Computer Vision News - January 2021

8 This network showed efficiency with a mean DSC score of 0.8958 for the radiomics- based segmentation task of axial T2W MR images. The prostate lesions in different stages (T1, T2, T3, T4) were also well segmented with a mean score of 0.9176 and it is shown in the Fig 4. The segmentation was more accurate compared to the current state-of-the-art and a probable reason explaining this may be the extraction of local and global features using the radiomics approach. The Promise12 challenge dataset was also used to apply the proposed pipeline and some of the segmented results are shown in Fig 5. This work can be useful for tasks that require prostate lesion segmentation and detection and their respective treatments. I would really like to thank the authors of this paper for giving permission to present this work in this month’s review section! Research Figure 5 (from left to right): the input T2W image, result using U-Net, proposed deeply supervised U-Net, zoomed and cropped sections with green is the plain U-Net and red the proposed pipeline.