MedMLX
/

segresnet-brats-mlx

@@ -2,77 +2,123 @@
 license: apache-2.0
 library_name: mlx
 tags:
-- medical-imaging
 - apple-silicon
 - research
 language:
 - en
 ---
-# MONAI brats_mri_segmentation (SegResNet, 4-channel MRI)
-Native MLX package: [segresnet-mlx](https://github.com/MedMLX/segresnet-mlx.git).
-Evaluated revision: `40cbce54523632fbc8cdfc564a9cce48d8df3424`.
-This card describes the one configuration evaluated in the MedMLX paper. It is research software; the single-case comparison does not estimate clinical accuracy.
-## Source and checkpoint
-Original publisher: MONAI.
-Original implementation: [https://huggingface.co/MONAI/brats_mri_segmentation](https://huggingface.co/MONAI/brats_mri_segmentation).
-Source revision or version: `370f7f9d062745fbac445e7fe6d6616d35df04ec`.
-Original checkpoint: https://huggingface.co/MONAI/brats_mri_segmentation/resolve/370f7f9d062745fbac445e7fe6d6616d35df04ec/models/model.pt
-Conversion: segresnet_mlx.convert_checkpoint(spec=BUNDLES['brats_mri_segmentation'].spec).
-Conversion changes tensor layout or storage as specified by the native package; it does not retrain the model.
-## Recorded example
-MSD Task01 BrainTumour (training) / BRATS_001; MSD Task01 BrainTumour (training) / BRATS_001; MSD Task01 BrainTumour (training) / BRATS_001; MSD Task01 BrainTumour (training) / BRATS_001
 ```python
 from segresnet_mlx.medmlx import run
-# Replace file and directory names with your locally obtained inputs and weights.
-kwargs = {
-    "bundle": "brats_mri_segmentation",
-    "weights_path": "segresnet-brats.safetensors",
-    "output_dir": "out",
-    "t1c_path": "T1gd.nii.gz",
-    "t1_path": "T1w.nii.gz",
-    "t2_path": "T2w.nii.gz",
-    "flair_path": "FLAIR.nii.gz",
-    "overwrite": True
-}
-result = run(**kwargs)
 print(result["outputs"])
 ```
 Native result: **success**. Original-code comparison: **pass**, `minimum_per_label_dice` = `1.0`.
 Native whole-volume output compared with the original MONAI bundle; the October 12 renal operator repair also has a full-case BraTS regression.
-The complete [native run](native-run.json) and [original-code comparison](reference-comparison.json) retain commands, geometry, precision, and output-specific checks. Use the [replay harness](https://github.com/MedMLX/Med-MLX/blob/main/benchmarks/README.md) to substitute local paths. The package README documents installation and conversion.
-## Distribution terms
-Upstream code: **Apache-2.0**. Weights: **Apache-2.0**.
-Converted weights are hosted in this repository.
-Native whole-volume output compared with the original MONAI bundle; the October 12 renal operator repair also has a full-case BraTS regression. Converted weights may retain the identified upstream license and required notices; the accompanying notices describe the redistribution terms.
-Full [upstream license](LICENSE) accompanies any weight distribution.
-Required notices:
-- Retain the applicable license, copyright and attribution notices; identify the upstream checkpoint, authors, and MLX conversion changes.
-- Include Apache-2.0, retain applicable upstream NOTICE content, and mark modifications.
-Review sources:
-- https://huggingface.co/MONAI/brats_mri_segmentation
-- https://huggingface.co/MONAI/brats_mri_segmentation/resolve/370f7f9d062745fbac445e7fe6d6616d35df04ec/models/model.pt
-- https://huggingface.co/MONAI/brats_mri_segmentation/blob/370f7f9d062745fbac445e7fe6d6616d35df04ec/LICENSE
-## Packaged files
-- `segresnet-brats.safetensors`

