A Hybrid Deep Learning Method for Early Recognition of Oral Malignancy Cell Carcinoma in Image Data

Abstract

The seventh most prevalent cancer in the world is oral squamous cell carcinoma (OSCC), often diagnosed through histopathological images, which require expert interpretation due to tumor heterogeneity. This work tackles the urgent need for automated and precise early identification of OSCC using histopathology images. Utilizing a publicly accessible dataset of 5192 biopsy images (2494 normal, 2698 malignant OSCC) captured at 100x magnification, the research proposes a novel hybrid deep learning (DL) framework. The methodology involves comprehensive data preprocessing, including resizing images to 224x224x3, followed by feature extraction and dimensionality reduction using PCA. To combat overfitting and class imbalance, data augmentation, notably SMOTE applied. A Support Vector Machine (SVM) classifier and a ResNet-18 model for robust feature learning form the system's foundation. With a remarkable accuracy of 98.1%, the hybrid model also had good precision (98.22%), recall (98.61%), and F1-score (98.44%). These results highlight the substantial potential of the ResNet-18+SVM model to serve as a reliable and efficient computer-aided diagnostic tool for OSCC, contributing to improved patient prognosis through timely intervention