Table 3. Review findings of the image processing techniques using different imaging modalities.
| Author & Year | Relevant review findings | Total images | Detection/Identification |
|---|---|---|---|
| Imaging modality: Bitewing X-rays | |||
| (Mahoor & Abdel-Mottaleb, 2004) | For segmentation, adaptive thresholding methods is being used, then features are extracted, and teeth numbering is done using the Bayesian classification technique. | 50 | Teeth numbering |
| (Zhou & Abdel-Mottaleb, 2005) | Proposed segmentation using a window-based adaptive thresholding scheme and minimum Hausdorff distance used for matching purposes. | Training =102 images Testing = 40 images |
Human identification |
| (Nomir & Abdel-Mottaleb, 2005) | Results are improved by using a signature vector in conjunction with adaptive and iterative thresholding. | 117 | Human identification |
| (Nomir & Abdel-Mottaleb, 2007) | Iterative followed by adaptive thresholding used for the segmentation and features extracted using fourier descriptors after forcefield transformation then matching is done by using euclidian distance | 162 | Human identification |
| (Lai & Lin, 2008) | The B-spline curve is used to extract intensity and texture characteristics for K-means clustering to locate the bones and teeth contour. | N.A | Teeth detection |
| (Nomir & Abdel-Mottaleb, 2008) | The procedure starts with an iterative process guided by adaptive thresholding. Finally, the Bayesian framework is employed for tooth matching. | 187 | Human identification |
| (Harandi, Pourghassem & Mahmoodian, 2011) | An active geodesic contour is employed for upper and lower jaws segmentation. | 14 | Jaw identification |
| (Huang et al., 2012) | An adaptive windowing scheme with isolation-curve verification is used to detect missing tooth regions. | 60 | Missing teeth detection |
| (Prajapati, Desai & Modi, 2012) | A region growing technique is applied to the X-rays to extract the tooth; then, the content-based image retrieval (CBIR) technique is used for matching purposes. | 30 | Human identification |
| (Pushparaj, Gurunathan & Arumugam, 2013) | The tooth area's shape is extracted using contour-based connected component labeling, and the Mahalanobis distance (MD) is measured for matching. | 50 | Person identification |
| Imaging modality: Periapical X-rays | |||
| (Huang & Hsu, 2008) | Binary image transformations, thresholding, quartering, characterization, and labeling were all used as part of the process. | 420 | Teeth detection |
| (Oprea et al., 2008) | Simple thresholding technique applied for segmentation of caries. | N.A | Caries detection |
| (Harandi & Pourghassem, 2011) | Otsu thresholding method with canny edge detection is used to segment the root canal area. | 43 | Root canal detection |
| (Lin, Huang & Huang, 2012) | The lesion is detected using a variational level set method after applying otsu’s method. | 6 | Lesion detection |
| (Sattar & Karray, 2012) | Phase congruency based approach is used to provide a framework for local image structure + edge detection | N.A | Teeth detection |
| (Niroshika, Meegama & Fernando, 2013) | Deformation and re-parameterize are added to the contour to detect the tooth comer points. | N.A | Teeth detection |
| (Ayuningtiyas et al., 2013) | Dentin and pulp are separated using active contour, and qualitative analysis is conducted using the dentist’s visual inspection, while quantitative testing is done by measuring different statistic parameters. | N.A | Tooth detection |
| (Nuansanong, Kiattisin & Leelasantitham, 2014) | Canny edge detection was initially used, followed by an active contour model with data mining (J48 tree) and integration with the competence path. | Approx. 50 | Tooth detection |
| (Lin et al., 2014) | The otsu’s threshold and connected component analysis are used to precisely segment the teeth from alveolar bones and remove false teeth areas. | 28 | Teeth detection |
| (Purnama et al., 2015) | For root canal segmentation, an active shape model and thinning (using a hit-and-miss transform) were used. | 7 | Root canal detection |
| (Rad et al., 2015) | The segmentation is initially done using K-means clustering. Then, using a gray-level co-occurrence matrix, characteristics were extracted from the X-rays. | 32 | Caries detection |
| (Jain & Chauhan, 2017) | First, all parameter values defined in the snake model then initial contour points initializes, and at last canny edge detection extract the affected part. | N.A | Cyst detection |
| (Singh & Agarwal, 2018) | The color to mark the carious lesion is provided by the contrast limited adaptive histogram (CLAHE) technique combined with masking. | 23 | Caries detection |
| (Rad et al., 2018) | The level set segmentation process (LS) is used in two stages. The first stage is the initial contour creation to create the most appropriate IC, and the second stage is the artificial neural network-based smart level approach. | 120 | Caries detection |
| (Obuchowicz Rafałand Nurzynska et al., 2018) | K-means clustering applied considering intensity values and first-order features (FOF) to detect the caries spots. | 10 | Caries detection |
