Deep Vein Thrombosis Segmentation via Deep Learning based U-Net using MRI Images

Deep vein thrombosis (DVT) is a prevalent medical condition. For pulmonary embolism to prevent potentially fatal acute consequences, an accurate identification of deep vein thrombosis is essential. In this paper, a novel VT-Seg has been proposed to segment the DVT using MRI images. Initially, the input MRI images are pre-processed using Multi-Scale Retinex (MSR) for enhancing the image. The enhanced images are classified using MobileNet to identify normal and deep vein thrombosis. The next phase uses U-Net to segment the deep vein thrombosis cases. Efficacy is determined by analysing F-measure, specificity, accuracy, Jaccard index, and Dice coefficient enhancement. Compared to the U-Net, traditional networks like Grabcut, Otsu method, and SegNet obtain less accuracy. The accuracy ranges of 99.25% are preserved with U-Net. According to the proposed VT-Seg approach, the overall efficiency of MSMT-DL, ResNet-101, and construct fusion models is improved by 3.15%, 26.35%, and 34.42%, respectively.