Multi-modal Femoral Head Imaging Dataset

We have uploaded data from 10 patients, and other data will be uploaded gradually in the future.In this work, we construct a Multi-modal Femoral Head Imaging (MFHI) dataset. To the best of our knowledge, this is the first multi-modal medical image dataset for ONFH staging. Specifically, we collect complete CT, MRI, and X-ray data of 209 patients from the Chinese PLA General Hospital. Each patient underwent imaging examination with a multi-slice CT scanner to obtain CT data, a 3.0T superconducting magnetic resonance imaging system to obtain T1-weighted MRI data, and a digital X-ray imaging system to obtain X-ray data. To ensure the structure consistency between multi-modal data, we select to use the coronal-view of CT and MRI data, and the AP-view of X-ray data. For each CT or MRI data, considering that the frames at the start and end positions are irrelevant to the femoral head, we discard the first 20% and last 20% frames, and then select 20 frames (left: 10, right: 10) from the remaining frames as the data of interest. These data are combined with the corresponding X-ray data to form 20 groups of multi-modal data. In addition, as the femoral head only exists in a local region of CT, MRI, and X-ray data, we pre-train a YOLOv8 model to detect the femoral head region to generate the pre-processed multi-modal data, where each image has a resolution of 512\u00d7512. Finally, we invite two orthopedics experts to annotate the pre-processed multi-modal data into four stages: I, II, III, and IV. In summary, we obtain 4,180 groups of multi-modal data, where 3,140 groups from 157 patients and 1,040 groups from 52 patients are defined as the training and test sets, respectively.