Deep Learning for Evaluation of Ultrasound Images in Developmental Hip Dysplasia

The dataset here can be used only for scientific research purposes with the written consent of the publisher (contact: [e-mail]).

The dataset was obtained with the approval of the ethics committee from TOBB Hospital. Ultrasound (US) imaging was performed using the EPIQ 5G device, manufactured by Philips Medical Systems, at TOBB ETU Hospital. The acquired US images were in DICOM format with a .dcm extension, representing 2D image slices. These slices were aggregated to form a comprehensive image pool.

Since the images were in their raw format, a filtering and evaluation process was conducted. Some images were excluded due to being too degraded or complex to be utilized in training. To export the raw images, the RadiAnt DICOM Viewer software was used, converting them into JPEG format. All metadata was removed from the images to comply with personal data protection regulations, ensuring full anonymization.

Subsequently, the images were aligned to center the relevant structures of interest. Expert physicians then reviewed and approved the cropping of the images to a resolution of 640x640 pixels. The annotations and masks were generated using the CVAT tool, producing outputs in both Segmentation Mask 1.1 and COCO formats. For specific applications, the COCO format was further converted to YOLO format.

These steps ensured the dataset's readiness for further research and analysis. The methodology, including data acquisition, preprocessing, and annotation, was designed to facilitate reproducibility in similar studies.