2020-2021 CAIIMI Stimulation Project Grants
- Machine Learning Models for the Prognostication of Patients after Cardiac Arrest
PI: Dooman Arefan, PhD (firstname.lastname@example.org)
Cardiac arrest is the sudden and unexpected cessation of effective cardiac contraction and blood flow, resulting in rapid injury to the brain and vital organs. Among those resuscitated from cardiac arrest, brain injury is the major cause of mortality. Most post-arrest patients are initially comatose, of whom only 3 in 10 will awaken and recover. Accurately identifying patients who can recover and predicting time-to-awakening from a coma could help inform clinicians and avoid premature withdrawal of life-sustaining therapies. In this study, we leveraged artificial intelligence (AI) techniques to build interpretable machine learning models for prognostication using computed tomographic images of head and clinical data for post-arrest patients. Our study contributed to AI-based image analyses towards building clinically meaningful interpretation of model predictions, which is essential to gain physicians’ trusts and promote clinical adoption of AI.
- Super Resolution Ultrasound (SRU) Imaging using Deep Learning (DL) Approach: Further Breaking the Acoustic Diffraction Limit
PI: Kang Kim, PhD (email@example.com)
Ultrasound imaging (US) is one of the most favored imaging modalities in clinics in general. Recent development of super-resolution ultrasound (SRU) imaging provides unprecedented high spatial resolution that breaks the diffraction of conventional ultrasound imaging. It holds a great potential for noninvasively evaluating the microvascular changes that are associated with pathology in a broad spectrum of diseases. However, conventional center localization-based SRU imaging algorithm requires sparsely and uniformly distributed microbubble, leading to a long scan time and poor robustness for different organs. The deconvolution-based SRU developed previously by our group enables a short scan time but the spatial resolution is compromised. In this project, we developed a deep learning-based SRU imaging using U-net architecture that can reconstruct the microvascular images under the relatively high concentration and uneven microbubble distributions to maintain both the high spatial resolution and short scan time. The deep learning network was trained using simulated data and the deep learning-based SRU algorithm was evaluated on the data from in vivo human kidney scan. The SRU images were successfully reconstructed by the deep learning approach in which the spatial resolution is much improved compared to the deconvolution approach. Further improvement and validation is ongoing toward the clinical translation.
- Development of a Longitudinal ML model for the Clinical Management of Cerebral Aneurysm
PI: David A. Vorp, PhD (firstname.lastname@example.org)
- Utility of Machine Learning in Evaluating Musculoskeletal Imaging for Pathology
PI: Gene Kitamura, MD (email@example.com)
Please see the People page to check research interests and projects of individual members of the Center.