 license: apache-2.0
 library_name: mlx
 tags:
+- mlx
+- native-mlx
 - apple-silicon
+- medical-imaging
+- mri
+- brain-tumor
+- segmentation
 - research
 language:
 - en
+inference: false
+pipeline_tag: image-segmentation
+base_model: MONAI/brats_mri_segmentation
 ---
+# SegResNet Brain Tumor Segmentation
+Native MLX port of [MONAI BraTS MRI segmentation](https://huggingface.co/MONAI/brats_mri_segmentation).
+Segment brain tumor regions from four MRI sequences locally on Apple Silicon.
+- T1, contrast-enhanced T1, T2 and FLAIR inputs
+- Native whole-volume segmentation with MLX
+- Converted MONAI SegResNet checkpoint included
+MLX implementation by [MedMLX](https://github.com/MedMLX).
+[GitHub](https://github.com/MedMLX/segresnet-mlx) · [Usage](https://github.com/MedMLX/segresnet-mlx#quick-start) · [Upstream](https://huggingface.co/MONAI/brats_mri_segmentation)
+**Requires the companion `segresnet-mlx` runtime.** Follow the quick start below.
+## Requirements
+Apple Silicon Mac with Metal GPU access, Python 3.12 and [uv](https://docs.astral.sh/uv/).
+The weights are included here; checkpoint conversion is not needed.
+## Quick start
+```bash
+uv venv --python 3.12
+source .venv/bin/activate
+uv pip install "segresnet-mlx @ git+https://github.com/MedMLX/segresnet-mlx.git" huggingface_hub
+hf download MedMLX/segresnet-brats-mlx --local-dir models
+```
+Authenticate with `hf auth login` if repository access is required.
+Run this example in the same environment, replacing the input paths with your own:
 ```python
 from segresnet_mlx.medmlx import run
+result = run(
+    bundle='brats_mri_segmentation',
+    weights_path='models/segresnet-brats.safetensors',
+    output_dir='outputs',
+    t1c_path='T1gd.nii.gz',
+    t1_path='T1w.nii.gz',
+    t2_path='T2w.nii.gz',
+    flair_path='FLAIR.nii.gz',
+)
 print(result["outputs"])
 ```
+## Input and output
+Input is four 3D NIfTI MRI volumes on the same voxel grid: T1,
+contrast-enhanced T1, T2 and FLAIR. The example names each modality explicitly.
+Output is `brats_mri_segmentation_mask.nii.gz` on the input grid. Exported
+labels use the bundle’s BraTS convention: 0 background, 1 tumor core,
+2 whole tumor outside the core, 4 enhancing tumor.
+## Technical details
 Native result: **success**. Original-code comparison: **pass**, `minimum_per_label_dice` = `1.0`.
 Native whole-volume output compared with the original MONAI bundle; the October 12 renal operator repair also has a full-case BraTS regression.
+These are recorded results for one configuration, not a new benchmark or an
+estimate of clinical accuracy. The [native run](https://huggingface.co/MedMLX/segresnet-brats-mlx/blob/main/native-run.json)
+and [original-code comparison](https://huggingface.co/MedMLX/segresnet-brats-mlx/blob/main/reference-comparison.json)
+retain the commands, precision, geometry and output checks.
+<details>
+<summary>Checkpoint provenance and recorded configuration</summary>
+Recorded configuration: **MONAI brats_mri_segmentation (SegResNet, 4-channel MRI)**.
+Evaluated package revision: `40cbce54523632fbc8cdfc564a9cce48d8df3424`.
+Recorded input: MSD Task01 BrainTumour (training) / BRATS_001.
+Original publisher: MONAI.
+Original implementation: [upstream source](https://huggingface.co/MONAI/brats_mri_segmentation).
+Source revision or version: `370f7f9d062745fbac445e7fe6d6616d35df04ec`.
+Original checkpoint: [pinned checkpoint](https://huggingface.co/MONAI/brats_mri_segmentation/resolve/370f7f9d062745fbac445e7fe6d6616d35df04ec/models/model.pt).
+Weights were converted with the companion MLX package, without retraining.
+Conversion changes tensor layout or storage as required by that implementation.
+The [replay harness](https://github.com/MedMLX/Med-MLX/blob/main/benchmarks/README.md)
+describes how to reproduce the recorded configuration with local paths.
+</details>
+## Intended use
+Research and education. Outputs require human review and are not intended for
+clinical diagnosis or treatment decisions.
+## License
+Weights: **Apache-2.0** ([license](https://huggingface.co/MedMLX/segresnet-brats-mlx/blob/main/LICENSE)).
+Upstream code: **Apache-2.0**. MLX port code:
+[Apache-2.0](https://github.com/MedMLX/segresnet-mlx/blob/main/LICENSE).
+Retain the [attribution and notices](https://huggingface.co/MedMLX/segresnet-brats-mlx/blob/main/NOTICE) when redistributing
+the bundle or a conversion:
+- Retain the applicable license, copyright and attribution notices; identify the upstream checkpoint, authors, and MLX conversion changes.
+- Include Apache-2.0, retain applicable upstream NOTICE content, and mark modifications.