and deep learning
  • Explainability and trust in AI for healthcare
  • Data privacy, ethics, and regulatory frameworks (GDPR, HIPAA, AI Act)
  • Personalization through knowledge-aware systems
  • Knowledge graph prompting & reasoning
  • LLMs for knowledge graph construction
  • Knowledge-augmented LLMs
  • Retrieval-augmented generation with KGs

Emerging topics highlight:

The workshop will particularly encourage innovative submissions in cutting-edge research areas, including retrieval-augmented generation (RAG) applied to healthcare, low-resource fine-tuning for specialized medical tasks, and the development of privacy-preserving language models through federated learning approaches.

2. Relevance and complementarity to AICCSA 2025

The proposed HealthLLM workshop is closely aligned with several key tracks of AICCSA 2025:

  • Track 3 – Data Science, Knowledge Engineering, and Ontologies: by exploring how structured knowledge and reasoning mechanisms can enhance the interpretability and applicability of LLMs in clinical settings.
  • Track 4 – Artificial Intelligence and Cognitive Systems: through the design and deployment of intelligent systems that can reason, learn, and support human decision-making in healthcare contexts.
  • Track 5 – Natural Language Processing: with a focus on biomedical and clinical NLP tasks, such as summarization, named entity recognition, dialogue systems, and question answering for medical data.

Healthcare is a data-rich, safety-critical domain where accuracy, transparency, and contextual understanding are paramount. As such, it serves as an ideal and high-stakes application field for testing and advancing the capabilities of AI systems. The complexity and sensitivity of medical data also highlight the importance of hybrid approaches, combining deep learning with knowledge-based reasoning and formal ontologies; a theme deeply rooted in the AICCSA tradition.

Moreover, this workshop complements the main conference by:

  • Promoting interdisciplinary collaboration between AI researchers, clinicians, data scientists, and ethicists.
  • Encouraging contributions that go beyond proof-of-concept to tackle real-world challenges in clinical workflows and decision-making.
  • Addressing pressing concerns around privacy, explainability, and fairness in healthcare AI, in light of evolving international regulations such as the EU AI Act and GDPR.
  • Creating a space to critically examine the limits of current generative AI tools in health, and propose innovative, responsible, and impactful solutions.

By bridging foundational AI research with applied health technology, the HealthLLM workshop supports the mission of AICCSA 2025 to be a forum for forward-thinking and societally relevant computing research.

3. Keynote Talk

If the HealthLLM workshop reaches a satisfactory number of accepted submissions and registered participants, we plan to host two keynote talks (Jashore University of Science and Technology, Bangladesh), entitled: ““.

Keynote 1: Dr. Md. Kamrul Islam

Title: Empowering Drug Discovery with LLMs and Knowledge Graphs: Towards Explainable and Scalable Biomedical Inference

Abstract

Recent advances in biomedical AI have highlighted the synergy between large language models (LLMs) and knowledge graphs (KGs) for accelerating drug discovery and improving healthcare outcomes. This keynote presents an integrated framework that combines ensemble KG embeddings with LLM-driven knowledge extraction and reasoning to support explainable drug repurposing in healthcare and bioinformatics applications. By constructing a biomedical KG and learning contextualized embeddings of its entities and relations, the framework enables a deep learning model to identify and prioritize potential therapeutic candidates. LLMs enhance the system by extracting biomedical relations from unstructured literature and facilitating interpretable predictions through natural language rationales. Moreover, the approach incorporates molecular docking simulations for biological validation and leverages KG-derived rules and paths for explainability. This talk will explore how the fusion of symbolic and neural paradigms KGs and LLMs can advance trustworthy and scalable biomedical AI, with broader implications for personalized medicine and knowledge-driven healthcare systems.

About the speaker

Dr. Md. Kamrul Islam is the Head of the Department of Computer Science and Engineering at Jashore University of Science and Technology (JUST), Bangladesh, where he also serves as an Associate Professor. He earned his Ph.D. in Computer Science from the University of Lorraine, France, in 2023. He holds a Master’s degree in Computing from Universiti Malaysia Pahang (2019) and a Bachelor’s degree in Computer Science and Engineering from Khulna University of Engineering and Technology (2009). Dr. Islam’s research interests span big data processing, data mining, graph representation learning, knowledge graphs, drug repurposing, and social network analysis. His recent work focuses on the integration of knowledge graph embeddings and deep learning for explainable drug discovery. He has led and contributed to several national research projects funded by the University Grants Commission (UGC) and the Ministry of Education (MoE) in Bangladesh. As an active researcher and academic leader, Dr. Islam is committed to advancing the intersection of symbolic and neural AI methods and fostering international collaboration in biomedical and data-driven research.

