Abdominal Multi-Organ Segmentation Via UnetR Model with Multi-Scale Feature Fusion

Accurate medical image segmentation significantly contributes to the development of computer-aided diagnosis and computer-aided intervention. In this paper, we propose an improved UnetR model for computed tomography (CT) image segmentation of multi-organ in abdominal regions. In order to take full advantage of the different scales of features in the UnetR model, a multi-scale feature map fusion module is integrated into the original model. This module can adaptively fuse the semantic features of different scales of each organ, and the fusion result is used as the final multi-organ segmentation result. We evaluated the proposed model on the publicly available BTCV abdominal multi-organ segmentation dataset. The experimental results show that the multi-scale feature fusion module can effectively improve the accuracy of the UnetR model for abdominal multi-organ segmentation.