Kidney Ct Classifier Efficientnet

Model Description

Custom EfficientNet-style CNN for kidney CT scan classification (101M params, 95%+ accuracy)

This is a custom EfficientNet-style CNN trained from scratch for kidney CT scan classification. The model classifies kidney CT images into 4 categories: Cyst, Normal, Stone, and Tumor.

Model Details

• Model Type: Custom EfficientNet-style Convolutional Neural Network
• Architecture: 101M parameters, 7 stages with MBConv blocks
• Input Resolution: 384x384x3 RGB images
• Number of Classes: 4 (Cyst, Normal, Stone, Tumor)
• Framework: PyTorch 2.0+
• Training Precision: BF16 mixed precision on NVIDIA A100
• No Pretrained Weights: Trained from scratch on medical imaging data

Performance

Test Set Results

• Accuracy: 95.00%
• F1-Score: 0.9400

Per-Class Performance

Training Details

The model was trained on the CT Kidney Dataset with the following approach:

• Custom EfficientNet-style architecture built from scratch
• 101 million trainable parameters
• Width multiplier: 1.4, Depth multiplier: 1.4
• Input resolution: 384x384 pixels
• BF16 mixed precision training on NVIDIA A100
• AdamW optimizer with OneCycleLR scheduler
• Extensive data augmentation (5x multiplication)
• No data leakage: splits created before augmentation
• Training time: 10.5 hours on A100 40GB

Training Configuration

• Epochs: 40
• Batch Size: 48
• Optimizer: AdamW (lr=2e-3, weight_decay=2e-4)
• Scheduler: OneCycleLR with cosine annealing
• Loss Function: CrossEntropyLoss with label smoothing (0.1)
• Regularization: Dropout (0.35), Stochastic Depth (0-0.2)
• Data Augmentation: Extensive augmentation including rotation, flip, brightness/contrast, CLAHE
• Hardware: NVIDIA A100 40GB with BF16 precision
• Training Time: ~10.5 hours for 40 epochs

Data Split

• Training: 80% (augmented to ~78,000 images)
• Validation: 10% (original, no augmentation)
• Testing: 10% (original, no augmentation)
• No Data Leakage: Splits created before augmentation

Intended Use

Primary Use Cases

• Medical research and educational purposes
• Kidney disease classification from CT scans
• Computer-aided diagnosis (CAD) system development
• Medical imaging research

Limitations

• Model trained on specific dataset distribution
• Should not be used as sole diagnostic tool
• Requires clinical validation before medical use
• Performance may vary on images from different scanners or protocols

How to Use

Installation

pip install torch torchvision pillow

Inference Example

import torch
from PIL import Image
from torchvision import transforms

# Load model
model = torch.load('model.pth')
model.eval()

# Prepare image
transform = transforms.Compose([
    transforms.Resize((384, 384)),
    transforms.ToTensor(),
    transforms.Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225])
])

# Predict
image = Image.open('kidney_ct.jpg').convert('RGB')
image_tensor = transform(image).unsqueeze(0)

with torch.no_grad():
    output = model(image_tensor)
    probs = torch.softmax(output, dim=1)
    pred = output.argmax(1).item()

classes = ['Cyst', 'Normal', 'Stone', 'Tumor']
print(f"Prediction: {classes[pred]} ({probs[0][pred].item()*100:.1f}% confidence)")

Dataset

Name: nazmul0087/ct-kidney-dataset-normal-cyst-tumor-and-stone

The model was trained on the CT Kidney Dataset containing 12,446 CT scan images across 4 classes. The dataset consists of coronal and axial cuts from PACS systems, verified by medical professionals.

Ethical Considerations

• This model is for research and educational purposes only
• Not FDA approved or clinically validated
• Should not replace professional medical diagnosis
• Requires human oversight and clinical validation
• May have biases from training data distribution

Citation

If you use this model in your research, please cite:

@misc{kidneyctclassifierefficientnet,
author = {Arko007},
title = {Kidney Ct Classifier Efficientnet},
year = {2025},
publisher = {Hugging Face},
howpublished = {\url{[https://huggingface.co/](https://huggingface.co/)Arko007/Kidney Ct Classifier Efficientnet}}
}

License

Apache License 2.0 - See LICENSE file for details

Contact

For questions or issues, please open an issue on the model repository.

Acknowledgments

• Dataset: nazmul0087/ct-kidney-dataset-normal-cyst-tumor-and-stone
• Training infrastructure: NVIDIA A100 GPU
• Framework: PyTorch