Vox2Vox: 3D-GAN for Brain Tumour Segmentation

Vox2Vox] Vox2Vox: 3D-GAN for Brain Tumour Segmentation · Issue #5 ·  e4exp/paper_manager · GitHub

Gliomas are the most common primary brain malignancies, with different degrees of aggressiveness, variable prognosis and various heterogeneous histological sub-regions, i.e., peritumoral edema, necrotic core, enhancing and non-enhancing tumor core. Although both these brain tumour types can easily be detected using multi-modal MRI, and exact them doing image segmentation is a challenging task. Hence, using the data provided by the BraTS Challenge 2020, we propose a 3D volume-to-volume Generative Adversarial Network for segmentation of brain tumours. The model, called Vox2Vox, generates realistic segmentation outputs from multi-channel 3D MR images, detecting the whole, core and enhancing tumor with median values of 93.39%, 92.50%, and 87.16% as dice scores and 2.44mm, 2.23mm, and 1.73mm for Hausdorff distance 95 percentile for the training dataset, and 91.75%, 88.13%, and 85.87% and 3.0mm, 3.74mm, and 2.23mm for the validation dataset, after ensembling 10 Vox2Vox models obtained with a 10-fold cross-validation.


神经胶质瘤是一种最常见的脑部恶性肿瘤,它的特点是多变的恶性程度,难以预测的预后,以及异构的子结构,例如,瘤边水肿,坏死核心,以及在发展的和非发展的肿瘤核心。虽然这些脑部肿瘤可以轻易的被多模态MRI发现,但是对它们进行精确分割却是一个挑战。因此,我们提出了一种3D体到体的生成对抗网络并且在BraTS Challenge 2020数据上进行了测试。模型的名称为Vox2Vox,它可以在不同条件下为3D MR图像生成逼真的分割结果,检测全体、核心或者正在发展的肿瘤核心。


邮箱地址不会被公开。 必填项已用*标注