Outcome Prediction

Therapeutic strategy might be variable depending on the reliable outcome modeling about toxicity and reseponse following the radiation therapy. Learning-based approach has been enhancing the predictive performance of the model for the treatment outcome, which involved the three-dimensional medical images and dose distribution as well as the patient-specific clinical information. Our group is interested in developing robust learning-based predictive model for the appropriate clinical applications.

Image Translation

There exist a variety of image generative networks available, which could be applied to the medical image processing, ranging from image reconstruction to image translation. MR-only radiotherapy, for instance, has been one of the most interesting topics in radiation therapy with no use of CT images for treatment planning, which was enabled by the generative networks translating MR to CT images. Our group is interested in developing such a network algorithm that can generate clinically feasible translated images, and achieving the artifcat correction possibly throughout the translation

Auto-Segmentation

Contouring normal organs and target volume is important, but labor-intensive. Deep learning-based autosegmentation recently helped boost efficiency in the preparation for radiation therapy. We have been interested in the normal organ autosegmentation, which are expanded to the target volume segmentation throughout the state-of-art network algorihtms such as vision transformer.