Abstract
Muscle tissue at the level of the third lumbar vertebrae (L3) slice in the abdomen Computed Tomography (CT) images is often used to evaluate body composition in clinical process, which commonly requires manual annotation by physicians using software toolkits. Image segmentation methods, especially deep learning models, have achieved much progress and been applied to medical image analysis in recent years. In this work, a fully convolutional networks (FCN) is applied to muscle segmentation at the L3 slice in the abdomen CT images. A total of 216 abdomen CT scans were collected and annotated by physicians. The total muscle region is divided into 4 sub-regions based on its anatomical structure. Two FCN models are trained and tested on the same dataset, one with data augmentation, and the other without. Experimental results show that both the FCN models achieve accurate muscle segmentation on the total muscle region and the sub-regions (Jaccard index > 0.97). Moreover, the FCN model trained with data augmentation achieves better performance than the one without except on the third sub-region, which indicates that applying data augmentation may not improve the segmentation performance for all the muscle regions.
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have