Table 2.
Data extraction.
| Author/year | Country | Architecture | Objective | Sample size | SDR (successful detection rate) |
|---|---|---|---|---|---|
| Alshamrani et al. (2022) (Alshamrani et al., 2022) | Saudi Arabia | CNN (autoencoder-based Inception layers) | Generate a Bjork–Jarabak and Ricketts cephalometrics automatically. | 100 | Basic autoencoder model trained on Set 1 2.0 mm: 64% 2.5 mm: 69% 3.0 mm: 72% 4.0 mm: 77% |
| 150 | Model autoencoder wider Paddup box set 2 2.0 mm: 71% 2.5 mm: 75% 3.0 mm: 78% 4.0 mm: 84% |
||||
| El-Fegh et al. (2008) (El-Fegh et al., 2008) | Libya/ Canada | CNN | A new approach to cephalometric X-ray landmark localization | > 80 | 2.0 mm: 91% |
| El-Feghi et al. (2003) (El-Feghi et al., 2003) | Canada | MLP | A novel algorithm based on the use of the Multi-layer Perceptron (MLP) to locate landmarks on a digitized X-ray of the skull | 134 | 2.0 mm: 91.6% |
| Hwang et al. (2021) (Hwang et al., 2021) | South Korea | CNN (YOLO version 3) |
To compare an automated cephalometric analysis based on the latest deep learning method | 200 | 2.0 mm: 75.45% 2.5 mm: 83.66% 3.0 mm: 88.92% 4.0 mm: 94.24% |
| Jiang et al. (2023) (Jiang et al., 2023) | China | CNN (A cascade framework “CephNet”) | Utilizing artificial intelligence (AI) to achieve automated landmark localization in patients with various malocclusions | 259 | 1.0 mm: 66.15% 2.0 mm: 91.73% 3.0 mm: 97.99% |
| Kafieh et al. (2009) (Kafieh et al., 2009) | Iran | ASM | As a new method for automatic landmark detection in cephalometry, they propose two different methods for bony structure discrimination in cephalograms. | 63 | 1.0 mm: 24.00% 2.0 mm: 61.00% 5.0 mm: 93.00% |
| Kim et al. (2020) (Kim et al., 2020) | South Korea | CNN | Develop a fully automated cephalometric analysis method using deep learning and a corresponding web-based application that can be used without high-specification hardware. | 100 | 2.0 mm: 84.53% 2.5 mm: 90.11% 3.0 mm: 93.21% 4.0 mm: 96.79% |
| Kim et al. (2021) (Kim et al., 2021) | South Korea | CNN | Propose a fully automatic landmark identification model based on a deep learning algorithm using real clinical data | 50 | 2.0 mm: 64.30% 2.5 mm: 77.30% 3.0 mm: 85.50% 4.0 mm: 95.10% |
| Lee et al. (2020) (Lee et al., 2020) | South Korea | BCNN | Develop a novel framework for locating cephalometric landmarks with confidence regions | 250 | 2.0 mm: 82.11% 2.5 mm: 88.63% 3.0 mm: 92.28% 4.0 mm: 95.96% |
| Oh et al. (2021) (Oh et al., 2020) | South Korea | CNN | They proposed a novel framework DACFL that enforces the FCN to understand a much deeper semantic representation of cephalograms | 150 | 2.0 mm: 86.20% 2.5 mm: 91.20% 3.0 mm: 94.40% 4.0 mm: 97.70% |
| 100 | 2.0 mm: 75.90% 2.5 mm: 83.40% 3.0 mm: 89.30% 4.0 mm: 94.70% |
||||
| Ramadan et al. (2022) (Ramadan et al., 2022) | Saudi Arabia | CNN | Detection of the cephalometric landmarks automatically | 150 | 2.0 mm: 90.39% 3.0 mm: 92.37% |
| 100 | 2.0 mm: 82.66% 3.0 mm: 84.53% |
||||
