Automatic whole mouse segmentation for cryo-imaging data using DRLSE model

Cryo-imaging is a novel and powerful imaging technique that enables 3D visualization of an entire mouse with single cell resolution. However, the current methods to segment a whole animal from the cryo-imaging data is not yet optimal. In this paper, we developed a fully-automatic software for segmenting a whole mouse in fluorescent cryo-images using Distance Regularized Level Set Evolution (DRLSE) model. In our experiment, we used masks that were manually created by experts as the gold standard to evaluate segmentation performance (sensitivity and specificity). We also tested the algorithm against a thresholding-based algorithm which was developed based on our previous work. The results suggest that DRLSE-based segmentation algorithm was more robust to noises and weak boundaries than the thresholding-based algorithm. The mean specificity of the DRLSE-based algorithm in the long exposure data (500 ms) and the short exposure data (250 ms) were 98.32% and 98.46%, respectively. The mean sensitivity in the long exposure data and the short exposure data were 97.08% and 93.93%, respectively. The drop in sensitivity was mostly due to the increased numbers of weak boundaries in the low contrast images. The 3D visualization results show similar results between the body masks generated by the DRLSE-based algorithm and the gold standard. This work is significant as it can increase through-put of cryo-imaging analysis and visualization workflow. Hopefully, it will have a significant impact on the advancement of biomedical image processing and analysis.