Upload vit_base_patch16_224 multi-label model (AUC: 0.7595)

bf0e12b verified 5 months ago

4.13 kB

tags:
  - pytorch
  - image-classification
  - medical-imaging
  - chest-x-ray
  - multi-label-classification
  - chexnet
  - vit-base-patch16-224
license: mit
datasets:
  - alkzar90/NIH-Chest-X-ray-dataset
language:
  - en
metrics:
  - accuracy
  - f1
  - roc_auc

🫁 ViT-Base-Patch16-224 - Multi-Label Chest X-ray Classification (14 Pathologies)

Ce modèle a été entraîné pour la classification multi-label de 14 pathologies thoraciques à partir de radiographies X-ray du dataset ChestX-ray14.

📋 Description

Architecture: ViT-Base-Patch16-224
Tâche: Classification multi-label (14 pathologies)
Dataset: NIH Chest X-ray (ChestX-ray14)
Framework: PyTorch
Image Size: 224×224

📊 Performance Globale

Métrique	Valeur
AUC-ROC (macro)	0.7595
AUC-ROC (micro)	0.8191
F1 (macro)	0.0916
mAP	0.2041

Comparaison avec l'article CheXNet

Modèle	AUC Macro	Δ
CheXNet (article)	0.8414	-
Notre modèle	0.7595	-0.0819

📈 Performance par Pathologie

Pathologie	AUROC	F1	Support
Atelectasis	0.7253	0.0457	3279
Cardiomegaly	0.8585	0.2737	1069
Effusion	0.7918	0.3613	4658
Infiltration	0.6662	0.2098	6112
Mass	0.7563	0.1563	1748
Nodule	0.6822	0.0322	1623
Pneumonia	0.6705	0.0000	555
Pneumothorax	0.7889	0.0729	2665
Consolidation	0.7221	0.0129	1815
Edema	0.8229	0.0796	925
Emphysema	0.7516	0.0159	1093
Fibrosis	0.7701	0.0000	435
Pleural_Thickening	0.7429	0.0000	1143
Hernia	0.8836	0.0227	86

🏷️ Les 14 Pathologies

ID	Pathologie
0	Atelectasis
1	Cardiomegaly
2	Effusion
3	Infiltration
4	Mass
5	Nodule
6	Pneumonia
7	Pneumothorax
8	Consolidation
9	Edema
10	Emphysema
11	Fibrosis
12	Pleural_Thickening
13	Hernia

⚙️ Configuration d'entraînement

{
  "data_variant": "full",
  "batch_size": 16,
  "image_size": 224,
  "num_classes": 14,
  "learning_rate": 0.0001,
  "num_epochs": 50,
  "scheduler": "ReduceLROnPlateau (factor=0.5, patience=5)",
  "optimizer": "AdamW (weight_decay=0.01)",
  "loss": "BCEWithLogitsLoss (non pond\u00e9r\u00e9e)"
}

🚀 Utilisation

import torch
from torchvision import transforms
from PIL import Image

# Charger le modèle
# Pour ViT
import timm

model = timm.create_model(
    'vit_base_patch16_224',
    pretrained=False,
    num_classes=14
)
model.load_state_dict(torch.load('pytorch_model.bin', map_location='cpu'))
model.eval()

# Préprocessing
transform = transforms.Compose([
    transforms.Resize((224, 224)),
    transforms.ToTensor(),
    transforms.Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225])
])

# Pathologies
PATHOLOGIES = [
    "Atelectasis", "Cardiomegaly", "Effusion", "Infiltration",
    "Mass", "Nodule", "Pneumonia", "Pneumothorax", "Consolidation",
    "Edema", "Emphysema", "Fibrosis", "Pleural_Thickening", "Hernia"
]

# Prédiction
image = Image.open('chest_xray.png').convert('RGB')
input_tensor = transform(image).unsqueeze(0)

with torch.no_grad():
    logits = model(input_tensor)
    probs = torch.sigmoid(logits)

# Afficher les probabilités
for name, prob in zip(PATHOLOGIES, probs[0]):
    print(f"{name}: {prob:.4f}")

📚 Citation

@inproceedings{Wang_2017,
    title = {ChestX-Ray8: Hospital-Scale Chest X-Ray Database and Benchmarks},
    author = {Wang, Xiaosong and Peng, Yifan and Lu, Le and Lu, Zhiyong and Bagheri, Mohammadhadi and Summers, Ronald M},
    booktitle = {IEEE CVPR},
    year = {2017}
}

@article{rajpurkar2017chexnet,
    title={CheXNet: Radiologist-Level Pneumonia Detection on Chest X-Rays with Deep Learning},
    author={Rajpurkar, Pranav and others},
    journal={arXiv preprint arXiv:1711.05225},
    year={2017}
}

📄 License

MIT License