Predictive Cloud-Based Application for the Early Detection and Risk Stratification of Oral Cancer using Keratin Pearls, Multiple Nucleoli and Vascular Invasion

Oral cancer is a significant public health concern, with high morbidity and mortality rates. Early detection is critical for improving patient outcomes and reducing the burden of the disease. Histopathological images, essential for distinguishing between normal, oral precancerous, and cancerous lesions, may miss critical diagnostic features of aggressiveness. Additionally, routine histopathology struggles to effectively identify high-risk OSCC patients. This work was conducted in collaboration with Ramaiah Institute of Technology and Ramaiah Dental College to improve OSCC detection through a pipeline-based computer-assisted screening approach involving classification, feature detection, and risk stratification. First, the classification module achieved an accuracy of 99.25% with EfficientNet in classifying cancerous and non-cancerous lesions. The feature detection module successfully identified key indicators of aggressiveness, including keratin pearls and multiple nucleoli using YOLO-NAS, while vascular invasion was detected by the EfficientNet classifier. Finally, we stratified cancer risk based on these detected features, producing a comprehensive pathology report that provides clinicians with valuable insights for informed diagnosis. Ultimately, this work aims to provide doctors with insights from analyses using models trained on key features like keratin pearls, multiple nucleoli, and vascular invasion.