Covid-19 Automatic Detection from CT Images through Transfer Learning

During the last decade, several studies have been conducted to improve efficiency and robustness in the detection and segmentation of brain tumors based on different parameters like size, shape, location, and contrasts. This study proposes Multimodal Attention-gated Cascaded U-Net (MAC U-Net) model to address the performance issues observed in the detection and segmentation of low-grade tumors. The effectiveness of group normalization with attention gate is also explored with skip connections to segment small-scale brain tumors using several highlighted salient features. The model is evaluated on the brain tumor benchmark dataset BraTS2018 over various performance metrics such as Dice, IoU, Sensitivity, Specificity, and Accuracy. Experimental results illustrate that the proposed MAC U-net on BraTS 2018 dataset outperforms baseline U-nets with 94.47, 84.12, and 82.72 dice similarity coefficient values on HGG and 85.71, 78.85 and 74.16 on LGG subjects with Ground
Truth values of Complete Tumor, Tumor Core, and Enhancing tumor, respectively. The proposed model is also evaluated on BraTS 2019 and BraTS 2020 datasets. Moreover, MAC U-net achieves superior performance over typical conventional brain tumor segmentation methods especially in terms of low-grade gliomas.