Ensembles of Multiple Models and Architectures for Robust Brain Tumour Segmentat Brats17 NO.1 :Ensembles of Multiple Models and Architectures for Robust Brain Tumour Segmentation
论文Intramodality Domain Adaptation Using Self Ensembling and Adversarial Training 不同机构或者肿瘤程度导致数据集的domain shift问题: 论文Intramodality Domain Adaptation Using Self Ensembling and Adversarial Training,本论文提出方案来缓解这个问题.
newtest.py 在此之前,脑肿瘤专栏中的2D网络预测的时候,是把所有的切片预测完指标再求平均值,这样测的值极容易收到一些差的切片而影响整体的指标.所以以后的2D网络预测都采用下面方式进行计算指标,即把所有预测的切片拼接回3D,然后对3D数据整体进行计算指标.这样计算的值会偏高点.不只是2D网络这样,3D网络也是如此
Automatic Brain Tumor Segmentation using Convolutional Neural Networks with Test 论文:测试时数据增强(TTA):Automatic Brain Tumor Segmentation using Convolutional Neural Networks with Test
Brain tumor segmentation using deep learning 论文: Brain tumor segmentation using deep learning | Gal Peretz , Elad Amar
论文3D MRI brain tumor segmentation using autoencoder regularization Brats18 NO.1: 3D MRI brain tumor segmentation using autoencoder regularization
GetTrainingSets.ipynb 博主用的训练集和验证集均来自BraTs2018的训练集(其中HGG:210个病人,LGG:75个病人),在预处理中我主要有三个步骤:1、对各个模态进行标准化2、对各模态及其GT数据进行裁剪3、对各模态及其GT数据进行切片,并抛无病灶切片,最后合并各模态的切片,然后保存为Numpy