Glioma segmentation of optimized 3D U-net and prediction of multi-modal survival time

It is difficult to segment Glioma and its internal structure because the Glioma boundaries have edemas and complex internal structures. This paper proposes a new optimized, integrated 3D U-Net network to achieve accurate segmentation of Glioma and internal subareas. The contribution of this paper is twofold, it studies the clinical path of patients with Glioma and constructs an optimized 3D U-Net deep learning algorithm by combining them with the radiologic feature set. The proposed model was validated in the published Glioma operation data set of multi-modal MRI resonance images and clinicians manual segmentation data. The model can accurately segment the MRI multi-modality images of Glioma and intra-tumour nodes and achieve the multi-modality prediction of the overall survival period of patients. The experimental results further indicated that the segmentation accuracy of the proposed method was higher than other sophisticated methods. The Dice similarity coefficients of the whole tumor (WT) region, the core tumor (CT) region, and the augmentation / enhanced tumor (ET) region, were 0.9632, 0.8763, and 0.8421, respectively, which are better than the clinical experts’ manual segmentation results. Hence, this research can effectively promote the development of deep learning clinical precise diagnosis and medical technology for Glioma.