Transillumination and HDR Imaging for Proximal Caries Detection

The purpose of this study was to develop an in vitro model for the validation of near-infrared transillumination (NIRT) for proximal caries detection, to enhance NIRT with high-dynamic-range imaging (HDRI), and to compare both methods, using micro-computed tomography (µCT) as a reference standard. Both proximal surfaces of 53 healthy or decayed permanent human teeth were examined using the Diagnocam (DC) (KaVo) and NIRT with HDRI (NIRT-HDRI). NIRT was combined with HDRI to improve the diagnostic performance by reducing under- and overexposed image areas. For NIRT-HDRI, an exposure series was captured and merged into a single HDR image. A classification was applied according to lesion depth. All surfaces were assessed twice by 2 trained examiners, and additionally with µCT for validation. The Kappa statistic was used to calculate inter-rater reliability and agreement between DC and NIRT-HDRI. Inter-rater reliability (weighted Kappa, wκ) showed very good agreement for the DC (0.90) and NIRT-HDRI (0.96). The overall agreement (wκ) was almost perfect (0.85). In the individual categories (0 to 4), the agreement (simple Kappa) ranged from almost perfect (category 4) to moderate (1 and 2) to substantial (categories 0 and 3). Sensitivity and specificity of sound surfaces, enamel, and dentin caries ranged from 0.57 to 0.99 and were similar for both methods in the different categories. NIRT-HDRI had a higher sensitivity for sound surfaces and enamel caries, as well as a higher specificity for dentin caries. Regarding the obtained images, HDRI allowed for the detection of caries within a greater range of luminance levels, resulting in a more detailed visualization of structures without under- or overexposure. However, HDRI this did not improve the diagnostics significantly. Distinguishing between a processed demineralized enamel and dentin lesions appears to be a problem specific to NIRT and cannot be balanced using HDRI.