Keynote 2: Dr Bishnu Sarker

Title: LLM-based Retrieval-Augmented Generation (Rag) For ICD-10 Assignment

Abstract

Accurate assignment of ICD-10 codes is essential for clinical research, billing, and public health surveillance. However, manual coding is time-consuming and prone to errors, while many automated tools fall short in handling the nuanced and complex language of clinical documentation. In this talk, I will present a novel approach that leverages a Retrieval-Augmented Generation (RAG) framework to improve the precision and scalability of ICD-10 code prediction. Our method integrates dense vector retrieval with the generative power of large language models (LLMs), enabling more accurate semantic matching between clinical notes and ICD-10 code descriptions.

We preprocess and embed both clinical text and ICD-10 definitions using LLMs to capture deep contextual representations. These embeddings are indexed to support efficient retrieval of relevant codes. The retrieved information is then used as input context for a generative model that predicts the most appropriate ICD-10 codes. Trained and evaluated on a curated dataset of clinical narratives and associated codes, our RAG-based approach outperforms traditional classification models, particularly in handling ambiguous or complex clinical scenarios.

The results demonstrate the framework’s potential to enhance coding accuracy while reducing manual burden. Looking ahead, we aim to integrate patient demographic data, explore more advanced language models, and incorporate explainability techniques to increase system transparency and clinician trust. This work highlights how modern AI can be harnessed to address real-world challenges in healthcare documentation and informatics.

About the speaker

Bishnu Sarker is an Assistant Professor of Computer Science and Data Science at Meharry Medical College, Nashville, TN, USA. His research focus is on applying AI, deep learning, natural language processing (NLP), and graph-based reasoning approaches to effectively describe biomedical entities numerically and to infer their functional characteristics from complex, heterogeneous, and interconnected biomedical data. Dr. Sarker is currently principal investigator to multiple projects funded by NSF and other Federal agencies accumulating more than $2M. He received his BS in Computer Science from Khulna University of Engineering and Technology, Bangladesh; MS in Data Mining and Knowledge Mining from Sorbonne University, France; and PhD in Computer Science from INRIA, France. He was visiting researcher at MILA – Quebec AI Institute and University of Montreal, Canada.

Important Notice:

The organization of the keynote talk is conditional upon achieving a sufficient number of accepted papers and registered participants for the workshop. A final decision will be made after the submission and registration phases are completed.

4. Organizing committee

5. Tentative technical program committee

  • Tuna Hacaloglu, ETS Montréal, Canada
  • Nejat Arinik, CRIL-CNRS, University of Artois, France
  • Akrem Sallami University of Lille, France
  • Anis Tissaoui, University of Pau and Pays de l’Adour, France
  • Abderrazek Jemai , INSA of Tunis, Tunisia
  • Marouen Kachroudi, University of Tunis El Manar, Tunisia
  • Sami Zghal University of Jendouba, Tunisia
  • Chaker Katar, University of Jendouba, Tunisia
  • Imen Ben Sassi, LIRMM, Université de Montpellier, CNRS, France
  • Sarra Hasni, University of Tunis, Tunisia
  • DONATIEN KOULLA MOULLA, University of Maroua, Cameroun, Afrique du Sud
  • Montassar Ben Massoud, University of Tunis, Tunisia
  • Sabeur Aridhi, University of Lorraine, France

6. Submission and review process

  • Submissions will be handled through the AICCSA EasyChair platform, under the category Workshops and Special Tracks.
  • Papers may be in the form of:
    • Extended Abstracts (2 pages)
    • Full Papers (5 pages)
  • All papers must follow the IEEE conference format.
  • A double-blind review process will be conducted, with at least two independent reviewers per paper.
  • Accepted contributions will be included in the AICCSA 2025 proceedings and submitted for publication in IEEE Xplore.