Your Donations in Action: Sean Nurmsoo, MD

Using Deep Machine Learning to Automate Hematoma Volume Segmentation 

Sean Nurmsoo, MD, resident in the Department of Radiology and Diagnostic Imaging at University of Alberta in Canada

Intracerebral hemorrhage (ICH), constituting 15% of all strokes, is associated with high mortality and morbidity, and disability in survivors at six months. Accurate and quick measurement of hematoma volumes is crucial due to its association with hematoma expansion and poor outcomes. Volume information aids prognosis for families and supports the study of treatments, including medical hemostatic therapies and minimally invasive surgical evacuation.

The most widely used method of ICH volume estimation is the ABC/2 method, involving manual measurement of perpendicular hematoma diameters. Although comparatively quicker than manual segmentation, ABC/2 estimates are often inaccurate, resulting in overestimation and poor reliability.

For his 2019 RSNA Research Medical Student Grant project, “Development and Validation of a Deep Machine Learning Tool for Automated Hematoma Volume Segmentation Using 3D Convolutional Neural Networks,” Sean Nurmsoo, MD, a resident in the Department of Radiology and Diagnostic Imaging at University of Alberta, in Edmonton, Alberta, Canada, investigated the feasibility of training a 3D convolutional neural network (CNN), to recognize and take accurate volume measurements of ICHs.

Dr. Nurmsoo and colleagues used data from patients in the PREDICT study with ICH to train DeepMedic, an open-source 3D CNN. The trained model was validated using data from patients in the SPOTLIGHT/STOP-IT trials. The researchers evaluated the agreement between ICH volume assessed by DeepMedic and ABC/2, comparing it to semi-automated manual segmentation.

DeepMedic was more accurate when compared to ABC/2 and manual measurements, performing better specifically for larger, more irregular and varied types of bleeding in the brain.

“We hope this tool will serve as the new standard for both clinical and research hematoma measurement, and greatly facilitate ICH research worldwide,” Dr. Nurmsoo said.

Funding from the RSNA R&E Foundation grant helped Dr. Nurmsoo obtain essential resources and created an opportunity for collaboration. 

“My outstanding mentor, Dr. Thien Huynh, provided invaluable guidance through the research process, and encouraged me to accept the R&E Foundation’s invitation to present at RSNA,” he said. “This experience has led to lasting connections with other researchers. It also resulted in my participation in the Introduction to Academic Radiology (ITAR) program at RSNA.”

For More Information

Learn more about R&E Funding opportunities.

Read our previous Your Donations in Action article.