Machine Learning in Vascular Imaging– Forum, May 27, 2022

1000 563 MaLMIC - Machine Learning in Medical Imaging Consortium

Join the Machine Learning in Medical Imaging Consortium (MaLMIC) for an opportunity to network

Machine Learning in Vascular Imaging

Friday, May 27, 2022
3:30 to 5:30 p.m. ET

David Spence, a clinician, will talk about the need for machine learning applications in vascular imaging. His presentation will be followed by two researchers, Eran Ukwatta and Fumin Guo, talking about their work in vascular imaging machine learning. The talks will be followed by a discussion focused on lessons learned, opportunities to collaborate and sharing of resources.

Interested in joining? Please contact us.

Machine Learning in Vascular Imaging

Friday, May 27, 2022
3:30 to 5:30 p.m. ET

David Spence, a clinician, will talk about the need for machine learning applications in vascular imaging. His presentation will be followed by two researchers, Eran Ukwatta and Fumin Guo, talking about their work in vascular imaging machine learning. The talks will be followed by a discussion focused on lessons learned, opportunities to collaborate and sharing of resources.

Interested in joining? Please contact us.

Why measure carotid plaque burden?  How could automated methods based on AI improve the situation?

David Spence, MD
Professor of Neurology and Clinical Pharmacology, Western University

Talk summary: David will talk about use of carotid plaque burden for management of patients (risk stratification, evaluation of effects of therapy) and uses in research into the genetics and biology of atherosclerosis.

Machine learning for medical imaging: Towards translating algorithms into clinical care

Eran Ukwatta, PhD
Assistant Professor, University of Guelph

Talk summary: With the recent developments in medical imaging devices capable of acquiring high-resolution, multi-dimensional (i.e., 3D + time) images of the human body, automated image analysis methods are becoming increasingly essential for extracting previously inaccessible quantitative biomarkers from medical images. Parallel to this development, recent advancements in machine learning methods have availed a wealth of novel research opportunities in knowledge discovery and analysis of large medical databases. In this talk, Eran will provide a brief overview of research conducted in his lab. Eran will start with a description of conventional imaging analysis methods and then discuss the latest algorithms based on machine learning. He will then describe development of image analysis methodologies specifically for cardiovascular imaging.

Segmentation of carotid plaques in ultrasound images

Fumin Guo, PhD
Banting Postdoctoral Fellow, University of Toronto

Talk summary:Carotid ultrasound provides a number of important measurements for determining long term risk for stroke and monitoring carotid plaque progression, including total plaque area (TPA) and vessel-wall-volume (VWV). Deep learning can provide automatic carotid ultrasound segmentation. However, this technique typically requires heavy computational costs, large datasets and costly manual annotations for training, and yields sub-optimal generalizability and reproducibility. We developed a computationally efficient method combining a novel Voxel-FCN with simple network architecture and continuous max-flow with edge regularization for VWV segmentation. We also employed a UNet++ ensemble using relatively small datasets for algorithm training and evaluated the effect of using different training datasets. We further applied the UNet++ ensemble method for TPA measurements in 2D carotid ultrasound datasets acquired in London, Canada and Wuhan, China. In this presentation, Fumin will describe the algorithm, methods, results and potential clinical use.