Root fracture detection on digital images: effect of the zoom function.

Abstract

To assess the accuracy of root fracture detection at various levels of image magnification. Extracted single-rooted human teeth (100 non-fractured and 101 fractured) were mounted singly in wax blocks and a charge-coupled device (CCD)-based digital Sidexis imaging system was used to take images using the paralleling technique. Image sizes 1 : 1, 2 : 1, and 1 : 2 were prepared for each tooth. All 603 images were examined randomly and interpreted by an experienced radiologist unaware of which images were of fractured roots. The degree of agreement between image interpretation and the actual condition was estimated using a kappa statistic. A modified goodness-of-fit chi2-test was used to assess the significance of the difference between the three image sizes. The sensitivity, specificity, predictive value, and likelihood ratio were used to evaluate the accuracy of root fracture detection at each level of magnification. The degree of agreement between pairs of viewing conditions was also calculated using a kappa statistic. There was substantial agreement between image interpretation and the actual condition for both the 1 : 1 and 1 : 2 image sizes but only moderate agreement for the 2 : 1 size. The kappa values (and 95% CIs) were 0.65 (0.51-0.79), 0.63 (0.49-0.77), and 0.58 (0.44-0.72), respectively. The differences were not statistically significant (P-value = 0.79). The sensitivity (95% CI) of the image magnification ranged between 79.0% (70.0-87.5%) and 81.0% (72.2-87.5%). The 1 : 1 image size gave the highest specificity of 86.1% (95% CI: 78.1-91.6%). The positive predictive values (95% CI) of the 1 : 1, 1 : 2, and 2 : 1 images sizes were 84.9% (76.3-90.8%), 81.8% (73.1-88.2%), and 77.9% (69.0-84.8%), respectively. The negative predictive values (95% CI) were 80.6% (72.1-86.9%), 81.4% (73.8-88.9%), and 80.4% (71.4-87.1%) for the 1 : 1, 1 : 2, and 2 : 1 image sizes, respectively. The likelihood ratios of the positive test (95% CI) were 5.70 (3.47-9.36), 4.55 (3.00-6.98), and 3.56 (2.45-5,16) for the 1 : 1, 1 : 2, and 2 : 1 image sizes, respectively. The likelihood ratios of the negative test (95% CI) were 0.24 (0.17-0.36), 0.23 (0.15-0.35), and 0.25 (0.16-0.37) for the 1 : 1, 1 : 2, and 2 : 1 image sizes, respectively. The reliability between pairs of viewing conditions was substantial. The kappa statistics (95% CI) was 0.80 (0.72-0.88), 0.73 (0.64-0.83), and 0.77 (0.68-0.86), respectively. In this in vitro study, the three digital magnifications used did not affect the detection of root fractures.