Table 8. Dental X-ray image dataset description used for deep learning methods.
| Authors & Year | Dataset description |
|---|---|
| (Eun & Kim, 2016) | Periapical X-rays: 500 periapical images used for training where each Image is containing five teeth and 100 images used for testing with corresponding ground truth. |
| (Wang et al., 2016) | Total number of patients: 400 (100 additional patients) Cephalometric radiographs: 400 images .tiff format dimension of 1,935 × 2,400 pixels, 120 bitewing radiographs (new) (Age group 6 to 60 yrs) Software used: Soredex CRANEXr Excel Ceph machine (Tuusula, Finland) and Soredex SorCom software (3.1.5, version 2.0) |
| (Prajapati, Nagaraj & Mitra, 2017) | Periapical RadioVisio Graphy (RVG) X-ray: 251 images (labeled dataset) where 180 used for training, 26 images for testing & 45 images validation. |
| (Rana et al., 2017) | Color images: Training and validation data consist of 258 images & 147 images. |
| (Lee, Park & Kim, 2017) | A total of 300 Dental X-ray images with resolution 1,935 × 2,400 pixels and 150 images used for training, and 150 images used for testing. |
| (Srivastava et al., 2017) | Bitewing images: More than 3,000 images |
| (Miki et al., 2017a) | CBCT dataset taken from Asahi University Hospital, Gifu, Japan. Two dental units: Veraviewepocs 3D (J.Morita Mfg, Corp., Kyoto, Japan) and Alphard VEGA (Asahi Roentgen Ind. Co., Ltd., Kyoto, Japan |
| (Miki et al., 2017b) | CT data: 52 images, Training group: 42 images, testing group: 10 images |
| (Oktay, 2017) | Panoramic Images: Dataset taken from 3-different X-ray machines have image dimensions 2871 × 1577, 1435 × 791, or 2612 × 1244 pixels. Testing and validation are done using images of 100 different people. |
| (Yang et al., 2018) | A small dataset of 196 periapical images used, and also augmentation is performed. |
| (Zhang et al., 2018) | Periapical Images: Initially for training, 800 images and 200 used for testing. and data is annotated with the help of bounding box labels in 32 teeth position. |
| (Wirtz, Mirashi & Wesarg, 2018) | Panoramic X-rays: 14 test images used. Image augmentation is used to increase training images up to 4000. |
| (Choi, Eun & Kim, 2018) | Periapical X-rays: 475 images dimension of 300 × 413 from 688 × 944 or 1200 × 1650. |
| (Jader et al., 2018) | Panoramic X-ray images: A total of 193 images used for training containing 6987 teeth and 83 images for validation containing 3,040. |
| (Lee et al., 2018b) | Periapical Images: 3,000 images .jpeg format dimension resized to 299 × 299 pixels The training and validation dataset was 2,400 and a test dataset of 600. The training and validation dataset consisted of 1,200 dental caries and 1,200 non-dental caries, and the test dataset consisted of 300 dental caries and 300 non-dental caries. Augmentation is done up to ten times for training. |
| (Hatvani et al., 2018) | Micro CT images: a training set consists of 5,680 slices and a test set of 1,824 slices was used. |
| (Torosdagli et al., 2018) | CBCT dataset of 50 patients and qualitative visual inspection were done for 250 patients with high variability. |
| (Karimian et al., 2018) | Training is performed using different batches containing ten optical coherence tomography (OCT) images per batch. |
| (Lee et al., 2018a) | Periapical X-ray images resized to 224 × 224 pixels (from the original 1,440 × 1,920 pixels) in .png format : For training (n = 1,044), validation (n = 348), and test (n = 348) datasets. |
| (Egger et al., 2018) | CT dataset containing 45 images as DICOM files with dimension 512 × 512 from a department of craniomaxillofacial surgery in Austria. 1st Image set containing 1,680 slices, 2nd one with induced noise images 6720 slices, 3rd after transformation 13,440 slices, and 4th covered augmentation 18,480 slices |
| (Chu et al., 2018) | Panoramic X-ray: 108 images. |
| (Hiraiwa et al., 2019) | CBCT images and panoramic radiographs used for 760 mandibular first molars (400 patients) |
| (Lee et al., 2019) | Panoramic X-rays: Dimensions of 2,988 × 1,369 pixels. Total 846 annotated tooth images. Training group: 30 radiographs, Validation & testing: 20 images. Augmentation technique used to reduce overfitting ( obtained 1,024 training samples from 846 original data points ) |
| (Kim et al., 2019) | Panoramic Images:12,179 images (annotated by experts) Trained, validated, and tested using 11, 189, 190, and 800. |
| (Tuzoff et al., 2019) | Panoramic radiographs: 1,352 images Training group: 1,352 images, Testing group: 222 images |
| (Murata et al., 2019) | Panoramic X-rays: Total patients: 100 (50 men and 50 women), Training data for 400 healthy and 400 inflamed maxillary sinuses and data augmentation used to make 6,000 samples |
| (Kats et al., 2019) | Panoramic X-ray:65 images and augmentation performed. |
| (Fukuda et al., 2019) | Panoramic X-ray: 300 images with 900 × 900 pixels. Training set: 240 images, Testing set: 60 images |
| (Singh & Sehgal, 2020) | Panoramic X-rays: Total 400 images. Training group: 240 images, Testing group: 160 images. Also, augmentation is done by using transformation. |
| (Muramatsu et al., 2020) | Panoramic X-rays: 100 images dimension of 3,000 × 1,500 pixels used for testing and training both. |
| (Geetha, Aprameya & Hinduja, 2020) | Periapical X-rays: 105 images saved as in .bmp format dimension resized to 256 × 256, where caries identified 49 images Training, validation, and testing consists of 49 caries and 56 sound dental X-ray images. |
| (Banar et al., 2020) | Panoramic (OPGs) image dataset of 400 images used. |