We compared the agreement and variance of the CEJ localized by ultrasound and the clinical measurements (visual examination and tactile sensation) of all 153 teeth. The sample size of 153 had a power of 80% (sufficiently high by convention31) to detect a minimum difference of 0.25 mm (Fig. S2). We performed sub-group analysis across classifications of A- and B-, CL-S and CL-D, and three types of teeth. The descriptive statistics of the teeth used in this study based on sub-groups of classifications and tooth types are summarized in Table 1.
Identification of the CEJ via ultrasound images and comparison to clinical methods
Our first step was to use Bland-Altman analysis to compare the visually-identified distance with the ultrasound-identified distance (Fig. 1b). We found that 95% of the values fell within a range of 2.44 mm with a bias of 0.08 mm (ultrasound-based measurements are larger); the correlation analysis of the two distances showed a significant correlation with Pearson’s r = 0.6650 (Fig. S3a). Similarly, when ultrasound-based values were compared to tactile-based assessment, the 95% range was 2.66 mm with no systemic bias (Fig. 1c); the correlation analysis still showed a significant correlation but the Pearson’s r (0.6286) was 5.5% lower (Fig. S3b). The visual and tactile measurements showed a 95% CI of 1.42 mm with a bias of − 0.08 mm (Fig. 1d), and a Pearson’s r of 0.8924 (Fig. S3c). The ANOVA test indicated that the ultrasound measurements were not significantly different from either the visual or tactile measurements (p = 0.2434 and 0.9990, respectively), while the tactile measurements were significantly higher than the visual measurements (p = 0.0218) (Fig. 1e). To further quantify the error between the three different distances, we prepared a histogram (Fig. S3d), which showed that 71% (n = 108) had a relative difference within ± 20% for tactile versus ultrasound and 73% (n = 112) for visual versus ultrasound.
Comparisons of the CEJ identification between ultrasound and clinical measurements on Class A- (left) and Class B- (right). (a) and (f) show representative teeth in Class A- (visually detectable CEJ) and Class B- (visually undetectable CEJ), respectively. The agreement was characterized by Bland-Altman plot and Pearson correlation on teeth in (b)–(e) Class A- and (g)–(j) Class B-. Class A- shows higher agreement than Class B- for both US vs. Visual and US vs. Tactile.
Impact of CEJ integrity on ultrasound measurement accuracy
We next studied how the CEJ integrity impacted the accuracy of ultrasound assessment. Subgroup analysis thus used the American Academy of Periodontology classifications (Class A- and Class B-)27.
For Class A- (Fig. 2a), comparing the distances measured by ultrasound versus visual, the Bland-Altman analysis presented a mean bias of − 0.07 mm with 95% limits of agreement from − 0.93 mm to 0.79 mm (Fig. 2b). The correlation coefficient was 0.7690 (p < 0.0001) (Fig. 2c). Comparing the distances measured by ultrasound versus tactile, the mean bias was − 0.12 mm and 95% limits of agreement was from − 1.00 mm to 0.77 mm (Fig. 2d). The correlation coefficient was 0.7632 (p < 0.0001) (Fig. 2e). For Class B- (Fig. 2f), the mean bias of ultrasound versus visual was 0.23 mm and the 95% limits of agreement was from − 1.20 mm to 1.66 mm (Fig. 2g). The correlation coefficient was 0.6349 (p < 0.0001) (Fig. 2h). The mean bias of ultrasound versus tactile was 0.12 mm and the 95% limits of agreement was from − 1.51 mm to 1.74 mm (Fig. 2i). The correlation coefficient was 0.5742 (p < 0.0001) (Fig. 2j). Comparing ultrasound versus visual, the 95% range of Class B- (2.86 mm) was 66% larger than that of Class A- (1.72 mm), and he correlation coefficient of Class A- was 21% higher than that of Class B-. Comparing ultrasound versus tactile, the 95% range of Class B- (3.25 mm) was 84% larger than that of Class A- (1.77 mm), and the correlation coefficient of Class A- was significantly higher than that of Class B-.
The ANOVA test indicated that the ultrasound measurements were not significantly different from either the visual or tactile measurements for Class A- (Fig. S4a). For Class B-, the ultrasound measurements were significantly higher (p < 0.05) than the visual measurements. There was no significant difference (p > 0.05) between ultrasound measurement and tactile sensation on B- teeth (Fig. S4b). On 80% of Class A- teeth (n = 60), the ultrasound and visual measurements showed a relative difference within ± 20%; while for Class B- the ratio was only 62% (n = 48) (Fig. S4c). 81% teeth in Class A- (n = 61) had a relative difference between ultrasound and tactile measurements within ± 20%—this was only 66% for Class B- (n = 51) (Fig. S4d).
