Radiographs play a crucial role in establishing an accurate diagnosis and detecting various issues that would be otherwise difficult to observe. Within the field of orthodontics, a commonly utilized form of dental radiography is the panoramic image, which offers a comprehensive view of the teeth and jaws. This image is essential for early identification and assessment of multiple conditions and problems. Dental caries is an infectious ailment that initially targets the tooth enamel, resulting in the formation of a cavity. If left untreated, the decay progresses further into the tooth, potentially affecting the underlying dentin layer (Diniz et al., 2012).Discomfort starts to emerge, particularly when exposed to hot, cold, or sugary substances. The decay has the potential to impact the tooth's inner pulp, resulting in what is commonly referred to as a toothache. In more advanced cases, the invasion of bacteria into the periodontal ligament, bone, or gum can lead to the development of a dental abscess.
Within the biomedical context, this agent's actions translate into supplementary information assisting dentists in their decision-making processes. Patient management involves multiple stages, spanning diagnosis, treatment selection, and even surgical interventions. In this study's context, the focal agent aims for the classification of biomedical images,deep learning, and machine learning algorithms for the identification of pathological patterns in clinical studies. These classification tasks rely on decision-making tools. Even though, the inherent variation among challenges in patient diagnostic images for conventional algorithms used for classification. The algorithms have predominantly been trained and adjusted on limited datasets or specific cohorts. Over the past decade, representation learning, a subset of machine learning, has undergone a substantial resurgence, particularly within the realm of computer vision. These representation algorithms have notably propelled significant progress in object recognition, marking a noteworthy advancement (Ali et al., 2016). Recent advancements in deep learning algorithms and the availability of clinical dental images have markedly enhanced the reliability and accuracy of deep learning algorithms in aiding the prediction of dental caries using X-rays.
Dental cavities and gum disease are among the most prevalent dental issues worldwide. Dental cavities, commonly known as tooth decay, result from a complex interplay between bacteria in dental plaque and dietary sugars, generating acids that corrode tooth structure (Pitts et al., 2017). In contrast, periodontitis is a persistent inflammatory ailment affecting the teeth's supportive tissues, encompassing gums, periodontal ligaments, alveolar bone, and dental root cement. This multifaceted condition arises due to the buildup of dental calculus and bacterial plaque around teeth, provoking an immune reaction that damages tissues (Kinane et al., 2017).
Dental X-rays play a pivotal role in dental diagnostics by revealing hidden internal aspects of teeth and jaws that aren't apparent during a standard clinical check-up. Various types of dental X-rays existthat include cone beam completed tomography, panoramic, periapical, and bitewing X-rays. Bitewing X-rays are valuable for spotting dental cavities, while periapical X-rays identify cavities, bone loss fromperiapical lesions, and periodontitis. Panoramic and CBCT X-rays assess overall oral health, including the presence ofanomaliesas well asperiodontal diseases (The Use of Dental Radiographs., 2006).
Dental practices are revolutionized with dental X-rays, furnishing in-depth insights into oral structures. Early detection of cavities, gum diseases, and other issues aids in averting complications and enhancing the outcomes of treatment. Additionally, X-rays unveil concealed conditions like cysts, tumors, and impacted teeth that may elude clinical detection. They also facilitate treatment planning and monitoring of ongoing therapies (Tadinada.,2019). The rapid emergence of artificial intelligence (AI) technology is reshaping numerous sectors, including dentistry. This transformation is particularly pronounced in oral health due to the prevalent utilization of electronic health records and digitized imaging with the use of AI algorithms(Shan et al., 2021). Although this science is nascent and necessitates human oversight, the potential benefits in healthcare are significant (Mahdi et al., 2023).In recent times, the presence of clinical dental images and the advancement of deep learning algorithms have resulted in substantial enhancements in the precision and resilience of these algorithms for aiding in the diagnosis of a range of dental conditions.
A machinelearning strategy that employs a pre-trained model, like a CNN model, which has already learned pertinent features from an extensive image is referred to as transfer learning. This pre-trained model is then adjusted for a specific task using a smaller dataset. The initial pre-trained image models serve as a starting point for training the new model, with notable architectures such asXception, VGG, ResNet, and GoogLeNet Inception. Google LeNet Inception V 3 architecture was introduced in 2014 in oral health applications and stands out due to its impressive performance in the ImageNet Challenge for recognition of large-scale images, where it was trained on millions of images from the ImageNet dataset across 1000 object categories. Its original architecture, containing 22 deep layers, enables the extraction of diverse scale features by utilizing convolutional filters of varying sizes within the same layers (Martins et al., 2023).
ResNet, introduced in 2015, has become a cornerstone in deep learning, serving as a fundamental framework for subsequent advancements (Liu et al., 2020). It addresses the challenge of vanishing gradients in training very deep neural networks through the incorporation of residual connections. These shortcuts enable the bypassing of one or more layers, mitigating gradient-related issues. Alternately, some approaches integrate both traditional machine learning techniques like extreme gradient boosting, random forest, K-nearest neighbor, and support vector machine for classification.
1.1 Research Problem
The problem of the research study is to evaluate the application of deep learning models for the classification of medical images to diagnose dental caries obtained using bitewing X-rays.
1.2 Research Objective
The objectives of the research study are as given below;
-
To evaluate the algorithms like Naïve Bayes Classifier and K-Nearest Neighbour in the classification of X-ray images
-
To examine the transfer learning algorithms like LeNet, VGGNet, ShuffleNet, and MobileNet for image classification