Preserving Tumor Volumes for Unsupervised Medical Image Registration

A computer-implemented method for performing deformable registration of a moving image and a fixed image, the moving image being a medical image that images an organ having one or more tumor regions, each of the moving and fixed images being a three-dimensional (3D) image, the method comprising:

• performing a first computing process for generating a soft tumor mask (STM), the STM being a map of probability distribution for indicating a likelihood of an individual voxel of the organ in the moving image being located in the one or more tumor regions;
• and
• performing a second computing process for generating a warped image, the warped image being a 3D image obtained by warping the moving image to align with the fixed image, the second computing process being a machine-learning process trainable under an objective of minimizing a training loss computed as a weighted sum of a plurality of component losses, wherein the plurality of component losses includes:
• a volume-preserving loss for measuring a loss due to failure of keeping the one or more tumor regions unchanged in volume during warping the moving image, the volume-preserving loss being computed according to at least the STM;
• and
• a weighted similarity loss for measuring a loss of similarity over a non-tumor part of the organ between the warped image and the fixed image, the weighted similarity loss being computed according to at least the STM;
• such that the one or more tumor regions are preserved in volume during generating the warped image from the moving image while promoting similarity in the non-tumor part between the moving image and the fixed image during aligning the moving image with the fixed image.