Automatic multi-organ abdominal CT segmentation can improve the efficiency of clinical workflows such as disease diagnosis, prognostic analysis, and treatment planning. Artificial intelligence-based medical image segmentation techniques have significantly improved segmentation accuracy recently. However, segmentation models rely on a large amount of clinical CT data with manual annotation, and the annotation of data is expensive and time-consuming. In addition, the size and shape of organs vary from patient to patient, and the boundaries of adjacent organs are blurred. Therefore, the automatic and accurate multi-organ segmentation of abdominal CT remains a challenge. 

A new study published in the Medical Image Analysis on Jul. 9 shows that the proposed hybrid deformable model (HDM) and multi-scale feature fusion network (MFSS) can improve the performance of abdominal CT segmentation. The proposed method achieves the average Dice Similarity Coefficient (DSC) 0.852, which outperformed the other state-of-the-art results. The study was conducted by Dr. XIE Yaoqin and Dr. ZHOU Shoujun's group from the Shenzhen Institute of Advanced Technology (SIAT) of the Chinese Academy of Sciences. 

To alleviate the burden of annotating many training images, the researchers developed a new data augmentation method called HDM for training image augmentation. The HDM is essential to learn the appearance variance to the deep network. 

Dr. LIANG Xiaokun from SIAT, the first author of this study, explained that the proposed HDM achieves higher accuracy of the results, indicating that it may potentially become a generalized technique for deep learning-based segmentation, even for image registration and reconstruction. 

"In addition, to extract image features more effectively and accelerate the convergence of the algorithm, we designed an MSFF framework. The MSFF enables to extracts multi-scale features from the pre-trained network, which can reduce the training difficulty of the network and improving the segmentation accuracy," said the Ph.D. candidate Li Na from SIAT, the other co-first author.