vishalgokani/hipvascai-classification

HIPVASCAI MONAI 3D Classification Models

This repository hosts the released HIPVASCAI MONAI 3D classification model package for LCPD MRI research workflows.

Model Details

• Model family: MONAI 3D image classification
• Architectures: resnet18, resnet50, densenet121
• Release artifact: hipvascai_monai_classification_model.zip
• Input format: two-channel nnU-Net-style NIfTI files in imagesTs
• Optional masks: matching masks in labelsTs may be used for ROI cropping

Input Channels

• _0000.nii.gz: precontrast MRI
• _0001.nii.gz: subtraction MRI

Classes

• class_0
• class_1
• class_2
• class_3

Preprocessing

When a matching mask is available, inference uses a 10 percent expanded 3D mask bounding box before resizing to [128, 128, 32]. When no mask is available, inference uses the full image volume according to the pipeline defaults.

Intended Use

These models are intended for research use in automated HIPVASCAI MRI classification workflows. They are not intended for standalone clinical diagnosis or treatment decisions.

How to Run Inference

Clone or download the HIPVASCAI code repository, create the classification inference environment, then run:

python 3dclassification_inference\run_inference.py `
  --data-dir X:\path\to\my_inference_cases `
  --scratch-dir C:\scratch\hipvascai_classification_inference `
  --model-repo vishalgokani/hipvascai-classification `
  --models resnet18 resnet50 densenet121 `
  --batch-size 1

For local inference from a downloaded zip:

python 3dclassification_inference\run_inference.py `
  --data-dir X:\path\to\my_inference_cases `
  --scratch-dir C:\scratch\hipvascai_classification_inference `
  --model-zip C:\path\to\hipvascai_monai_classification_model.zip `
  --models resnet18 resnet50 densenet121 `
  --batch-size 1

Expected input layout:

my_inference_cases/
  imagesTs/
    case_001_0000.nii.gz
    case_001_0001.nii.gz
  labelsTs/
    case_001.nii.gz

labelsTs is optional. Outputs are written back to my_inference_cases/hipvascai_monai_classification_results/.

Package Contents

hipvascai_monai_classification_model/
  training/
    resnet18/
      final_average_5fold_model.pt
      cross_validation_summary.json
      fold_summary.csv
    resnet50/
      final_average_5fold_model.pt
      cross_validation_summary.json
      fold_summary.csv
    densenet121/
      final_average_5fold_model.pt
      cross_validation_summary.json
      fold_summary.csv
    model_comparison_summary.json
  training_manifest.json
  model_manifest.json

Citation

If you use these models, cite the associated HIPVASCAI repository and any accompanying publication or dataset documentation.