Localized higher-order shimming is a common method for improving image quality in phase sensitive sequences used in cardiac imaging, but requires the manual placement of a three-dimensional bounding box around the heart in which the localized shimming is performed. We present an automated method for detecting such a bounding box from the localizer images using Deep Learning. Two-dimensional bounding boxes are first detected in each localizer slice and then combined to one three-dimensional bounding box. We compare two approaches, either training individual models for each localizer orientation or a joint model for all orientations.
This abstract and the presentation materials are available to members only; a login is required.