Hybrid Mask R-CNN and Feature Fusion with MSLBP-AlexNet for Enhanced Caries Detection Using Ensemble AdaBoost Classifier

Abstract

Reduction in oral health quality demands professional dentists to achieve appropriate dental caries detection and classification skills as core elements of dental healthcare practice. The usage of traditional diagnostic approaches depends on manual interpretation by experts although they have proven effective but this approach makes it difficult to expand dental services and results in inconsistent findings. Automated detection systems have become more popular because deep learning and image processing technology advances make them better suited for improving diagnostic accuracy and efficiency. This method achieves its novelty by using collaborative advanced methods which include mask region-based convolutional neural network (Mask R-CNN) for image segmentation and AlexNet-based convolutional neural network (CNN) for deep feature extraction alongside ensemble adaptive boosting (ADABoost) for classification. A systematic method for dental caries detection in radiographs is proposed which unites state-of-the-art image segmentation and feature extraction and classification approaches. The proposed method begins with the Mask R-CNN technique because it achieves accurate dental image segmentation. For in-depth examination of individual teeth it is necessary to execute this segmentation process. The proposed method merges deep feature extraction with multi-scale local binary patterns (MSLBP) texture analysis through an implementation of AlexNet-CNN as its post-segmentation feature extraction technique. A combined vector featuring the retrieved elements serves as input for classification. Ensemble ADABoost provides high prediction accuracy through its combination of numerous weak classifiers. The proposed method demonstrates high precision for dental caries detection because of its validation process conducted through extensive data analysis. Multiple modern techniques are surpassed by this approach because it uses deep learning-based features and traditional texture analysis to develop complex dental image comprehension. The proposed method reaches a maximum accuracy level of 97.4% in its performance for the dental-bitewing X-Ray dataset.