Learning deep abdominal CT registration through adaptive loss weighting and synthetic data generation
This work addresses the problem of accurate image registration in medical imaging for clinicians, but it is incremental as it builds on existing methods like VoxelMorph.
The study tackled improving deep learning-based deformable image registration for abdominal CT by exploring training strategies, loss functions, and transfer learning, resulting in performance gains such as improved registration accuracy through segmentation guidance and finetuning from brain MRI data.
Purpose: This study aims to explore training strategies to improve convolutional neural network-based image-to-image deformable registration for abdominal imaging. Methods: Different training strategies, loss functions, and transfer learning schemes were considered. Furthermore, an augmentation layer which generates artificial training image pairs on-the-fly was proposed, in addition to a loss layer that enables dynamic loss weighting. Results: Guiding registration using segmentations in the training step proved beneficial for deep-learning-based image registration. Finetuning the pretrained model from the brain MRI dataset to the abdominal CT dataset further improved performance on the latter application, removing the need for a large dataset to yield satisfactory performance. Dynamic loss weighting also marginally improved performance, all without impacting inference runtime. Conclusion: Using simple concepts, we improved the performance of a commonly used deep image registration architecture, VoxelMorph. In future work, our framework, DDMR, should be validated on different datasets to further assess its value.