Optimization of Convolutional Neural Networks for Detecting Oral Cancer-Induced Bone Invasion: A Genetic Algorithm Approach

Introduction: For patients diagnosed with oral squamous cell carcinoma (OSCC), determining whether the tumor has penetrated the bone is crucial for therapy planning and surgical procedures. Computed Tomography (CT) imaging is preferred by radiologists due to its high sensitivity and specificity in detecting bone invasion. Objectives In this study, we present a deep learning-based Convolutional Neural Network (CNN) model, optimized using a genetic algorithm, to automatically detect bone invasion in OSCC cases. Methods: First, CT-scan images were collected fro S. M.S. hospital, Jaipur and annotated by the hospital’s experts. Further, all the collected images were changed from Digital Imaging and Communications in Medicine format (DICOM) to Portable Network Graphics (PNG) format as DICOM is very bulky to handle. Finally, GA is used to hyper-tune CNN parameters to get a good CNN architecture. The proposed model is compared with standard models like VGG16, VGG19, ResNet-50, MobileNetV2, DenseNet-121, and ResNet-101. Results: The proposed model achieved a classification accuracy of 95.6%, outperforming six state-of-the-art transfer learning models, including VGG16, VGG19, ResNet-50, MobileNetV2, DenseNet-121, and ResNet-101. Conclusions: The results demonstrate the effectiveness of our model in improving bone invasion detection accuracy, providing a valuable tool for clinical decision-making.