A comprehensive review on artificial intelligence-driven preprocessing, segmentation, and classification techniques for precision furcation analysis in radiographic images

The early-stage detection of dental problems from radiolucency images was important for a healthy oral function. X-Ray image was the most commonly used image modality in the diagnosis by providing detailed and overall contrast of teeths. Medical practitioners are preferring Computer Aided Diagnosis Systems(CADs) over the traditional manual systems. This review presents a comprehensive evaluation of CAD systems for dental radiographic images, with a specific emphasis on key elements including preprocessing, segmentation, and classification. The paper delved to provide an in-depth comparative analysis of different advanced techniques used in each phase of the CADs system. This analysis aims to evaluate their efficacy and potential applications in the field of medical image analysis especially furcation analysis. This study systematically reviews the existing literature to evaluate the performance of machine learning and deep learning-based methods in the three phases of CAD system development: preprocessing, segmentation, and classification. Periodontitis is a chronic inflammatory condition that affects the tissues surrounding and supporting the teeth, including the gums, periodontal ligament, and alveolar bone. It is one of the most prevalent dental diseases globally, with significant implications for oral health and overall well-being. A CAD is necessary so that the accuracy of the models can be increased. Preprocessing includes various methods to deal with different aspects of input images which increases the usability and acceptability of the dataset. A comprehensive investigation of both machine and deep learning denoising methods was carried out during the image preprocessing stage. Segmentation process was another important task to effectively crop or segment out the Region of Interest(ROI). Various AI techniques such as U-Net, Mask R-CNN, and FCN, were critically examined for segmentation for identifying and separating out Region of Interest (ROI). Classification was the last step of the process, for detection and prediction of periodontal problems using features extracted from ROI. In the classification phase, an extensive evaluation of machine learning and deep learning models was conducted. Classic classifiers, such as Support Vector Machines (SVM) and Random Forests(RF), and deep learning architectures, including Convolutional Neural Networks (CNN) and Recurrent Neural Networks (RNN) were analyzed and compared. The three processes, preprocessing, segmentation, classification can be validated and evaluated using different performance metrics like Peak Signal to Noise Ratio (PSNR), Structured Similarity Index (SSIM), dice metric, area overlap, accuracy, top-1/top-5 rate, mean Average Precision(mAP) and Mean Squared Error (MSE) etc. The paper presents the literature survey of the methods, process and evaluations used for detection or prediction of problems in respective radiolucency images datasets. A detailed overview and analytical comparison of different approaches and techniques used in preprocessing, segmentation and classification paradigms. A convolutional autoencoder based U-Net provided the improved results for denoising of medical images. Deep learning based models U-Net, RCNN showed better results than state-of-the-art. For classification, state of art AI models using Deep Neural Networks (DNN) exhibited promising results which increased their usability in the dental field. Further findings and challenges of respective CAD stages were provided, followed by discussion on the implications of AI in Dentistry. In terms of various approaches used for preprocessing,segmentation and classification, the deep learning approaches gave an edge to other paradigms, therefore becoming the root to create new methodologies involving them. Also, various research issues and challenges and future scope based on the study performed were discussed in the paper.