DenseNet121 CheXpert Multi-label (chexpert-densenet121-v1)

Model Description

This model is a fine-tuned DenseNet-121 (PyTorch) for multi-label classification of chest X-rays, trained on the Stanford CheXpert v1.0 dataset.
It predicts the presence of the following 14 labels (order preserved):

• No Finding
• Enlarged Cardiomediastinum
• Cardiomegaly
• Lung Opacity
• Lung Lesion
• Edema
• Consolidation
• Pneumonia
• Atelectasis
• Pneumothorax
• Pleural Effusion
• Pleural Other
• Fracture
• Support Devices

Author: Om Kumar (Hugging Face: @itsomk)

Model files included:

• chexpert_pytorch.safetensors — model weights saved with safetensors
• config.json — minimal config (backbone, num_labels, transforms)
• training_history.png — training curves

⚠️ Important: This model is provided for research and educational purposes only. Not for clinical use.

---

Intended Use

• Research in medical imaging and multi-label classification
• Educational use and reproducible baseline for further fine-tuning or adaptation
• NOT intended for clinical diagnosis or patient care. Use with caution; validate thoroughly before any downstream application.

---

Training Summary

• Backbone: DenseNet-121 (PyTorch torchvision.models.densenet121)
• Dataset: CheXpert v1.0 (Stanford)
• Uncertainty handling: U-Zeros (replace -1 with 0)
• Image size: 224 × 224
• Epochs: 20
• Batch size: 32
• Optimizer: Adam (lr=1e-4, weight_decay=1e-4)
• Loss: BCEWithLogitsLoss with per-class pos_weight
• Best validation mean AUC: 0.8176

Per-class AUC (validation)

• No Finding : 0.8762
• Enlarged Cardiomediastinum : 0.5959
• Cardiomegaly : 0.8165
• Lung Opacity : 0.8083
• Lung Lesion : 0.8230
• Edema : 0.8779
• Consolidation : 0.8527
• Pneumonia : 0.7559
• Atelectasis : 0.7117
• Pneumothorax : 0.8546
• Pleural Effusion : 0.9021
• Pleural Other : 0.9157
• Fracture : 0.7936
• Support Devices : 0.8622

---

Quick Usage (local safetensors)

import torch
from torchvision import models, transforms
from safetensors.torch import load_file
from huggingface_hub import hf_hub_download
from PIL import Image

REPO_ID = "itsomk/chexpert-densenet121"
FILENAME = "pytorch_model.safetensors"


local_path = hf_hub_download(repo_id=REPO_ID, filename=FILENAME)


class DenseNet121_CheXpert(torch.nn.Module):
    def __init__(self, num_labels=14, pretrained=False):
        super().__init__()
        self.densenet = models.densenet121(pretrained=pretrained)
        num_features = self.densenet.classifier.in_features
        self.densenet.classifier = torch.nn.Linear(num_features, num_labels)
    def forward(self, x):
        return self.densenet(x)


state = load_file(local_path)  

model = DenseNet121_CheXpert(num_labels=14, pretrained=False)

model.load_state_dict(state, strict=False)
model.eval()


preprocess = transforms.Compose([
    transforms.Resize((224, 224)),
    transforms.ToTensor(),
    transforms.Normalize(mean=[0.485,0.456,0.406], std=[0.229,0.224,0.225])
])


labels = [
 "No Finding","Enlarged Cardiomediastinum","Cardiomegaly","Lung Opacity",
 "Lung Lesion","Edema","Consolidation","Pneumonia","Atelectasis",
 "Pneumothorax","Pleural Effusion","Pleural Other","Fracture","Support Devices"
]

# inference
img = Image.open("path/to/xray.jpg").convert("RGB")
x = preprocess(img).unsqueeze(0)  
with torch.no_grad():
    logits = model(x)
    probs = torch.sigmoid(logits).squeeze().tolist()

results = {labels[i]: float(probs[i]) for i in range(len(labels))}
print(results)