Original Data

During image training, the original image files and annotation files within each folder under the "data" folder are first converted into an "img" folder for the raw images and a "label" folder for the annotated images using script codes. Subsequently, the images undergo enhancement and augmentation processes via script codes to bolster the image data. Following this, the dataset is divided into a training set and a validation set according to specific requirements using script codes. The training process is then executed with the "train.py" file to obtain trained weights, as well as data on the changes in MIoU and loss values throughout the training process, along with corresponding images.

For 3D reconstruction, the video files under the LSCM, LSSCap, RSCM, and RSSCap folders are first converted into image files. Then, the trained weights are used to predict all frame images, yielding segmentation results. These segmented images undergo correction and registration processes using calibration functions to determine the size and relative position of the ultrasound images within the MRI template. Finally, 3D reconstruction and visualization are performed using script files. The choice of visualization tools includes the Mayavi library within the script files or ITK-SNAP.