Oral disease ditection and cancer ditactionin mouth.pptx

oral disease ditection

1. Classification of Oral Images as Cancer and Non-Cancer Using Swin Transformers Presented by : Vinayak Mani

2. Problem Statement Current Diagnosis Methods: Primarily reliant on biopsies and visual examinations by experts. Challenges: These methods are invasive, subjective, and time-consuming. Need: An accurate, non-invasive, and quick method for early detection of oral cancer. Proposed Solution: Utilizing Swin Transformers, a cutting-edge deep learning model, to automate and improve the classification process.

3. Introduction Oral Cancer: A significant health issue with high mortality rates if not detected early. Importance of Early Detection: Improves survival rates and treatment effectiveness. Objective: To explore the use of Swin Transformers for accurately classifying oral images as cancerous or non-cancerous.

4. Why ST for Oral Image Classification Architecture: Utilizes hierarchical structure and self-attention mechanisms. Components: Patch embedding, multi-head self-attention, shifted windows. Advantages: Efficient at handling high-resolution images and complex patterns. Relevance: Designed to efficiently process large images common in medical imaging. Performance: Demonstrates superior accuracy and robustness in various image classification tasks compared to traditional CNNs.

5. Model Architecture Patch Embedding: Converts images into smaller patches. Self-Attention: Focuses on important features within each patch. Shifted Windows: Enhances efficiency by processing patches in a shifted manner, capturing global context.

6. Accuracy The accuracy of the model is 79.93% and the validation accuracy is 85% on datasets contain total 450 images that categorize as cancer & non-cancer.

7. Result

8. Result

9. Conclusions AI is more accurate in diagnosing oral cancer as compared to the conventional method of diagnosis. Current Limitations: Data Scarcity: Limited availability of labeled medical images. Model Interpretability: Understanding how the model makes decisions. Future Directions: Enhanced Data Collection: Increasing the size and diversity of the dataset. Model Improvements: Fine-tuning model parameters and exploring hybrid models. Clinical Trials: Validating the model in real-world clinical settings.

10. THANK YOU !