Developing interpretable predictive… (F99EB037629) – Project Grant to The Regents Of The University Of…

Abstract: Neurological and psychiatric disorders pose significant challenges for effective treatment due to their complex and multifaceted nature. Accurate, patient-specific predictive models are crucial for advancing precision medicine in these fields, yet current models often struggle with incomplete or messy clinical data and rely on 'black-box' machine learning techniques that lack interpretability. This limits their utility in clinical settings where transparency and adaptability are essential for patient- centered care. This project proposes the development of advanced predictive models that integrate features from multiple imaging techniques-including T1-weighted MRI, functional MRI, diffusion MRI, and quantitative susceptibility mapping-alongside clinical, demographic, cognitive, and other health data. By combining these diverse data types, the models aim to capture higher-order interactions and the complexity of neurological health. Importantly, they will be designed to handle missing and incomplete data, providing reliable, patient-specific predictions even when some information is unavailable. In the pre-doctoral phase, my research will focus on constructing hypergraph neural network architectures that generate interpretable predictions with associated confidence scores. This work will be conducted in a large cohort of over 230 parkinson's disease patients, predicting their response to deep brain stimulation (DBS). These models will enhance clinical decision-making by offering insights into treatment outcomes tailored to individual patients. In the post-doctoral phase, the research will focus on generalizing these models across diverse neurological conditions and clinical settings to ensure their reliability. The models will be refined and transformed into practical tools for clinical use. This phase will involve deeper collaboration with clinicians to better understand their needs, ensuring that the software developed is accessible, reliable, and capable of addressing equity in neurological disease care. The ultimate goal is to develop open-source software that supports precision medicine and improves patient outcomes across a range of disorders. By addressing critical gaps in handling complex, multi-modal data and providing patient-specific, interpretable tools, this project will foster the broader adoption of individualized treatment approaches in neurology and psychiatry, ultimately improving care across a wide range of conditions.

Developing interpretable predictive models using patient specific imaging and clinical biomarkers with missing data

Award facts

Description

Quick answers

Who received award F99EB037629?

How much is award F99EB037629 worth?

See this award's full record