|
|
--- |
|
|
license: mit |
|
|
language: |
|
|
- en |
|
|
metrics: |
|
|
- accuracy |
|
|
base_model: |
|
|
- facebook/convnextv2-tiny-1k-224 |
|
|
pipeline_tag: image-classification |
|
|
tags: |
|
|
- medical |
|
|
- transformers |
|
|
--- |
|
|
# MedConvNeXt: Optimized Skin Disease Classification |
|
|
|
|
|
## 📌 Introduction |
|
|
MedConvNeXt is a deep learning model based on ConvNeXt, optimized for skin disease classification using PyTorch Lightning. The model leverages hyperparameter tuning via Optuna to enhance its performance over multiple training iterations. |
|
|
|
|
|
## 📂 Dataset |
|
|
The dataset consists of images of various skin diseases, structured as follows: |
|
|
``` |
|
|
SkinDisease/ |
|
|
train/ |
|
|
class_1/ |
|
|
class_2/ |
|
|
... |
|
|
test/ |
|
|
class_1/ |
|
|
class_2/ |
|
|
... |
|
|
``` |
|
|
Data augmentation techniques such as **AutoAugment, horizontal flipping, rotation, color jittering, and random erasing** were applied to improve model generalization. |
|
|
|
|
|
## ⚙️ Model Architecture |
|
|
- **Base Model:** ConvNeXt-Base (pretrained on ImageNet) |
|
|
- **Optimizer:** AdamW with CosineAnnealingLR scheduler |
|
|
- **Loss Function:** CrossEntropyLoss / Focal Loss (for class imbalance handling) |
|
|
- **Evaluation Metrics:** Accuracy, Precision, Recall, and F1-score |
|
|
- **Hyperparameter Optimization:** Optuna (10 trials, 5 epochs per trial) |
|
|
|
|
|
## 📊 Training Process |
|
|
The model was trained using PyTorch Lightning with automatic logging to TensorBoard for real-time monitoring. The best hyperparameters were selected using Optuna, and the final model was trained over 23 epochs. |
|
|
|
|
|
## 🚀 Results |
|
|
Below are key performance graphs from TensorBoard: |
|
|
|
|
|
 |
|
|
|
|
|
- **Accuracy & Precision improved with hyperparameter tuning** |
|
|
- **Training loss consistently decreased, showing model convergence** |
|
|
|
|
|
## 🔗 How to Use |
|
|
To load and use the model: |
|
|
```python |
|
|
import torch |
|
|
from torchvision import transforms |
|
|
from PIL import Image |
|
|
|
|
|
# Load the model |
|
|
model = torch.jit.load("skinconvnext_scripted.pt") |
|
|
model.eval() |
|
|
|
|
|
# Define image transformation |
|
|
transform = transforms.Compose([ |
|
|
transforms.Resize((224, 224)), |
|
|
transforms.ToTensor(), |
|
|
transforms.Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225]) |
|
|
]) |
|
|
|
|
|
# Predict a sample image |
|
|
image = Image.open("sample.jpg").convert("RGB") |
|
|
image_tensor = transform(image).unsqueeze(0) |
|
|
output = model(image_tensor) |
|
|
predicted_class = torch.argmax(output, dim=1).item() |
|
|
print("Predicted Class:", predicted_class) |
|
|
``` |
|
|
|
|
|
## 📌 Future Work |
|
|
- **Clinical validation** on real-world medical datasets |
|
|
- **Model interpretability** via Grad-CAM or SHAP |
|
|
- **Deployment optimization** using ONNX and TensorRT |
|
|
|
|
|
## 📝 License |
|
|
This project is intended for research and educational purposes only. For clinical use, further validation is required. |
|
|
|
|
|
--- |
|
|
**Hugging Face Space:** [https://huggingface.co/spaces/Eraly-ml/Skin-AI] |
|
|
|
|
|
**Author:** [Eraly Gainulla] |
|
|
|
|
|
My telegram @eralyf |
