Table 3 Summary of recent literature
Sr. no. | Author | Year | Pre-processing | Technique | Diseases | Image type | Dataset | Results | Evaluation metrics |
|---|---|---|---|---|---|---|---|---|---|
1 | Korez et al. [33] | 2016 | 3D mesh shape model VB | Convolutional neural network 3D spatial VB probability maps | Vertebrae shape | MRI | 23 subjects | DC 93.4 ± 1.7% | Dice similarity coefficient |
2 | Arif et al. [34] | 2018 | Zero padding image dimension 100 × 100 | FCN deep probabilistic spatial regression shape-aware deep segmentation | Cervical vertebrae | X-ray | 296 images | DC 0.84 | Dice similarity coefficient |
3 | Shi et al. [35] | 2018 | Coronal spinal centerline intensity curve | 3D U-Net | Spine | CT | 61 images | DC 0.80 | Average dice coefficient |
4 | Jen-Tang Lu et al. [36] | 2018 | U-Net architecture spine-curve fitting | Multi-class convolutional neural network | Lumbar intervertebral discs | MRI | 4075 patients | 0.93 | Mean DSC |
5 | Davison et al. [37] | 2018 | – | CNN (CLM and RFRV) weighted heatmap loss | Pelvic | X-rays | 1696 images age 2–11 years | 6.92% and 5.85% | CLM method, having median error of 100–1000 training set, respectively |
6 | Kim et al. [38] | 2018 | Correlation map for ROI Hough transform and canny edge filtering | Line-based and graph cut method | Lumbar spine | MRI | 19 images | 90% | Dice similarity coefficient |
7 | Rehman et al. [39] | 2019 | – | FU-Net framework region-based deep U-Net | Osteoporotic | CT CSI 2014 CSI 2016 | 20 images 25 images | 92.8 ± 1.9% 95.4 ± 2.1% | Dice score of CSI2014 with fractured cases CSI 2016 |
8 | Chuang et al. [40] | 2019 | – | 3D U-Net and DeconvNet | Vertebral lumbar | CT xVertSeg.v1 | 25 Images | 88.4% | Dice coefficient |
9 | Lessmann et al. [41] | 2019 | Random elastic deformations, Gaussian smoothing | Fully convolutional neural network | Lumbar spine | CT CSI14 xVertSeg.v1 | 15 images 15 images | 94.9 ± 2.1% | Average dice score |
10 | Aubert et al. [42] | 2019 | Statistical shape modeling PCA | Simplified parametric model (CNN) | Scoliosis | X-rays | 400 images | 1.6 (1.3) mm, 1.8 (1.3) mm, and 2.3 (1.4) mm | Location error from 3D Euclidean distances vertebral center endplate centers pedicle centers |
11 | Pasha et al. [43] | 2019 | 3D coordinate of superior and inferior endplates centroids | 3D reconstruction of spine curve agglomerative hierarchical clustering | Adolescent idiopathic scoliosis (AIS) | X-rays | 103 AIS images 20 normal | 44% 56% | Maximum dissimilarity hypo-thoracolumbar kyphotic flat sagittal profile |
12 | Chen et al. [44] | 2019 | – | Fully convolutional neural network 3D HMM | Vertebrae identification and localization | CT scans MICAAI challenge | 242+60 images | 87.97% 2.56 mm | Mean identification rate mean error distance |
13 | Pastor et al. [45] | 2020 | – | Decision forest morphological operational refinement | Vertebrae identification and localization | CT scans | 232 images | 79.6% | Identification rate |
14 | Vergari et al. [46] | 2020 | – | CNN inspired by LeNet-5 | Scoliosis | Radiographs | 1892 images training 204 images validation | 96.5% | Correctly classified average accuracy |
15 | Alharbi et al. [47] | 2020 | CLAHE method | ResNet CNN Cobb angle with center points | Scoliosis | X-rays | 243 images | 90% | Accuracy |