Personal Computer-Based Cephalometric Landmark Detection With Deep Learning, Using Cephalograms on the Internet

Abstract

Background: Cephalometric analysis has long been, and still is one of the most important tools in evaluating craniomaxillofacial skeletal profile. To perform this, manual tracing of x-ray film and plotting landmarks have been required. This procedure is time-consuming and demands expertise. In these days, computerized cephalometric systems have been introduced; however, tracing and plotting still have to be done on the monitor display. Artificial intelligence is developing rapidly. Deep learning is one of the most evolving areas in artificial intelligence. The authors made an automated landmark predicting system, based on a deep learning neural network. Methods: On a personal desktop computer, a convolutional network was built for regression analysis of cephalometric landmarks’ coordinate values. Lateral cephalogram images were gathered through the internet and 219 images were obtained. Ten skeletal cephalometric landmarks were manually plotted and coordinate values of them were listed. The images were randomly divided into 153 training images and 66 testing images. Training images were expanded 51 folds. The network was trained with the expanded training images. With the testing images, landmarks were predicted by the network. Prediction errors from manually plotted points were evaluated. Results: Average and median prediction errors were 17.02 and 16.22 pixels. Angles and lengths in cephalometric analysis, predicted by the neural network, were not statistically different from those calculated from manually plotted points. Conclusion: Despite the variety of image quality, using cephalogram images on the internet is a feasible approach for landmark prediction.

Machine Learning for the Diagnosis of Orthodontic Extractions: A Computational Analysis Using Ensemble Learning

“…We plan to implement object recognition and landmark detection on images commonly used in the clinical practice. There is some existing work in the field of cephalometric X-ray landmark detection [ 16 , 17 ], and an even more extensive body of work in biomedical image detection [ 18 ]. One could also exploit internal features [ 19 , 20 , 21 ] extracted from image recognition models to predict other diagnostic outputs.…”

Automated Skeletal Classification with Lateral Cephalometry Based on Artificial Intelligence

“…Through the last decade, there have been continuous efforts to improve the accuracy of this procedure, one of which being the incorporation of deep learning algorithms. The application of deep learning algorithms to cephalometric analysis has shown better performance, yet many of these studies focus on the detection of cephalometric landmarks (Arik et al 2017; Lee et al 2017; Torosdagli et al 2018; Nishimoto et al 2019). Considering the high potential of errors and bias associated with the conventional diagnostic methods through landmark detection, we hypothesized that the elimination of the intermediary process replaced by a direct system will result in improved performance.…”

Locating cephalometric landmarks with multi-phase deep learning

“…Based on Keras with TensorFlow backend on Python, a regression neural network with 4 convolution layers, followed by 4 dense layers, which was used in our previous report 15 (Fig 3), was constructed, with input of 387 x 480 and output of 38 values (x, y each for 19 landmarks). Stochastic gradient descent was used as the optimizer.…”

“…Based on Keras with TensorFlow backend on Python, a regression neural network with 4 convolution layers, followed by 4 dense layers, which was used in our previous report 15…”

“…In a previous report, we built an automated landmark prediction system, based on a deep learning convolutional neural network 15 . Cephalograms were gathered through the internet and landmarks were plotted by ourselves.…”