Oral Cancer Detection By Using Tabular Data Synthesis and Classification

Area of Science:

• Artificial Intelligence
• Oncology
• Data Science

Background:

• Oral cancer screening relies on various data types, including clinical information.
• Distinguishing oral precancer from oral cancer is crucial for timely intervention.
• Clinical data imbalance poses a challenge for developing accurate AI screening models.

Purpose of the Study:

• To develop an AI-driven method for oral cancer screening using non-image clinical data.
• To improve classification performance by addressing data imbalance in clinical datasets.
• To evaluate the role of tabular clinical data in differentiating oral cancer from precancer.

Main Methods:

• Utilized deep learning techniques for classification of tabular clinical data.
• Implemented tabular data synthesis to balance imbalanced datasets.
• Conducted extensive experiments with multiple datasets and models to evaluate performance.

Main Results:

• Achieved a Youden index of ~0.74, balanced accuracy of ~0.83, and sensitivity of ~0.90.
• Demonstrated statistically significant improvements (p < 0.05) in key metrics using synthetic data.
• Confirmed the importance of tabular clinical information for oral cancer detection using AI.

Conclusions:

• AI analysis of tabular clinical data is effective for oral cancer screening.
• Synthetic data generation is a viable strategy to enhance AI model performance in oncology.
• This approach offers a promising direction for future clinical AI prediction tasks.