Computer Vision News - January 2021

Research 4 This paper examines a U-Net technique for prostate lesion detection. This task is especially challenging for axial T2 weighted MR images (T2W) for both prostate gland and prostate lesion segmentatio n. The main reason U-Net was introduced is for the segmentation task which makes it very applicable for this paper. You can read the introduction to U-Net in our previous articles but a quick reminder here. U-Net is created from an encoder network followed by a decoder network. Unlike classification where Prostate lesion segmentation in MR images using radiomics based deeply supervised U-Net Happy New Year for everyone and let’s hope that this year will bring new scientific and social solutions to the pandemic challenges we are facing! The reviewed article this month is on the field of urology and radiomics and it’s titled Prostate lesion segmentation in MR images using radiomics based deeply supervised U-Net by Praful Hambarde et al. by Ioannis Valasakis (@wizofe) Notably, in this paper, after the clinical evaluation and testing the proposed solution against public datasets the prostate cancer diagnosis is improved by using a U-Net radiomics- based approach. A 2D U-Net is introduced for prostate gland and lesions segmentation and an image-to-image approach is used for training/validation/ segmentation. the end result of the deep network is the only important thing, semantic segmentation not only requires discrimination at pixel level but also a mechanism to project the discriminative features learnt at different stages of the encoder onto the pixel space. In the encoder part convolution blocks are followed by a maxpool downsampling to encode the input image into feature representations at multiple different levels. The decoder attempts to semantically project the discriminative features (lower resolution) learnt by the encoder onto the pixel space (higher resolution) to get a dense classification. The decoder consists of upsampling and concatenation followed by regular convolution operations. To improve the network output approaches such as pixel- to-pixel and patch-to-patch have been used. The image-to-image approach has a lot of advantages: it