Oral cancer detection via Vanilla CNN optimized by improved artificial protozoa optimizer

Area of Science:

• Medical Imaging and Diagnostics
• Artificial Intelligence in Healthcare
• Computational Biology

Background:

• Oral cancer detection remains a critical challenge in healthcare, necessitating advanced diagnostic tools.
• Existing deep learning models show promise but require further optimization for improved accuracy and reliability.
• Metaheuristic algorithms offer potential for fine-tuning complex neural network architectures.

Purpose of the Study:

• To develop and evaluate a novel, optimized Convolutional Neural Network (CNN) for enhanced oral cancer detection.
• To introduce an Improved Artificial Protozoa Optimizer (IAPO) for optimizing the CNN architecture.
• To compare the proposed method's performance against established state-of-the-art models.

Main Methods:

• A modified Vanilla CNN architecture was designed, incorporating batch normalization, dropout regularization, and a custom convolutional block.
• The Improved Artificial Protozoa Optimizer (IAPO) algorithm was developed and employed to optimize the Vanilla CNN parameters.
• A dataset of 1000 oral cancer patient images underwent preprocessing, including contrast enhancement, noise reduction, and data augmentation.

Main Results:

• The IAPO-optimized Vanilla CNN achieved a high accuracy of 92.5% in oral cancer detection.
• Performance was superior to ResNet-101 (90.1%) and DenseNet-121 (89.5%).
• Evaluated using standard metrics: accuracy, precision, recall, F1-score, and AUC-ROC.

Conclusions:

• The proposed IAPO-optimized Vanilla CNN presents a highly accurate and trustworthy method for oral cancer detection.
• The IAPO algorithm effectively optimizes CNNs, demonstrating superior search space exploration and avoidance of local optima.
• This advanced approach holds significant potential for improving early diagnosis and patient outcomes in oral oncology.