A Dialogue-Grounded Clinical Evidence Panel

This project investigates how longitudinal in-home sensing data can be transformed into clinically relevant evidence and presented to healthcare professionals during outpatient consultations. A compact sensing system will be deployed in participants’ homes to collect long-term multimodal data, potentially including depth data, millimeter-wave radar signals, inertial measurement unit data, audio recordings or audio-derived features, and activity annotations, subject to ethical approval and informed consent. The system will process the collected data to characterize daily activities and longitudinal behavioral patterns, such as mobility, sleep-related behavior, falls, physical activity, and other activities of daily living. During a clinical consultation, the evolving doctor–patient dialogue will be used to identify the current information need. The system will then select, aggregate, and present relevant historical evidence, including recent observations, individual baselines, long-term trends, and notable deviations. The proposed system is intended to operate as a passive and non-diagnostic clinical side panel. It will not replace clinical judgment or independently generate diagnoses. The research will focus on context-to-evidence mapping, multimodal and temporal alignment, evidence selection and summarization, presentation of uncertainty or conflicting evidence, and privacy-preserving management of sensitive longitudinal data. The system will be evaluated in terms of evidence relevance, information coverage, response latency, clinician usability, and its potential to support clinical history-taking and reduce information omissions.