White Light Medical Image Based Oral Cancer Diagnosis Using an Ensemble Deep Learning Model

Oral cancer is a severe disease that significantly decreases the quality of life, as it is accompanied by noticeable symptoms such as communication difficulties, swallowing problems, and changes in facial appearance at advanced stages. With a five-year survival rate of only 63%, the disease demands greater attention toward improving early diagnosis and treatment rates. In this project, an ensemble deep learning model combining ResNet101, InceptionV3, and DenseNet201 is explored to diagnose white light images, a widely available and commonly used imaging technique in clinics. ResNet101 emphasizes hierarchical features, InceptionV3 provides multiscale analysis, and DenseNet201 excels in feature reuse. The system improves early cancerous lesion detection by operating in parallel and using a voting mechanism. This approach improves upon traditional methods for detecting early-stage cancerous lesions, as the ensemble model mitigates the weaknesses and errors of individual models more effectively. This project aims to leverage the wide availability of white light images alongside the advanced capabilities of multiple deep learning models for diagnosis of oral cancer.