Multimodel biosensing in intelligent healthcare systems

Abstract

Recent advances in biomedical sensing technologies have enabled continuous monitoring of biological processes across multiple levels, ranging from cellular imaging to physiological and clinical health indicators. However, the heterogeneous nature of these data sources presents significant challenges for effective integration and interpretation in intelligent healthcare systems. Moreover, many artificial intelligence (AI) models used in biomedical analysis operate as opaque “black-box” systems, limiting their transparency and reliability in clinical applications. This study proposes a multimodal explainable artificial intelligence framework for real-time cellular biosensing in intelligent healthcare systems. The proposed approach integrates cellular microscopy imaging, wearable physiological biosensor signals, and clinical monitoring data to capture complementary biological information across multiple sensing layers. Cellular imaging data from the Broad Bioimage Benchmark Collection (BBBC021), wearable biosensor signals from the WESAD dataset, and clinical physiological measurements from the MIMIC-IV database were used to evaluate the framework. A multimodal fusion architecture was developed to combine modality-specific feature representations, while explain ability mechanisms such as feature attribution and saliency visualization were incorporated to enhance model transparency. Findings demonstrate that the proposed multimodal model achieves improved predictive performance compared with traditional machine learning and single-modality deep learning approaches. In addition, the explain ability module provides interpretable insights into the relationships between cellular morphology, physiological biosignals, and prediction outcomes. These findings highlight the potential of explainable multimodal AI frameworks to support trustworthy and transparent biomedical sensing systems for next-generation intelligent healthcare applications.