🦷 Oral Diseases Image Classification

A ResNet50-based deep learning model fine-tuned to classify six common oral diseases from intraoral images. This repository contains the best-performing model from a benchmark of four convolutional neural network architectures trained and evaluated under identical conditions.

🏆 Best Model: ResNet50

✅ Accuracy: 94.77%

🎯 Macro F1-Score: 0.9411

🧠 Framework: PyTorch

Model Overview

The model classifies the following six oral conditions:

• Calculus
• Caries
• Gingivitis
• Ulcers
• Tooth Discoloration
• Hypodontia

The final model was obtained using transfer learning with an ImageNet-pretrained ResNet50 and fine-tuned using a two-stage training strategy.

Benchmark Results

Repository Structure

checkpoints/
│── best_model.pth

notebooks/
│── oral-disseases-image-classification.ipynb

outputs/
│── models_comparison.csv
│── resnet50_confusion_matrix.png
│── resnet50_history.png
│── densenet121_confusion_matrix.png
│── densenet121_history.png
│── efficientnet_b0_confusion_matrix.png
│── efficientnet_b0_history.png
│── scratch_cnn_confusion_matrix.png
│── scratch_cnn_history.png

Gradio.py
README.md

Download

Model Weights

The trained checkpoint is available in:

checkpoints/best_model.pth

or can be downloaded directly from this repository.

Dataset

Training dataset:

https://huggingface.co/datasets/nsr51324/Oral_Diseases

Original source:

Oral Diseases Dataset (Kaggle)

How to Load the Model

from huggingface_hub import hf_hub_download
import torch

weights_path = hf_hub_download(
    repo_id="nsr51324/Oral_Diseases_Image_Classification",
    filename="checkpoints/best_model.pth"
)

checkpoint = torch.load(weights_path, map_location="cpu")
class_names = checkpoint["class_names"]

Inference

import torch
import torch.nn as nn
from torchvision.models import resnet50
from torchvision import transforms
from PIL import Image

model = resnet50(weights=None)

model.fc = nn.Sequential(
    nn.Dropout(0.3),
    nn.Linear(model.fc.in_features, len(class_names))
)

model.load_state_dict(checkpoint["state_dict"])
model.eval()

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

image = Image.open("sample.jpg").convert("RGB")
tensor = transform(image).unsqueeze(0)

with torch.no_grad():
    probabilities = torch.softmax(model(tensor), dim=1)[0]

prediction = class_names[probabilities.argmax().item()]

print(prediction)

Interactive Demo

A standalone Gradio application is included.

Run:

pip install torch torchvision gradio pillow huggingface_hub

python Gradio.py

Training Details

Training consisted of two stages:

1. Freeze the ResNet50 backbone and train the classifier head.
2. Unfreeze the backbone and fine-tune the entire network.

Data Augmentation

The following augmentations were applied during training:

• Random Resized Crop
• Horizontal Flip
• Rotation
• Color Jitter
• Random Erasing

Evaluation

The repository includes:

• Confusion matrices
• Training history
• Classification metrics
• Model comparison
• CSV benchmark results

See the outputs/ directory for complete evaluation results.

Intended Use

This model is intended for research, educational purposes, and AI experimentation.

It is not a certified medical device and must not be used as a substitute for professional clinical diagnosis.

License

This project is released under the MIT License.

Please refer to the dataset license before commercial use.

Author

Nasr Mohamed

AI Engineer

🤗 https://huggingface.co/nsr51324