| (Devi, Banumathi & Ulaganathan, 2019) | The hybrid algorithm is applied using isophote curvature and the fast marching method (FMM) to extract the cyst. | 3 | Cyst detection |
| (Datta, Chaki & Modak, 2019) | The geodesic active contour method is applied to identify the dental caries lesion. | 120 | Caries detection |
| (Osterloh & Viriri, 2019) | It uses unsupervised model to extract the caries region. Jaws partition is done using thresholding and an integral projection algorithm. The top and bottom hats, as well as active contours, were used to detect caries borders. | N.A | Caries detection |
| (Kumar, Bhadauria & Singh, 2020) | The various dental structures were separated using the fuzzy C-means algorithm and the hyperbolic tangent gaussian kernel function. | 152 | Dental structures |
| (Datta, Chaki & Modak, 2020) | This method converts the X-ray image data into its neutrosophic analog domain. A custom feature called 'weight' is used for neutrosophication. Contrary to popular belief, this feature is determined by merging other features. | 120 | Caries detection |
| Imaging Modality: Panoramic X-rays | |||
| (Patanachai, Covavisaruch & Sinthanayothin, 2010) | The wavelet transform, thresholding segmentation, and adaptive thresholding segmentation are all compared. Where, the results of wavelet transform show better accuracy as compare to others. | N.A | Teeth detection |
| (Frejlichowski & Wanat, 2011) | An automatic human identification system applies a horizontal integral projection to segment the individual tooth in this approach. | 218 | Human identification |
| (Vijayakumari et al., 2012) | A gray level co-occurrence matrix is used to detect the cyst. | 3 | Cyst detection |
| (Pushparaj et al., 2013) | Horizontal integral projection with a B-spline curve is employed to separate maxilla and mandible | N.A | Teeth numbering |
| (Lira et al., 2014) | Supervised learning used for segmentation and feature extraction is carried out through computing moments and statistical characteristics. At last, the bayesian classifier is used to identify different classes. | 1 | Teeth detection |
| (Banu et al., 2014) | The gray level co-occurrence matrix is used to compute texture characteristics (GLCM) and classification results obtained in the feature space, focusing on the centroid and K-mean classifier. | 23 | Cyst detection |
| (Razali et al., 2014) | This study aims to compare the edge segmentation methods: Canny and Sobel on X-ray images. | N.A | Teeth detection |
| (Amer & Aqel, 2015) | The segmentation process uses the global Otsu’s thresholding technique with linked component labeling. The ROI extraction and post-processing are completed at the end. | 1 | Wisdom teeth detection |
| (Abdi, Kasaei & Mehdizadeh, 2015) | Four stages used for segmentation: Gap valley extraction, canny edge with morphological operators, contour tracing, and template matching. | 95 | Mandible detection |
| (Veena Divya, Jatti & Revan Joshi, 2016) | Active contour or snake model used to detect the cyst boundary. | 10 | Cyst detection |
| (Poonsri et al., 2016) | Teeth identification, template matching using correlation, and area segmentation using K-means clustering are used. | 25 | Teeth detection |
| (Zak et al., 2017) | Individual arc teeth segmentation (IATS) with adaptive thresholding is applied to find the palatal bone. | 94 | Teeth detection |
| (Alsmadi, 2018) | In panoramic X-ray images that can help in diagnosing jaw lesions, the fuzzy C-means concept and the neutrosophic technique are combinedly used to segment jaw pictures and locate the jaw lesion region. | 60 | Lesion detection |
| (Dibeh, Hilal & Charara, 2018) | The methods use a shape-free layout fitted into a 9-degree polynomial curve to segment the area between the maxillary and mandibular jaws. | 62 | Jaw separation + teeth detection |
| (Mahdi & Kobashi, 2018) | Quantum Particle Swarm Optimization (QPSO) is employed for multilevel thresholding. | 12 | Teeth detection |
| (Ali et al., 2018) | A new clustering method based on the neutrosophic orthogonal matrix is presented to help in the extraction of teeth and jaws areas from panoramic X-rays. | 66 | Teeth detection |
| (Divya et al., 2019) | Textural details extracted using GLCM to classify the cyst and caries. | 10 | Dental caries & cyst extraction |
| (Banday & Mir, 2019) | Edge detection method for the segmentation then, the Autoregression(AR) model is adopted, and AR coefficients are derived from the feature vector. At last, matching is performed using euclidean distance. | 210 | Human identification |
| (Fariza et al., 2019) | For tooth segmentation, the Gaussian kernel-based conditional spatial fuzzy c-means (GK-csFCM) clustering algorithm is used. | 10 | Teeth detection |
| (Aliaga et al., 2020) | The region of interest is extracted from the entire X-ray image, and segmentation is performed using k-means clustering. | 370 | Osteoporosis detection, mandible detection |
| (Avuçlu & Bacsçiftçi, 2020) | The Image is converted to binary using Otsu's thresholding, and then a canny edge detector is used to find the object of interest. | 1,315 | Determination of age and gender |
