Large Scale Medical Image Dataset

Multiple dataset from different sources has been aggregated to create a large-scale medical image benchmark dataset in order to measure its performance. As each of the dataset’s images are of different sizes, the images are resized to 3 × 224 × 224 before the training process. This dataset contains total of 35 diseases of 4 different modality and is divided into train, validation, and test with a ratio of 7 : 1 : 2.

为构建用于模型性能评测的大规模医学影像基准数据集，研究人员聚合了多源数据集。由于各源数据集内的图像尺寸各不相同，在训练前需将所有图像统一调整为3×224×224的尺寸。该数据集涵盖4种不同成像模态下的35种疾病，并按照7:1:2的比例划分为训练集、验证集与测试集。