Segmentation of lesions from magnetic resonance images of patients with multiple sclerosis is a challenging task, especially when involving multi-center or multi-scanner data. State-of-the-art lesion segmentation algorithms require training data to use identical acquisition protocols as the input data, but this is often difficult to control. In this work, we employ image synthesis to allow data from one scanner to resemble the data acquired in a different scanner. Overall lesion segmentation accuracy improves and the amount of false positives are reduced using synthesized images, indicating image synthesis can improve segmentation consistency in a heterogeneous dataset.
This abstract and the presentation materials are available to members only; a login is required.