SCH: Robust and Equitable Clinical Decision Support in Glaucoma Detection and Progression Prediction

University Of Pennsylvania, Philadelphia PA

Investigators

Abstract

This project seeks to address the pressing issue of vision morbidity and vision impairment caused by glaucoma. Glaucoma, a complex and multifactorial disease that develops over extended periods, presents significant challenges in terms of early diagnosis and effective treatment within the existing healthcare infrastructure. The research aims to harness the potential of artificial intelligence (AI) to not only enhance our understanding of the disease but also to develop a clinical decision support tool that can significantly improve glaucoma diagnosis and treatment. The research seeks to innovate by tackling three specific challenges. The first challenge involves the development of multi-modal approaches for robust glaucoma detection assessment and vision field loss prediction. To accurately diagnose a patient, clinicians collect many types of data from several patient visits. This research recognizes the critical importance of leveraging multiple sources of data to improve the robustness and performance of glaucoma detection models. By integrating various data types, including various retinal imaging and vision field measurement data, across different points in time, the proposed clinical decision support tool is expected to provide a more comprehensive and accurate analysis of the disease, leading to better treatment outcomes. The second challenge is how to promote trust and improve adoption of AI-based decision support by clinicians and patients. The human-AI interface should provide information about the performance and limitations of the model predictions to enable users to make safe and effective decisions. Recent work has shown that uncertainty quantification can be effective in fostering trust in AI systems. However, these methods need to be extended to be integrated with our proposed multi- modal prediction pipeline to ensure robustness across multiple modalities with varying levels of reliability and correlation structures. The third challenge focuses on mitigating bias and improving fairness in glaucoma detection. By incorporating multi-modal information, the research aims to address these biases and improve fairness in glaucoma diagnosis.