A customm CNN model for the detection of PeriodontaDisease
Periodontal disease is a significant concern for global oral health, affecting a substantial portion of the population. Early detection and intervention are crucial to halt its progression and prevent associated complications. Deep Learning methods have emerged as promising tools for medical image analysis, including periodontal disease detection.
In this study, we present an innovative convolutional neural network (CNN) design customized for the diagnosis of periodontal disease. Our architecture is designed to optimize detection accuracy while maintaining computational efficiency. We performed an extensive comparative analysis, assessing our model against classic CNN architectures like VGG16, VGG19, AlexNet, ResNet50 etc. Our analysis covers accuracy metrics, computational performance, and model robustness. The results demonstrate the superior performance of our model, accurately detecting periodontal disease while remaining computationally efficient. Comparaative analysis with existing models further confirms the consistent outperformance of our proposed system.
This research contributes to advancing automated periodontal disease detection systems, potentially enhancing early diagnosis and treatment planning. The findings underscore the importance of leveraging deep learning techniques in dental imaging analysis and emphasize the significance of developing specialized architectures tailored to specific medical applications.
- Custom CNN architecture (PeriodontalNet), achieving 97% accuracy.
- Comparative analysis with 11 other models, including ResNet, VGG16, VGG19 and MobileNet.
- Visualization of training and testing accuracies using Matplotlib and Seaborn.
- Detailed performance metrics, including confusion matrix and classification report.
- Python
- PyTorch
- Matplotlib
- Seaborn
- Scikit-learn
-
Install PyTorch -pip install torch torchvision torchaudio
-
Install Matplotlib, Seaborn, Scikit-learn. -pip install matplotlib seaborn scikit-learn
Our custom CNN architecture (PeriodontakNet) is desiged to optimize detection accuracy while maintaining computational efficiency.
We compared our model with the following architectures:
- VGG16
- VGG19
- AlexNet
- ResNet50
- SqueezeNet
- MobileNet
- DenseNet121
- DenseNet201
- ResNet152
- ResNet101
The comparative analysis covers accuracy metrics, computational performance, and model robustness.Our model demonstrated superior performance across various metrics.
We used Matplotlib and Seaborn for visualizing training and testing accuracies, as well as for plotting the comparative analysis graph.
Contributions are welcome! Please fork the repository and submit a pull request.