Enhanced Detection of Oral Squamous Cell Carcinoma from Histopathological Images Using Convolutional Neural Network Architectures
Gurjot Kaur, Neha Vaishnavi Sharma, Rupesh Gupta · 2024
A major worldwide health issue marked by high morbidity and death; oral squamous cell carcinoma is caused mostly by delayed detection. Early identification of oral cancer using immediate medical intervention improves patient survival rates and therapy results. In this work, we built an image classification system to detect oral squamous cell carcinoma (OSCC) in histopathology images using a customized convolutional neural network (CNN). Normal and OSCC classes comprised the dataset's 4,946 training images, 126 test images, and 120 validation images. The model's resilience was improved using rescaling, shearing, zooming, and horizontal flipping among data augmentation methods. Comprising many convolutional, pooling, and dropout layers, the CNN model was trained using the Adam optimizer and category cross-entropy loss function. Following thirty epochs, the model attained 86.67% validation accuracy and 94.64% training accuracy. The model's performance on the test set was further evaluated generating an overall accuracy of 72%, a recall of 92%, and a precision of 76%. These results highlight the potential of the CNN-based approach in supporting pathologists in early oral cancer detection, hence improving clinical judgment and patient treatment. The outstanding accuracy highlights how well deep learning models analyze medical images and their possible use in clinical operations to raise diagnosis accuracy and results. The CNN-based image classification system heralds a revolution in early oral cancer detection by providing a consistent and fast tool for doctors fighting this common illness.