Automated oral cancer detection with histopathological images using efficient deep-learning models

Oral cancers that frequently arise include those with squamous cell carcinoma (OSCC), which is a worldwide public health concern. Machine learning-based categorization is constrained by time, subjectivity, inconsistency, errors, and application, as well as by existing algorithms for segmentation, clustering, and methods, which have an influence on stability. This research presents an innovative approach for automated oral cancer detection using efficient deep-learning techniques employed to histopathological images. The proposed method focuses on detecting OSCC utilizing the oral cancer annotated dataset. In the preprocessing phase, a weighted median filter and contrast-limited adaptive histogram equalization are used to enrich picture quality by reducing noise and enhancing contrast. The segmentation phase utilizes the U-Net architecture, consisting of contracting and expansive paths with multilevel skip connections, to accurately segment the preprocessed images. The third phase employs an augmented convolutional neural network model, incorporating 22 layers, including convolutional, max-pooling, flattening, dense layers, and activation functions, to perform the classification. Evaluation of the suggested model is conducted based on various evaluation metrics, such as accuracy, intersection over union, and dice coefficient, demonstrating its potential as a reliable tool for automated monitoring for oral cancer and aiding in prevention and treatment at an early stage planning.