| Song et al. (2020) (Song et al., 2020) | Japan | CNN (with a backbone of ResNet50) | A two-step method for the automatic detection of cephalometric landmarks | 150 | 2.0 mm: 86.40% 2.5 mm: 91.70% 3.0 mm: 94.80% 4.0 mm: 97.80% |
| 100 | 2.0 mm: 74.00% 2.5 mm: 81.30% 3.0 mm: 87.50% 4.0 mm: 94.30% |
||||
| Song et al. (2021) (Song et al., 2021) | Japan/ China | CNN (Deep convolutional neural networks) | A coarse-to-fine method to detect cephalometric landmarks | 150 | 2.0 mm: 85.20% 2.5 mm: 91.20% 3.0 mm: 94.40% 4.0 mm: 97.20% |
| 100 | 2.0 mm: 72.20% 2.5 mm: 79.50% 3.0 mm: 85.00% 4.0 mm: 93.50% |
||||
| Song et al. (2020) (Song et al., 2019) | Japan/ China | CNN (Resnet50) | A semi-automatic method for detection of cephalometric landmarks using deep learning. | 150 | 2.0 mm: 85.00% 2.5 mm: 90.70% 3.0 mm: 94.50% 4.0 mm: 98.40% |
| 100 | 2.0 mm: 81.80% 2.5 mm: 88.06% 3.0 mm: 93.80% 4.0 mm: 97.95% |
||||
| Tanikawa et al. (2009) (Tanikawa et al., 2009) | Japan | N/A | Evaluate the reliability of a system that performs automatic recognition of anatomic landmarks and adjacent structures on lateral cephalograms using landmark-dependent criteria unique to each landmark | 65 | 88.00% |
| Ugurlu, (2022) (Uğurlu 2022) | Turkey | CNN | Develop an artificial intelligence model to detect cephalometric landmark, automatically enabling the automatic analysis of cephalometric radiographs | 180 | 2.0 mm: 76.20% 2.5 mm: 83.50% 3.0 mm: 88.20% 4.0 mm: 93.40% |
| Wang et al. (2018) (Wang et al., 2018) | China | Multiscale decision tree regression voting using SIFTbased patch | Develop a fully automatic system of cephalometric analysis, including cephalometric landmark detection and cephalometric measurement in lateral cephalograms. | 150 | 2.0 mm: 73.37% 2.5 mm: 79.65% 3.0 mm: 84.46% 4.0 mm: 90.67% |
| 165 | 2.0 mm: 72.08% 2.5 mm: 80.63% 3.0 mm: 86.46% 4.0 mm: 93.07% |
||||
| Yao et al. (2022) (Yao et al., 2022) | China | CNN | Develop an automatic landmark location system to make cephalometry more convenient | 100 | 1.0 mm: 54.05% 1.5 mm: 91.89% 2.0 mm: 97.30% 2.5 mm: 100.00% 3.0 mm: 100.00% 4.0 mm: 100.00% |
| Yoon et al. (2022) (Yoon et al., 2022) | South Korea | CNN (EfficientNetB0 (Eff-UNet B0) model) | Evaluate the accuracy of a cascaded two-stage (CNN) model in detecting upper airway soft tissue landmarks in comparison with the skeletal landmarks on lateral cephalometric images | 100 | 1.0 mm: 74.71% 2.0 mm: 93.43% 3.0 mm: 97.29% 4.0 mm: 98.71% |
| Yue et al. (2006) (Yue et al., 2006) | China | ASM | Craniofacial landmark localization and structure tracing are addressed in a uniform framework. | 86 | 2.0 mm: 71.00% 4.0 mm: 88.00% |
| Zeng et al. (2021) (Zeng et al., 2021) | China | CNN | A novel approach with a cascaded three-stage convolutional neural networks to predict cephalometric landmarks automatically. | 150 | 2.0 mm: 81.37% 2.5 mm: 89.09% 3.0 mm: 93.79% 4.0 mm: 97.86% |
| 100 | 2.0 mm: 70.58% 2.5 mm: 79.53% 3.0 mm: 86.05% 4.0 mm: 93.32% |
CNN: convolutional neural network, ASM: Active shape model, BCNN: Bayesian Convolutional Neural Networks, MLP: Multi-layer Perceptron.