Ceph-Net: automatic detection of cephalometric landmarks on scanned lateral cephalograms from children and adolescents using an attention-based stacked regression network

Abstract

Background: The success of cephalometric analysis depends on the accurate detection of cephalometric landmarks on scanned lateral cephalograms. However, manual cephalometric analysis is time-consuming and can cause inter- and intra-observer variability. The purpose of this study was to automatically detect cephalometric landmarks on scanned lateral cephalograms with low contrast and resolution using an attention-based stacked regression network (Ceph-Net).

Methods: The main body of Ceph-Net compromised stacked fully convolutional networks (FCN) which progressively refined the detection of cephalometric landmarks on each FCN. By embedding dual attention and multi-path convolution modules in Ceph-Net, the network learned local and global context and semantic relationships between cephalometric landmarks. Additionally, the intermediate deep supervision in each FCN further boosted the training stability and the detection performance of cephalometric landmarks.

Results: Ceph-Net showed a superior detection performance in mean radial error and successful detection rate, including accuracy improvements in cephalometric landmark detection located in low-contrast soft tissues compared with other detection networks. Moreover, Ceph-Net presented superior detection performance on the test dataset split by age from 8 to 16 years old.

Conclusions: Ceph-Net demonstrated an automatic and superior detection of cephalometric landmarks by successfully learning local and global context and semantic relationships between cephalometric landmarks in scanned lateral cephalograms with low contrast and resolutions.

Keywords: Cephalometric analysis; Cephalometric landmark; Deep learning; Landmark detection; Scanned lateral cephalogram.

Fig. 1

a Examples of scanned lateral cephalograms with labeling of 19 cephalometric landmarks. b 2D heatmap generations from manual labeling results

Fig. 2

The schematic diagram of the proposed method. a Data collection and manual labeling of cephalometric landmarks. b Dataset composition. c 2D heatmap generation from manual labeling results. d The training process of the Ceph-Net. e The prediction and evaluation process of the Ceph-Net

Fig. 3

a The network architecture of the proposed Ceph-Net. The schematics of (b) and (c) are the dual attention module and multi-path convolution module, respectively

Fig. 4

Bar plots for detection performance of cephalometric landmarks from different detection networks. a presents the mean radial error of each cephalometric landmark from different detection networks. b presents the successful detection rate (less than 2.0 mm errors) of each cephalometric landmark from different detection networks. The abbreviation of each cephalometric landmark is shown in Fig. 1

Fig. 5

a-h Show cumulative curves of MREs by different detection networks tested on patients aged 8 to 16 years old, excluding 12 years old sequentially. The orange, green, blue, pink, and red lines indicate cumulative MREs of U-Net, SegNet, Dense U-Net, Attention U-Net, and Ceph-Net, respectively

Fig. 6

a-f Show representative detection results of cephalometric landmarks from different detection networks. The red points denote the detected landmarks by detection networks, while the blue points indicate the ground truth of cephalometric landmarks

Fig. 7

Bar plot for detection performance of cephalometric landmarks from different detection networks on five specific conditions in scanned lateral cephalograms. The bracket means the number of samples

Fig. 8

a-h Show representative detection results of cephalometric landmarks produced by Ceph-Net on the test dataset split by specific age (8 to 16 except for 12 years old). The red points denote the detected landmarks by detection networks, while the blue points present the ground truth of cephalometric landmarks