Automated identification of cephalometric landmarks: Part 1-Comparisons between the latest deep-learning methods YOLOV3 and SSD

Abstract

Objective: To compare the accuracy and computational efficiency of two of the latest deep-learning algorithms for automatic identification of cephalometric landmarks.

Materials and methods: A total of 1028 cephalometric radiographic images were selected as learning data that trained You-Only-Look-Once version 3 (YOLOv3) and Single Shot Multibox Detector (SSD) methods. The number of target labeling was 80 landmarks. After the deep-learning process, the algorithms were tested using a new test data set composed of 283 images. Accuracy was determined by measuring the point-to-point error and success detection rate and was visualized by drawing scattergrams. The computational time of both algorithms was also recorded.

Results: The YOLOv3 algorithm outperformed SSD in accuracy for 38 of 80 landmarks. The other 42 of 80 landmarks did not show a statistically significant difference between YOLOv3 and SSD. Error plots of YOLOv3 showed not only a smaller error range but also a more isotropic tendency. The mean computational time spent per image was 0.05 seconds and 2.89 seconds for YOLOv3 and SSD, respectively. YOLOv3 showed approximately 5% higher accuracy compared with the top benchmarks in the literature.

Conclusions: Between the two latest deep-learning methods applied, YOLOv3 seemed to be more promising as a fully automated cephalometric landmark identification system for use in clinical practice.

Keywords: Artificial intelligence; Automated identification; Cephalometric landmarks; Deep learning; Machine learning.

Figure 1.

An image indicating the 80 cephalometric landmarks detected in the present study. Detailed landmark information is provided in Table 1.

Figure 2.

Comparison of the mean point-to-point errors between the You-Only-Look-Once version 3 (YOLOv3, red) and Single Shot Detector (SSD, blue) methods. The plot indicates that YOLOv3 was more accurate than SSD in general.

Figure 3.

Error scattergrams and 95% confidence ellipses from the YOLOv3 (red) and SSD (blue) methods. YOLOv3 resulted in a more uniformly distributed pattern of detection errors (more circular isotropic shaped ellipse) as well as higher accuracy (smaller sized ellipse) than SSD. (a) Errors after detecting porion. (b) Errors after detecting condylion.

Figure 4.

Compared with the top accuracy results in the previous literature, the proposed YOLOv3 shows approximately 5% higher success detection rates for all ranges.