A Novel Hybrid AI-Based System for Early Detection of Oral Squamous Cell Carcinoma via Histopathological Images
Mehran Ahmad, Department of Electrical Engineering, University of Engineering and Technology, Peshawar, Pakistan.
Muhammad Irfan Khattak, Department of Electrical Engineering, University of Engineering and Technology, Peshawar, Pakistan.
Atif Jan, Department of Electrical Engineering, University of Engineering and Technology, Peshawar, Pakistan.
Ihtisham Ul Haq, Intelligence Information Processing Lab, NCAI, University of Engineering and Technology, Peshawar, Pakistan.
Corresponding Author:
Mehran Ahmad (mehranahmad7242@gmail.com)
Abstract:
Oral cancer is a serious health condition that affects many people around the world. Oral cancer is a dangerous and potentially life-threatening form of cancer if not detected and treated early. It can spread quickly to other parts of the body, such as the lymph nodes and other organs, which can make treatment more difficult and increase the risk of mortality. Histopathological image analysis plays a crucial role in the diagnosis of Oral Squamous Cell Carcinoma (OSCC) by identifying abnormal cells. However, manual diagnosis is reliant on the skill and experience of doctors, which can be time-consuming when tracing all the tissues in a patient's biopsy. Furthermore, manual diagnosis can be limited by subjective differences in doctors’ opinions. To overcome these challenges and improve cure rates for patients, this research applied three different methodologies: Gabor+ CatBoost, ResNet50 + CatBoost, and Gabor+ ResNet50 + CatBoost. Here we extract 32 low-level features from the Gabor Filter and 100532 high-level features from the ResNet50 model. To reduce the dimensionality of the high-level features and avoid overfitting, we apply principal component analysis (PCA) and retain the top 4096 components. Firstly, the extracted features of Gabor and ResNet50 were classified individually through CatBoost. Secondly, the extracted features are then concatenated and fed into CatBoost for image classification. Among the three proposed strategies for classification, it was found that the third strategy, which involved the combination of Gabor filtering with ResNet50 feature extraction, and classification through CatBoost, achieved the best performance with the highest accuracy of 94.92%, 95.51% precision, 84.30% sensitivity, 95.54% specificity, 94.90% F1 score and 94.9% of AUC. These AI-based approaches offer promising solutions for more accurate and efficient OSCC diagnosis.
Keywords:
CatBoost; Gabor Filter; Histopathological Images; OSCC; PCA; ResNet50