Enhancing oral squamous cell carcinoma detection: a novel approach using improved EfficientNet architecture

Area of Science:

• Oncology
• Medical Imaging
• Artificial Intelligence

Background:

• Oral squamous cell carcinoma (OSCC) is a prevalent cancer globally, necessitating early diagnosis for effective treatment.
• Loss of structural integrity in oral cavity layers and membranes is a key characteristic of OSCC.
• Current diagnostic methods require timely and accurate detection to improve patient outcomes.

Purpose of the Study:

• To automate the early diagnosis of oral cancer using deep learning on histopathology images.
• To leverage advanced deep learning models for accurate classification of oral biopsy images.
• To facilitate prompt and precise detection of oral squamous cell carcinoma.

Main Methods:

• A deep learning convolutional neural network (CNN) model was employed for classifying benign and malignant oral biopsy images.
• Seventeen pretrained DL-CNN models were evaluated, with EfficientNetB0 identified as the superior model.
• The EfficientNetB0 model was further enhanced by integrating a dual attention network (DAN).

Main Results:

• The enhanced EfficientNetB0 model achieved high performance metrics: 91.1% accuracy, 92.2% sensitivity, and 91.0% specificity.
• The model demonstrated a low false-positive rate (1.12%) and a high F1 score (92.3%).
• The developed model outperformed existing state-of-the-art approaches in oral cancer detection.

Conclusions:

• Deep learning, particularly the enhanced EfficientNetB0 with DAN, shows significant promise for automated early oral cancer diagnosis.
• This approach facilitates early detection through oral histopathology image analysis.
• The advancement holds potential for improving the efficacy of oral cancer treatment strategies.