A cascading learning method with SegFormer for radiographic measurement of periodontal bone loss

Abstract

Objective: Marginal alveolar bone loss is one of the key features of periodontitis and can be observed via panoramic radiographs. This study aimed to establish a cascading learning method with deep learning (DL) for precise radiographic bone loss (RBL) measurements at specific tooth positions.

Materials and methods: Through the design of two tasks for tooth position recognition and tooth semantic segmentation using the SegFormer model, specific tooth's crown, intrabony portion, and suprabony portion of the roots were obtained. The RBL was subsequently measured by length through these three areas using the principal component analysis (PCA) principal axis.

Results: The average intersection over union (IoU) for the tooth position recognition task was 0.8906, with an F1-score of 0.9338. The average IoU for the tooth semantic segmentation task was 0.8465, with an F1-score of 0.9138. When the two tasks were combined, the average IoU was 0.7889, with an F1-score of 0.8674. The correlation coefficient between the RBL prediction results based on the PCA principal axis and the clinicians' measurements exceeded 0.85. Compared to those of the other two methods, the average precision of the predicted RBL was 0.7722, the average sensitivity was 0.7416, and the average F1-score was 0.7444.

Conclusions: The method for predicting RBL using DL and PCA produced promising results, offering rapid and reliable auxiliary information for future periodontal disease diagnosis.

Clinical relevance: Precise RBL measurements are important for periodontal diagnosis. The proposed RBL-SF can measure RBL at specific tooth positions and assign the bone loss stage. The ability of the RBL-SF to measure RBL at specific tooth positions can guide clinicians to a certain extent in the accurate diagnosis of periodontitis.

Keywords: Alveolar bone resorption; Deep learning; Periodontitis; Segmentation; Tooth position recognition.

Fig. 1

Flowchart of tooth position recognition and tooth semantic segmentation. The workflow begins with the utilization of SegFormer for tooth position recognition on panoramic images to identify the locations of 32 teeth. Based on the tooth position data, specific regions containing teeth were extracted from the panoramic radiographs. After this extraction, semantic segmentation of the teeth was performed on the cropped regions. Ultimately, combining the tooth position information with the semantic segmentation results yielded the semantic segmentation of individual teeth. The direction of the blue arrow indicates the process sequence

Fig. 2

Manual X-ray bone loss measurement method and PCA measurement method based on tooth semantic segmentation results. (a) Two manual methods for RBL measurement are described as max (L1/L2, R1/R2) ×100 and max (L3/L4, R3/R4) ×100. The red lines indicate the length of the CEL to the ABC, and the green lines indicate the length of the CEJ to the AP. The yellow lines indicate the distance from the CEL to the ABC, and the blue lines indicate the distance from the CEJ to the AP. L and R represent proximal and distal, respectively. (b) The external ellipse of the tooth body and its principal axis were obtained via PCA. The red line indicates the tooth’s smallest enclosing ellipse, and the dark blue line represents the tooth’s longitudinal axis. The green dot represents the center of the red ellipse. (c) Length A of the principal axis of the ellipse passed through the convex hull of the suprabony portion of the root. (d) Length B of the principal axis of the ellipse passed through the convex hull of both the suprabony and intrabony portions of the root. PCA, principal component analysis; CEJ, cemento-enamel junction; ABC, alveolar bone crest; AP, apical point

Fig. 3

Diagram of the SegFormer model. SegFormer incorporates four transformer blocks, represented by the orange parts, with each capturing feature at scales of 1/4, 1/8, 1/16, and 1/32 of the original image dimensions. These features were unified to a consistent scale using an MLP layer and then processed through another MLP layer for classification. H and W indicate the input image’s height and width, respectively. C₁ to C₄ and C denote the feature map channel counts. N_cls specifies the class count. MLP, multilayer perceptron

Fig. 4

Scatter plot of the PCA principal axis method versus the measurements taken by clinicians. (a) and (b) scatter plots for all teeth, whereas (c)-(j) scatter plots for four distinct types of teeth. The x-axis denotes the PCA principal axis ratio, specifically A/B from Fig. 2, and the y-axis represents the measurement values. PCA, principal component analysis

Fig. 5

Semantic segmentation of periapical radiographs. (a) and (b) display two periapical radiographs, while (c) and (d) represent their respective semantic segmentation results