Meeting Banner
Abstract #2882

EfficientNet-B0 U-Net Model for Improved Specificity Cerebral Microbleed Detection on QSM MRI

Pin-Yu Lin1, Sabina Iqbal1, Yan Wen2, Rafeeque Bhadelia1, Katherine L Tucker3, and Salil Soman1
1Department of Radiology, Beth Israel Deaconess Medical Center, Harvard Medical School, Boston, MA, United States, 2General Electric, New York, NY, United States, 3Biomedical & Nutritional Sciences, Health Assessment Laboratory (HAL), Center for Population Health (CPH), University of Massachusetts Lowell, Lowell, MA, United States

Synopsis

Keywords: Segmentation, Neuro, Deep Learning, Cerebral Microbleeds, Neuroimaging

Motivation: Cerebral microbleeds (CMBs) are critical markers of neurovascular health; yet their small size and low contrast hinder detection in MRI. This study leverages QSM MRI and deep learning to improve CMB segmentation accuracy, aiming to reduce false positives.

Goal(s): Develop a high-sensitivity, efficient model for CMB segmentation in QSM MRI, improving accuracy and workflow integration over conventional techniques.

Approach: Using 2D U-Net with EfficientNet-B0, pseudo-3D stacking, and Focal and Dice Loss, we enhance CMB visibility and segmentation accuracy, validated on a custom dataset.

Results: The model achieved a Dice score of 0.7290, sensitivity of 0.8171, and specificity of 0.9506, confirming reliable CMB segmentation.

Impact: This model offers a sensitive, but highly specific automated CMB detection method, well suited to augment human readers for the multiple clinical instances requiring careful CMB biomarker assessment. Its robust performance enables large-scale CMB studies, advancing neurovascular diagnostics and research.

How to access this content:

For one year after publication, abstracts and videos are only open to registrants of this annual meeting. Registrants should use their existing login information. Non-registrant access can be purchased via the ISMRM E-Library.

After one year, current ISMRM & ISMRT members get free access to both the abstracts and videos. Non-members and non-registrants must purchase access via the ISMRM E-Library.

After two years, the meeting proceedings (abstracts) are opened to the public and require no login information. Videos remain behind password for access by members, registrants and E-Library customers.

Click here for more information on becoming a member.

Keywords