AI-Powered Oral Cancer Detection System

Inspiration

The idea for this project emerged from recognizing how often oral cancer goes undetected in its early stages due to the lack of accessible and efficient screening tools. Early detection is crucial for improving survival rates, yet many cases are only diagnosed after significant progression. We were inspired to bridge this gap using AI to assist healthcare providers, especially in underserved areas, in identifying early signs of oral cancer and potentially saving lives.

What it does

Our AI-powered system analyzes medical images like CT scans, and X-rays to quickly and accurately detect oral cancer. It differentiates between cancerous and non-cancerous tissues, offering real-time diagnostic feedback through a simple web interface. This allows doctors to make faster, more informed decisions, reducing delays in diagnosis and treatment.

How we built it

We collected a balanced dataset from kaggle website of 1,962 high-quality oral images and preprocessed them using resizing, normalization, rotation, and flipping. EfficientNetB0, a pre-trained deep learning CNN model, was chosen for classification due to its high accuracy. We implemented transfer learning and fine-tuning to improve performance while avoiding overfitting with techniques like dropout and early stopping. The system was then deployed through a user-friendly web interface built with Gradio, making it accessible for healthcare professionals .

Challenges we ran into

We faced several challenges during the project. First, image quality and variability were major issues, as the medical images came from multiple sources with different equipment and conditions. Additionally, achieving high accuracy was challenging; we experimented with several pre-trained models such as VGG16, ResNet50, and InceptionV3 before ultimately selecting EfficientNetB0, which offered the best balance of accuracy, efficiency, and generalization for our dataset.

Accomplishments that we're proud of

The system achieved an impressive 97% accuracy, with balanced precision and recall for both cancerous and non-cancerous classes. It successfully differentiates subtle features in medical images that might be missed by human analysis, offering reliable, real-time feedback. The web interface was designed to be intuitive, making it practical even in resource-limited settings, which demonstrates our commitment to improving healthcare accessibility .

What we learned

Through this project, we gained strong experience in medical image processing and deep learning techniques. We learned how to apply transfer learning effectively and how to improve model performance through fine-tuning and regularization methods. We also understood the importance of data quality and preprocessing in achieving reliable results. In addition, we developed skills in deploying AI models into real-world applications using user-friendly interfaces.

What's next for AI-Powered Oral Cancer Detection System

In the future, we plan to expand our dataset to include more diverse and clinically validated images to further improve model accuracy and reliability. We also aim to integrate the system with real hospital databases and electronic health records. Another goal is to enhance the model to detect different stages of oral cancer and possibly other oral diseases. Finally, we hope to develop a mobile application version to make the system more accessible and usable in remote and low-resource areas.

Built With

• accuracy
• cnn
• confusion-matrix
• efficientnetb0
• google-colab
• kaggle
• precision
• python
• tensorflow