Contextual information extraction in brain tumour segmentation

Automatic brain tumour segmentation in MRI scans aims to separate the tumour's endoscopic core, edema, non-enhancing peritumoral and enhancing core from three-dimensional MR voxels. Due wide range of intensity, shape, location, size, it is challenging segment these regions automatically. UNet prime CNN network performance source for medical imaging applications like segmentation. This research proposes a context aware 3D ARDUNet (Attentional Residual Dropout UNet) network, modified version take advantage ResNet soft attention. A novel residual dropout block (RDB) implemented analytical encoder path replace traditional convolutional blocks extract more contextual information. unique Attentional Block (ARDB) decoder utilizes skip connections attention gates retrieve local global The gate enabled Network focus on relevant part input image suppress irrelevant details. Finally, proposed assessed BRATS2018, BRATS2019, BRATS2020 some best-in-class approaches. achieved dice scores 0.90, 0.92, 0.93 whole tumour. On BRATS2020, 0.93, 0.94.