Revolutionizing Oral Cancer Detection: An Approach Using Aquila and Gorilla Algorithms Optimized Transfer...

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computational Pathology

Background:

• Early oral cancer detection is crucial for patient survival but hindered by costly manual screenings.
• Deep learning presents a cost-effective, automated solution for oral cancer screening using histopathology images.

Purpose of the Study:

• To develop an accurate, automated framework for oral cancer classification from microscopic histopathology slides.
• To evaluate the efficacy of optimized deep learning models in improving oral cancer detection accuracy.

Main Methods:

• Utilized convolutional neural networks (CNNs) integrated with transfer learning (TL) models (VGG19, VGG16, MobileNet variants, NASNetMobile, DenseNet201).
• Fine-tuned CNNs using the Aquila Optimizer (AO) and Gorilla Troops Optimizer (GTO) metaheuristic algorithms.
• Trained and tested on the Histopathologic Oral Cancer Detection dataset (2494 normal, 2698 OSCC images).

Main Results:

• The Aquila Optimizer (AO) generally outperformed the Gorilla Troops Optimizer (GTO) across most transfer learning models.
• The DenseNet201 model achieved the highest accuracy, reaching 99.25% with AO and 97.27% with GTO.
• The optimized deep learning framework demonstrated superior performance in distinguishing between normal and oral squamous cell carcinoma (OSCC) images.

Conclusions:

• The proposed framework offers a significant advancement in automated oral cancer detection, enhancing diagnostic capabilities.
• Optimized deep learning models, particularly CNNs with metaheuristic algorithms like AO, show immense potential for medical image analysis and diagnostics.
• This approach addresses limitations in manual screening, paving the way for more accessible and accurate oral cancer diagnostics.