This repository contains a .zip file with a PyTorch-based deep learning model (.pth file) together with the training .json file, the model can be used for automated segmentation of the left and right kidneys from post-contrast Dixon MRI images. This Dixon MRI dataset was acquired with the following imaging parameters: a field of view (FoV) of 400 mm in both the read and phase directions, and a slice thickness of 1.5 mm. The repetition time (TR) was 4.01 ms, with two echo times (TE): TE1 at 1.34 ms and TE2 at 2.57 ms. A total of 144 slices per slab were obtained. The model requires 4D axial data with 4 input channels: out-phase, in-pase, water and fat. The model was trained using data from 143 patients with diabetic kidney disease (DKD), based on high-quality semi-automated segmentations. API The model can be applied to data using the function kidney_pc_dixon (using the keyword model='nnunet') from the miblab python package [documentation]. Model Details: Format: PyTorch .pth checkpoint Architecture: nnU-Net Input: 4D array [contrast, x, y, z] from post-contrast Dixon MRI (in-phase images, out-phase images, water map and fat map) Output: Binary segmentation masks for left and right kidneys Post-processing: Optionally retains only the largest connected component for each kidney to reduce false positives. Training Data: Dataset: 143 patients with diabetes Annotations: Semi-automated expert segmentations Modality: Post-contrast Dixon MRI Version History: Version Date Description v1.0 May 2, 2025 Initial release. four-channel nnunet (.pth model).