RP33 - From Retrospective Model Interaction to Prospective Monitoring

The paradigm shift from relying on retrospective data analysis to embracing prospective guidance in medical diagnostics marks a significant milestone in the evolution of healthcare technology. This project builds upon RP14, RP17 and RP25 and will provide real-time suggestions and automatic support, empowering clinicians with actionable insights based on current examination results [1], [2]. Healthcare providers can proactively respond to emerging trends and anomalies rather than reacting to historical data. For instance, when lab test results indicate increased tumour markers, a system should automatically suggest additional diagnostic procedures, such as imaging studies, to investigate the patient’s condition further. To facilitate the seamless execution of these recommendations, we plan to integrate agents that can fill out request forms, effectively merging workflow and diagnostics into a streamlined process. This harmonisation enhances efficiency by reducing manual data entry, minimising errors, and accelerating the diagnostic process. We will also consider the need to address the computational resource intensity and environmental impact of model execution. We will focus on developing innovative model distillation techniques to mitigate these concerns. We can significantly reduce their computational requirements and carbon footprint by automatically refining models to concentrate on the most pertinent and impactful data. Distilled models offer numerous benefits, including improved efficiency, reduced energy consumption, and enhanced deployability. These advantages make them an attractive solution for dynamic clinical environments where adaptability and responsiveness are crucial. Overall, this shift towards prospective guidance with real-time support, model distillation, and comprehensive integration fosters a more cohesive, efficient, and patient-centred approach.

[1] J. Keyl et al., “Decoding pan-cancer treatment outcomes using multimodal real-world data and explainable artificial intelligence,” Oct. 19, 2023, medRxiv. doi: 10.1101/2023.10.12.23296873.

[2] A. Brehmer et al., “Establishing Medical Intelligence - Leveraging FHIR to Improve Clinical Management: a retrospective cohort and clinical implementation study,” J. Med. Internet Res., Aug. 2024, doi: 10.2196/55148.