Text this: Enhancing Learning-Based Cross-Modality Prediction for Lossless Medical Imaging Compression