Alzheimer's disease

Alzheimer’s disease progresses over decades, but symptoms manifest only in the advanced stages, which makes early diagnosis and thereby, timely intervention difficult. The complexity and heterogeneity of the disease; sparse longitudinal data; and unknown relationships among factors such as disease pathology, brain structures, brain activity, and cognition present significant challenges in predicting disease progression.

We have developed models that provide a holistic view of disease progression by combining ML with multimodal data and clinical domain knowledge in the form of science-based differential equations (DEs). Basing the model on DEs provides biological interpretability to the results and using ML improves the model’s data fitting capacity.

The models predict personalized 5- to 10-year future cognitive decline in older adults, thus providing a tool to identify at-risk individuals. Our methods have also yielded novel insights into the physiological basis of compensatory processes that mitigate the effect of disease on cognition. Further investigation of these can guide the development of interventions.

Selected relevant publications

• K. V. Saboo, C. Hu, Y. Varatharajah, S. A. Przybelski, R. I. Reid, C. G. Schwarz, J. Graff-Radford, D. S. Knopman, M. M. Machulda, M. M. Mielke, R. C. Petersen, P. M. Arnold, G. A. Worrell, D. T. Jones, C. R. Jack Jr., R. K. Iyer*, P. Vemuri* (2022). Deep learning identifies brain structures that predict cognition and explain heterogeneity in cognitive aging. NeuroImage.

• K. V. Saboo, A. Choudhary, Y. Cao, G. A. Worrell, D. T. Jones, R. K. Iyer (2021). Reinforcement learning-based disease progression model for Alzheimer’s disease. Advances in Neural Information Processing Systems (NeurIPS).

• K. V. Saboo, C. Hu, Y. Varatharajah, P. Vemuri, R. K. Iyer (2020). Predicting longitudinal cognitive scores using baseline imaging and clinical variables. IEEE International Symposium on Biomedical Imaging (ISBI). (Oral presentation).