nitrain

Why we included this project

Training models on volumetric medical scans usually means writing your own loading, resampling, and augmentation code before you ever get to a network. Nitrain collapses that into a few lines: point an ImageReader at a NIfTI folder pattern, apply built-in transforms for normalization and resizing, assemble a loader with a SliceSampler, and pass it to a Trainer that already knows how to run segmentation. The same high-level API works with PyTorch, TensorFlow, or Keras, which matters if you might switch frameworks mid-project. It grew out of the older torchsample library, and that heritage shows in the defaults: modest configuration gets you a fitted model fast, which suits researchers and small teams more than assembling a bespoke pipeline. The tutorials page and an upcoming book make it a reasonable place to learn medical imaging AI as well.