Labelled in vitro, ex vivo, and in vivo IVUS imaging for "learned ultrasound segmentation and deformable CT fusion for augmented reality endovascular surgery", Dillon et al.

Labelled datasets for IVUS imaging for the purposes of training and evaluating segmentation algorithms. Recorded for in vitro (Volcano/Philips 10MHz IVUS PV Visions 0.035" transducer) and large animal aortic imaging (Volcano/Philips 10MHz IVUS PV Visions 0.018" transducer). Numpy datasets can be laoded using e.g., dataset_X = np.load("dataset_location/dataset_name.npy", allow_pickle=True). The following header data is provided for each element in the dataset: e.g., dataset_X[76] {'image': array([[[0, 0, 0], [0, 0, 0], [0, 0, 0], ..., [0, 0, 0], [0, 0, 0], [0, 0, 0]], [0, 0, 0], [0, 0, 0], [0, 0, 0]]], dtype=uint8), 'near_lumen_mask': array([[0, 0, 0, ..., 0, 0, 0], [0, 0, 0, ..., 0, 0, 0], [0, 0, 0, ..., 0, 0, 0], ..., [0, 0, 0, ..., 0, 0, 0], [0, 0, 0, ..., 0, 0, 0], [0, 0, 0, ..., 0, 0, 0]], dtype=uint8), 'far_lumen_mask': None, 'branch_mask': array([[0, 0, 0, ..., 0, 0, 0], [0, 0, 0, ..., 0, 0, 0], [0, 0, 0, ..., 0, 0, 0], ..., [0, 0, 0, ..., 0, 0, 0], [0, 0, 0, ..., 0, 0, 0], [0, 0, 0, ..., 0, 0, 0]], dtype=uint8), 'image_num': '686'} Where image is the original image, near_lumen_mask is the aortic lumen label, far_lumen_mask is the branch label, and image_num is the image number in the original dataset (not all images were labelled from the pullbacks to avoid highly repetitive segmentations). Note that "near_lumen_mask" and "far_lumen_mask" come from legacy naming conventions.