tamara-kostova/multiagentmed-multiclass-tumor

# Brain Tumor Multiclass Classifier

PyTorch checkpoint artifacts for the MultiAgentMedClassifier multiclass brain tumor
MRI task. Contains a DenseNet169 CNN classifier and a BiomedCLIP linear-probe
checkpoint for classifying brain MRI images into 12 categories (11 tumor subtypes
+ normal).

These are checkpoint files for the accompanying project loaders, not standalone
Transformers models.

## Model Description

- Task: multiclass brain tumor MRI classification (12 classes)
- CNN architecture: DenseNet169
- Vision-language backbone for probe: `microsoft/BiomedCLIP-PubMedBERT_256-vit_base_patch16_224`
- Framework: PyTorch

## Classes

| Index | Label |
|-------|-------|
| 0 | `glioma` |
| 1 | `meningioma` |
| 2 | `pituitary_tumor` |
| 3 | `carcinoma` |
| 4 | `germinoma` |
| 5 | `granuloma` |
| 6 | `medulloblastoma` |
| 7 | `neurocytoma` |
| 8 | `papilloma` |
| 9 | `schwannoma` |
| 10 | `tuberculoma` |
| 11 | `normal` |

## Files

- `multiclass_tumor/cnn/densenet169_MRI_tumor_multiclass_norm_final.pt`: DenseNet169 CNN checkpoint for multiclass brain tumor MRI classification.

• multiclass_tumor/biomedclip/linear_probe_BiomedCLIP_MRI_tumor_multiclass_norm_best.pt: BiomedCLIP linear-probe checkpoint for multiclass brain tumor MRI classification.

  Dataset

  Trained on a combination of three publicly available brain MRI datasets:

  • Brain Tumor MRI Images — 17 Classes (Kaggle): https://www.kaggle.com/datasets/fernando2rad/brain-tumor-mri-images-17-classes
  • Brain Tumor MRI Images — 44 Classes (Kaggle): https://www.kaggle.com/datasets/fernando2rad/brain-tumor-mri-images-44c
  • Brain Tumor Dataset (Figshare): https://figshare.com/articles/dataset/brain_tumor_dataset/1512427

  Training Details

  • Input size: 224 x 224 RGB
  • Normalization: ImageNet mean/std
  • CNN checkpoint: DenseNet169 fine-tuned for the multiclass_tumor task
  • BiomedCLIP probe: linear/MLP probe over frozen BiomedCLIP image features

  Inference Example

  from huggingface_hub import hf_hub_download
  from agents.cnn_tool import CNNClassifier
  from config import DEFAULT_CONFIG
  
  checkpoint_path = hf_hub_download(
      repo_id="tamara-kostova/multiagentmed-multiclass-tumor",
      filename="multiclass_tumor/cnn/densenet169_MRI_tumor_multiclass_norm_final.pt",
  )
  DEFAULT_CONFIG.model.cnn_checkpoints["multiclass_tumor"] = checkpoint_path
  classifier = CNNClassifier(DEFAULT_CONFIG.model, DEFAULT_CONFIG.preprocess)
  result = classifier.classify("path/to/brain_mri.png", task="multiclass_tumor")
  print(result)
  

  Intended Use

  Research and experimentation only. Not a medical device. Always validate on your own held-out test set before using in any pipeline.