Moreover, we rounded the visual and tactile measurements in accordance with the clinical standard32,33 for comparison (Fig. S5). In periodontal standard, the probing measurements were rounded to integers making it challenging to monitor small periodontal disease progression. For this reason, comparing rounded clinical measurements with the ultrasound imaging could be a problem as ultrasound imaging can reach sub-100 μm resolution.
The correlation between ultrasound and rounded visual examination decreased from 0.77 to 0.68 in Class A-. The 95% limits of agreement increased from [− 0.93, 0.79 mm] to [− 1.13, 0.96 mm], Similarly, the correlation decreased from 0.76 to 0.67 for tactile sensation in Class A-, and the 95% limits of agreement increased from [− 1.00, 0.77 mm] to [− 1.20, 0.96 mm]. Interestingly, minor changes were observed in Class B-. The correlation between rounded visual examination and ultrasound decreased from 0.63 to 0.61 in Class B-. the 95% limits of agreement increased from [− 1.20, 1.66 mm] to [− 1.35, 1.73 mm]. The correlation decreased from 0.57 to 0.53 for tactile sensation in Class B-, and the 95% limits of agreement increased from [− 1.51, 1.74 mm] to [− 1.65, 1.77 mm]. These results further confirmed that rounded clinical measurements lowered the agreement with ultrasound measurements and should not be taken as the gold standard in the future ultrasound-based periodontal diagnosing.
Comparison of CEJ identification between ultrasound and clinical measurements on Class CL-S (left) and Class CL-D (right). (a) and (f) show representative teeth in Class CL-S (shallow cervical lesions) and Class CL-D (deep cervical lesions), respectively. The agreement was characterized by Bland-Altman plot and Pearson correlation on teeth in (b)–(e) Class CL-S and (g)–(j) Class CL-D. Class CL-S shows higher agreement than Class CL-D for both US vs. Visual and US vs. Tactile. (k) Measurement of cervical lesion (i.e., the step) depth using ultrasound imaging. (l)–(m) The variance between US vs. Visual and US vs. Tactile showed a significant correlation with the depth of NCCL on teeth in Class CL-D, indicating that deeper cervical lesions can cause greater bias between ultrasound-based and clinical CEJ identification. Red line represents the regression line, and black dashed lines are boundaries of the 95% confidence interval.
Impact of NCCL depth on ultrasound measurement accuracy
We noted a subgroup of Class B- teeth with deeper cervical lesions demonstrating more discrepancies among the detection methods. Therefore, we combined Class A- (all with no steps or shallow steps ≤ 0.2 mm) and 27 teeth of Class B- with shallow steps into CL-S (Fig. 3a), and assigned Class B- teeth with deep steps > 0.2 mm to CL-D (Fig. 3f).
Next, we analyzed the impact of cervical lesions on the CEJ identification accuracy using ultrasound by comparing the measurements in CL-S and CL-D. Comparing the distances measured by ultrasound versus visual examination, the Bland-Altman analysis presented a mean bias of – 0.06 mm with 95% limits of agreement from − 1.01 mm to 0.98 mm for CL-S (Fig. 3b). The correlation coefficient was 0.7367 (p < 0.0001) (Fig. 3c). For CL-D, the mean bias was 0.35 mm and the 95% limits of agreement was from − 1.03 mm to 1.74 mm (Fig. 3g). The correlation coefficient was 0.6219 (p < 0.0001) (Fig. 3h). To compare the distances measured by ultrasound versus tactile sensation, the mean bias was − 0.12 mm and 95% limits of agreement was from − 1.17 to 0.93 mm for CL-S (Fig. 3d). The correlation coefficient was 0.7431 (p < 0.0001) (Fig. 3e). For CL-D, the mean bias was 0.25 mm and the 95% limits of agreement was from − 1.42 to 1.92 mm (Fig. 3i). The correlation coefficient was 0.5065 (p < 0.0001) (Fig. 3j). The 95% range of CL-D (2.77 mm) was 39% larger than that of CL-S (1.99 mm). The correlation coefficient of Class CL-S was 18% higher than that of Class CL-D.
The ANOVA test indicated that the ultrasound measurements were not significantly different from either the visual or tactile measurements for Class CL-S (Fig. S6a). For Class CL-D, the ultrasound measurements were significantly higher than the visual measurements (p < 0.05), but ultrasound showed no significant difference than the tactile measurements (Fig. S6b). In Class CL-S, 76% teeth (n = 78) showed that the relative difference between ultrasound and visual measurements was within ± 20%; while for Class CL-D the ratio was only 59% (n = 30) (Fig. S6c). Here, 80% of teeth in CL-S (n = 82) had a relative difference between ultrasound and tactile measurements within ± 20%; for Class CL-D the ratio was only 59% (n = 30) (Fig. S6d).
