Bridging the Gap Between AI and Reality • Rhodes, Greece
Time: Wednesday, 5.11
Room: Room C
Authors: Thamizhinyan Natarajan, Tiziana Margaria
Abstract: Artificial Intelligence (AI) has achieved remarkable accuracy in healthcare diagnostics, yet adoption in clinical practice remains limited. High-performing deep learning models are often opaque, difficult to deploy, and inaccessible to non-technical users. This gap between accuracy and usability poses a barrier to responsible AI adoption in sensitive domains such as healthcare. This paper presents a Low-Code/No-Code (LCNC) platform that enables usable AI by making advanced diagnostic models accessible to clinicians without programming expertise. The platform provides drag-and-drop workflows, reusable building blocks, and integrated explainable AI (XAI) components to bridge the divide between technical innovation and real-world practice. To demonstrate this approach, we integrate an AI-driven Parkinson’s Disease (PD) screening model into the LCNC environment. PD is a progressive neurological disorder where early detection is vital for patient care. Our prior work optimized Convolutional Neural Networks (CNNs), with the Xception model enhanced by a multi-head attention layer achieving 98.25\% accuracy and 100\% sensitivity on spiral and wave handwriting datasets. While these results are promising, clinicians face significant barriers in using such models directly. Within the LCNC platform, clinicians can upload handwriting samples through a simple interface. Pre-processing steps such as resizing and normalization are automated through workflow blocks, while the trained CNN model is invoked via a model node. The results are presented alongside XAI visualizations such as SHAP heatmaps, highlighting regions of the drawing most influential in the prediction. A human-in-the-loop checkpoint allows medical professionals to validate outputs before generating diagnostic reports, which can be exported into electronic health records. This case study demonstrates how LCNC platforms can transform accurate but complex AI models into responsible, transparent, and usable healthcare tools. By embedding explainability and human oversight into visual workflows, the approach advances the democratization of AI in healthcare and offers a scalable pathway for wider clinical adoption.