On-the-Fly Test-time Adaptation for Medical Image Segmentation
This addresses data-shift issues in medical imaging for clinical settings, offering a zero-shot, episodic adaptation method that avoids back-propagation during inference, though it is incremental as it builds on existing test-time adaptation techniques.
The paper tackles the problem of performance drop in medical image segmentation models when tested on data distributions different from training, proposing an on-the-fly adaptation method that achieves better performance compared to previous test-time adaptation methods, with validation on 2D and 3D data distribution shifts.
One major problem in deep learning-based solutions for medical imaging is the drop in performance when a model is tested on a data distribution different from the one that it is trained on. Adapting the source model to target data distribution at test-time is an efficient solution for the data-shift problem. Previous methods solve this by adapting the model to target distribution by using techniques like entropy minimization or regularization. In these methods, the models are still updated by back-propagation using an unsupervised loss on complete test data distribution. In real-world clinical settings, it makes more sense to adapt a model to a new test image on-the-fly and avoid model update during inference due to privacy concerns and lack of computing resource at deployment. To this end, we propose a new setting - On-the-Fly Adaptation which is zero-shot and episodic (i.e., the model is adapted to a single image at a time and also does not perform any back-propagation during test-time). To achieve this, we propose a new framework called Adaptive UNet where each convolutional block is equipped with an adaptive batch normalization layer to adapt the features with respect to a domain code. The domain code is generated using a pre-trained encoder trained on a large corpus of medical images. During test-time, the model takes in just the new test image and generates a domain code to adapt the features of source model according to the test data. We validate the performance on both 2D and 3D data distribution shifts where we get a better performance compared to previous test-time adaptation methods. Code is available at https://github.com/jeya-maria-jose/On-The-Fly-Adaptation