| C. Troussas, A. Krouska, Ph. Mylonas, C. Sgouropoulou, I. Voyiatzis |
| ClinLLM-XAI: Explainable Clinical Decision Support Using Large Language Models with Uncertainty-Aware Reasoning |
| 11th South-East Europe Design Automation, Computer Engineering, Computer Networks and Social Media Conference (SEEDA-CECNSM 2026), Thessaloniki, Greece, September 18-20, 2026 |
|
ABSTRACT
|
| AI-based Clinical Decision Support Systems (CDSS) raise high expectations regarding improving diagnostic accuracy and treatment planning. However, there are two major obstacles to the use of Deep Learning-based CDSS in practice: First, there is a lack of easily understandable and convincing explanations for physicians, and second, reliable uncertainty indicators to determine when a model's predictions should not be trusted. Therefore, this work proposes a Framework called ClinLLM-XAI. This system combines a large-scale language model (LLM) with Chain-of-Sort inference to generate well-explained clinical predictions with calibrated quantitative uncertainty ratings. It uses a multimodal encoder to process structured data from electronic health records (EHR) and unstructured medical notes, performs data retrieval and advanced inference using a clinical knowledge base, and calibrates confidence using Monte Carlo Dropout Ensemble procedures. We evaluated ClinLLM-XAI using 5 clinical prediction tasks with 38597 patient data points from hospital stays, as part of the MIMIC III benchmark. The results achieved AUROC values at the highest level (mortality forecast 0.892, readmission forecast 0.847, hospital length of stay forecast 0.874) compared to the base model, while generating explanations rated significantly more trustworthy by 24 clinicians compared to non-explainable baselines. Thanks to the uncertainty module, we were also able to specifically select predictions and achieve an accuracy of 0.94 in cases with high confidence. Unsuccessful cases were marked for manual review.
|
| 18 September, 2026 |
| C. Troussas, A. Krouska, Ph. Mylonas, C. Sgouropoulou, I. Voyiatzis, "ClinLLM-XAI: Explainable Clinical Decision Support Using Large Language Models with Uncertainty-Aware Reasoning", 11th South-East Europe Design Automation, Computer Engineering, Computer Networks and Social Media Conference (SEEDA-CECNSM 2026), Thessaloniki, Greece, September 18-20, 2026 |
[
BibTex] [
Print] [
 Back] |
