BenchMD

BenchMD是一个包含19个公开可用数据集的基准，由哈佛大学等机构创建，旨在评估统一学习在医疗图像和传感器数据上的表现。数据集涵盖了从1D传感器数据到3D体积扫描的7种医疗模式，包括心电图、脑电图、胸部X光、乳腺X光、皮肤镜图像、眼底图像和低剂量计算机断层扫描。该数据集通过模拟真实世界的数据限制，评估了在不同数据量和分布偏移下的方法性能，特别关注自监督学习和预训练技术的应用。BenchMD旨在推动医疗领域中通用、可泛化和标签高效的方法的发展。

BenchMD is a benchmark comprising 19 publicly available datasets, developed by institutions including Harvard University, with the goal of evaluating the performance of unified learning on medical images and sensor data. This benchmark covers 7 medical modalities ranging from 1D sensor data to 3D volumetric scans, including electrocardiogram (ECG), electroencephalogram (EEG), chest X-ray, mammography, dermoscopy images, fundus images, and low-dose computed tomography (LDCT). It evaluates the performance of methods under varying data volumes and distribution shifts by simulating real-world data constraints, with a particular focus on the application of self-supervised learning and pre-training techniques. BenchMD aims to advance the development of general, generalizable, and label-efficient methods in the medical domain.