Table 4. The table shows relevant review findings of conventional machine learning algorithms for different imaging modalities.
| Author & Year | Relevant review findings | Images | Feature classifier | Detection |
|---|---|---|---|---|
| Imaging modality: Periapical X-rays | ||||
| (Li et al., 2005, 2007) | To segment the dental Image into normal, abnormal, and potentially abnormal areas, the variational level set function is used. | 60 X-rays | Trained SVM is used to characterize the normal and abnormal regions after Segmentation. | Bone loss & root decay |
| Imaging modality: Panoramic X-rays | ||||
| (Pushparaj et al., 2013) | The geometrical features are used to classify both premolar and molar teeth, while for tooth numbering, the matching templates method is used effectively. | N.A | Feature extraction (Projected principal edge distribution (PPED) + Geometric properties + Region descriptors) + SVM | Teeth numbering and Classification are used to help Forensic odontologists. |
| (Sornam & Prabhakaran, 2017) | The Linearly Adaptive Particle Swarm algorithm is developed and implemented to improve the accuracy rate of the neural system classifier. | N.A | Back Propagation Neural Network (BPNN) and Linearly Adaptive Particle Swarm Optimization (LA-PSO) | Caries detection |
| (Bo et al., 2017) | A two-stage SVM model was proposed for the Classification of osteoporosis. | Dataset consists of 40 images | HOG (histogram of oriented gradients + SVM | Osteoporosis detection |
| (Vila-Blanco, Tomás & Carreira, 2018) | Segmentation of mandibular teeth carried out by applying Random forest regression-voting constrained local model (RFRV-CLM) in two steps: The 1st step gives an estimate of individual teeth and mandible regions used to initialize search for the tooth. In the second step, the investigation is carried out separately for each tooth. | Training images: 261 Testing images: 85 |
(RFRV-CLMs) | Adult age teeth detection or a missing tooth for person identification. |
| Imaging Modality: Photographic color images | ||||
| (Fernandez & Chang, 2012) | Teeth segmentation and classification of teeth palate using ANN gives better results as compared to SVM. It shows that ANN is seven-times faster than SVM in terms of time | N.A | ANN + Multilayer perceptrons trained with the error back-propagation algorithm. | Oral infecto-contagious diseases, |
| (Prakash, Gowsika & Sathiyapriya, 2015) | The prognosticating faults method includes the following stages: pre-processing, Segmentation, features extraction, SVM classification, and prediction of diseases. | N.A | Adaptive threshold + Unsupervised SVM classifier | Dental defect prediction |
| Imaging modality: CBCT or CT | ||||
| (Yilmaz, Kayikcioglu & Kayipmaz, 2017) | Classifier efficiency improved by using the forward feature selection algorithm to reduce the size of the feature vector. The SVM classifier performs best in classifying periapical cyst and keratocystic odontogenic tumor (KCOT) lesions. | A total of 50 CBCT 3D scans | Order statistics (median, standard deviation, skewness, kurtosis, entropy) and 3D Haralick Features + SVM | Periapical cyst and keratocystic odontogenic tumor |
| Imaging modality: Hybrid dataset images | ||||
| (Nassar & Ammar, 2007) | A hybrid learning algorithm is used to evaluate the binary bayesian classification filters’ metrics and the class-conditional intensities. | Bitewing & Periapical films | Feature extraction + Bayesian classification. | Teeth are matching for forensic odontology. |
| (Avuçlu & Bacsçiftçi, 2019) | First, image pre-processing and segmentation are applied to extract the features and quantitative information obtained from the feature extraction from teeth images. Subsequently, features are taken as input to the multilayer perceptron neural network. | A total of 1,315 Dental X-ray images,162 different age groups | Otsu thresholding + Feature extraction (average absolute deviation) + Multilayer perceptron neural network | Age and gender classification |