The depth of the cervical lesion was measured as shown in Fig. 3k. For teeth in CL-D, the depth and the relative difference between ultrasound versus visual examination showed a significant positive correlation with Pearson’s r = 0.6607 (P < 0.0001), indicating that deeper cervical lesions caused greater bias across two methods (Fig. 3l). The depth of the cervical lesions and the relative difference between ultrasound versus tactile sensation also showed a significant positive correlation (r = 0.3081, P = 0.0278) (Fig. 3m), but the correlation coefficient was 53% lower than ultrasound versus visual examination.
Comparisons of the CEJ identification between ultrasound and clinical methods on different tooth types. (a)–(c) and (d)–(f) show Bland-Altman plots of US vs. Visual and US vs. Tactile for incisors, cuspids, and molars/premolars, respectively. (g)–(i) Box-and-whisker plots showing no significant differences across ultrasound, visual, and tactile measurements for all three types of teeth (p > 0.05).
Impact of tooth type on ultrasound measurement accuracy
We next investigated whether ultrasound analysis was impacted by tooth type: incisors, cuspids, or molars/premolars.
In the aspect of ultrasound versus visual measurements, the incisors showed a mean bias of 0.15 mm and a 95% range from − 0.93 to 1.23 mm (Fig. 4a). The correlation coefficient was 0.8356 (p < 0.0001) (Fig. S7a). For cuspids, the mean bias was 0.17 mm and the 95% range was from − 1.38 to 1.72 mm (Fig. 4b). The correlation coefficient was 0.5997 (p = 0.0052) (Fig. S7b). For molars/premolars, the mean bias was − 0.09 mm and the 95% range was from − 1.17 to 1.24 mm (Fig. 4c), and the correlation coefficient was 0.6891 (p < 0.0001) (Fig. S7c). The range of 95% confidence interval of cuspids was 44% and 29% larger than for incisors and molars/premolars, respectively, because of the limited number of cuspids available from the clinic34,35,36.
Then the ultrasound and tactile measurements were compared. The incisors had a mean bias of 0.09 mm and 95% range of − 1.24 to 1.43 mm (Fig. 4d), and a correlation coefficient of 0.8771 (p < 0.0001) (Fig. S5d). The cuspids presented a mean bias of − 0.10 mm and 95% range of – 1.36 to 1.56 mm (Fig. 4e), and a correlation coefficient of 0.6589 (p = 0.0016) (Fig. S7e). The molars/premolars showed a mean bias of − 0.17 mm and 95% range of − 1.36 to 1.24 mm (Fig. 4f), and a correlation coefficient of 0.6616 (p < 0.0001) (Fig. S7f).
Distances measured using ultrasound, visual, and tactile measurements showed no significant difference for all three types of teeth (Fig. 4g–i). 59% incisors (20 out of 34), 56% cuspids (5 out of 9), and 77% molars/premolars (53 out of 69) had a relative difference between ultrasound and visual measurement within ± 20% (Fig. S8a). 65% incisors (n = 22), 67% cuspids (n = 6), and 78% molars/premolars (n = 54) had a relative difference between ultrasound and tactile measurement within ± 20% (Fig. S8b).
Inter-rater reliability of clinical probing reading
Next, the inter-rater reliability was studied by comparing the reading of visual and tactile measurements between E1 and E2.
The visual distances measured by E1 was higher than E2 with a mean bias of 0.18 mm. The 95% range was 1.30 mm, with limits of agreement from − 0.47 mm to 0.83 mm (Fig. S9a). The visual measurements by E1 were significantly higher than E1 (p < 0.05, paired t-test) (Fig. S9b). The ICC was 0.86.
Similarly, the tactile distances measured by E1 was higher than E2 with a mean bias of 0.17 mm. The 95% range was 1.18 mm, with limits of agreement from − 0.42 mm to 0.76 mm (Fig. S9c). The tactile measurements by E1 were significantly higher than E1 (p < 0.05, paired t-test) (Fig. S9d). The ICC was 0.89.
- SEO Powered Content & PR Distribution. Get Amplified Today.
- PlatoData.Network Vertical Generative Ai. Empower Yourself. Access Here.
- PlatoAiStream. Web3 Intelligence. Knowledge Amplified. Access Here.
- PlatoESG. Carbon, CleanTech, Energy, Environment, Solar, Waste Management. Access Here.
- PlatoHealth. Biotech and Clinical Trials Intelligence. Access Here.
- Source: https://www.nature.com/articles/s41598-024-79081-z