23 Generalized 3D Medical Image … Computer Vision News Previous research has shown that semantic segmentations can be inferred across different medical data types. The team faced the challenge of using this prior work for another task. Finding a generalized solution for different problems was difficult for the registration part. However, by proxying the problem to another field in the segmentation part, they could formulate a generalized approach to image registration without the limitations of traditional methods. “Our motivation stems from our experience evaluating and running the Learn2Reg challenge that we organized at MICCAI,” Mattias reveals. “We saw that we have this great variety of tasks. We have abdominal image registration, lung, or thorax. We have different modalities: MRI, CT, ultrasound. But so far, all the participants had unique and different solutions for each of them. This requires a lot of training, retraining, designing, and redesigning different methods for each challenge task. We thought having one joint approach would be a good step forward.”
RkJQdWJsaXNoZXIy NTc3NzU=