| Imaging modality: Hybrid dataset images | |||
| (Said et al., 2006) | Thresholding with mathematical morphology is performed for the segmentation. | A total of 500 Bitewing & 130 Periapical images. | Teeth detection |
| (Li et al., 2006) | The fast and accurate segmentation approach used strongly focused on mathematical morphology and shape analysis. | A total of 500 (Bitewing and Periapical images) | Person identification |
| (Al-sherif, Guo & Ammar, 2012) | A two-phase threshold processing is used, starting with an iterative threshold followed by an adaptive threshold to binarize teeth images after separating the individual tooth using the seam carving method. | A total of 500 Bitewing & 130 Periapical images | Teeth detection |
| (Ali, Ejbali & Zaied, 2015) | The Chan-vese model and an active contour without edges are used to divide an image into two regions with piece-constant intensities. | N.A | Teeth detection |
| (Son & Tuan, 2016) | The otsu threshold procedure, fuzzy C-means, and semi-supervised fuzzy clustering are all part of a collaborative framework (eSFCM). | A total of eight & 56 Image dataset (Bitewing + Panoramic) | Teeth structures |
| (Tuan, Ngan & Son, 2016) | It uses a semi-supervised fuzzy clustering algorithm – SSFC-FS based on the Interactive Fuzzy Satisficing method. | A total of 56 (Periapical & Panoramic) | Teeth structures |
| (Son & Tuan, 2017) | Semi-supervised fuzzy clustering algorithm combined with spatial constraints (SSFC-SC) for dental image segmentation. | A total of 56 (Periapical & panoramic images) | Teeth structures |
| (Tuan et al., 2018) | Graph-based clustering algorithm called enhanced affinity propagation clustering (APC) used for classification process and fuzzy aggregation operators used for disease detection. | A total of 87 (Periapical & Panoramic) | Disease detection |
| Imaging modality: Photographic color images | |||
| (Ghaedi et al., 2014) | Segmentation functions in two ways. In the first step, the tooth surface is partitioned using a region-widening approach and the Circular Hough Transform (CHT). The second stage uses morphology operators to quantify texture to define the abnormal areas of the tooth's boundaries. Finally, a random forest classifies the various classes. | 88 | Caries detection |
| (Datta & Chaki, 2015a) | The method uses a biometrics dental technique using RGB images. Segment individual teeth with water Shed and Snake’s help, then afterward incisors teeth features are obtained to identify the human. | A total of 270 images dataset | Person identification |
| (Datta & Chaki, 2015b) | The proposed method introduces a method for filtering optical teeth images and extracting caries lesions followed by cluster-based Segmentation. | 45 | Caries detection |
| (Berdouses et al., 2015) | The proposed scheme included two processes: (a) identification, in which regions of interest (pre-cavitated and cavitated occlusal lesions) were partitioned, and (b) classification, in which the identified zones were categorized into one of the seven ICDAS classes. | 103 | Caries detection |
| Imaging modality: CT & CBCT | |||
| (Gao & Chae, 2008) | The multi-step procedure using thresholding, dilation, connected component labeling, upper-lower jaw separation, and last arch curve fitting was used to find the tooth region. | N.A | Teeth detection |
| (Hosntalab et al., 2010) | Otsu thresholding, morphological operations, and panoramic re-sampling, and variational level set were used. Following that, feature extraction with a wavelet-Fourier descriptor (WFD) and a centroid distance signature is accomplished. Finally, multilayer perceptron (MLP), Kohonen self-organizing network, and hybrid structure are used for Classification. | 30 Multislice CT image (MSCT) dataset consists of 804 teeth | Teeth detection and Classification |
| (Gao & Chae, 2010) | An adaptive active contour tracking algorithm is used. In which the root is tracked using a single level set technique. In addition, the variational level was increased in several ways. | A total of 18 CT images | Teeth detection |
| (Mortaheb, Rezaeian & Soltanian-Zadeh, 2013) | Mean shift algorithm is used for CBCT segmentation with new feature space and is compared to thresholding, watershed, level set, and active contour techniques. | A total of two CBCT images | Teeth detection |
| (Gao & Li, 2013) | The volume data are initially divided into homogeneous blocks and then iteratively merged to produce the initial labeled and unlabeled instances for semi-supervised study. | N.A | Teeth detection |
| (Ji, Ong & Foong, 2014) | The study adds a new level set procedure for extracting the contour of the anterior teeth. Additionally, the proposed method integrates the objective functions of existing level set methods with a twofold intensity model. | A total of ten CBCT images | Teeth structure |
| (Hu et al., 2014) | Otsu and mean thresholding technique combinedly used to improve the segmentation. | Image dataset consists of 300 layers | Teeth detection |