Innovative Approaches in Oncology: YOLOv5 and EfficientNet for Improved Oral Cancer Diagnosis

• Published in: 2024 Asia Pacific Conference on Innovation in Technology (APCIT) pp. 1 - 4
• Main Authors: Bordoloi, Dibyahash, Joshi, Kireet, Kukreja, Vinay, Sharma, Rishabh
• Format: Conference Proceeding
• Language: English
• Published: IEEE 26.07.2024
• Subjects: Accuracy; Atmospheric modeling; Deep learning; EfficientNet; Oncology; Oral Cancer; Robustness; Training; YOLO; YOLOV5
• DOI: 10.1109/APCIT62007.2024.10673457

Oral cancer early and accurate detection is an area where traditional diagnostic methods fall short considerably. It is, therefore, important to seek help from a healthcare professional at the first sign of any symptoms and ensure that the treatment is correctly administered early enough. This new model is a union of two effective deep learning approaches - the object detection capabilities of YOLOv5 and the efficiency of EfficientNet are involved in the development of the model to establish oral cancer diagnosis with more accuracy and efficiency. The model was developed to determine the grade of oral cancer ranging from four severity levels using images from the oral cavity, which is expected to bridge the gap in the development of diagnostic tools that are non-invasive, rapid, and with the capacity to provide accurate diagnoses. The dataset of oral cavity images we captured was exhaustive and employed actually, noisy images alongside different base angles, image shifts, and image scaling to make the training, validation, and testing of the hybrid model successful. The positive effect of YOLOv5 implementation was seen with the accurate detection of lesions in the scans, and with the EfficientNet the classification process became more precise. The model had been put through the most demanding training sessions and validation procedure, exhibiting an extraordinary capacity to describe and discriminate with accuracy among oral cancer's different severity grades. This model obtained an overall accuracy of 97.88% in detecting oral cancer, which is superior to standalone YOLOv5 and EfficientNet v2 implementations as well as existing diagnostic methods of oral cancer. The future work will aim at the validation of the model in clinical surroundings and studying its applicability to other cancers or medical conditions. The functionality of the new AI-powered diagnostic tools developed for oncology significantly will influence the strategy of care and overall performance of the opposition to cancer.