Approximal caries diagnosis on ink-jet printed radiographs versus monitor: ROC-analysis

Methods: 100 corresponding teeth were paired and exposed on a ccd-sensor using standardized projection geometry (50 direct digital radiographs). Teeth were selected in such, that roughly 50% of the 100 approximal surfaces according to microscopic evaluation (= ground truth) after serial sectioning had a non-cavitated enamel or enamel-dentine lesion. After individual printer calibration by means of a specifically designed and pre-evaluated test image radiographs were printed on glossy paper by three off-the-shelf ink-jet printers (Canon S500 and iP4500, Epson Stylus Photo R2400). 12 experienced observers judged the surfaces on the prints as well as on a calibrated 17"CRT-viewing monitor with respect to the visibility of a carious lesion using a five-point confidence scale. All viewing was done in a quiet room with subdued light. Data were evaluated by means of an extensive non-parametric Receiver-Operating-Characteristics (ROC) according to Hanley and McNeill. Areas (Az) underneath the ROC-curves represent diagnostic performance. The critical z-value indicates significant differences if z >± 2 (Hanley and McNeill 1983).

Results: Differences were not significant (range mean z: -0.4611 and -0.0125), with mean Az = 0.643 for monitor, followed by 0.636 (iP4500), 0.633 (R2400) and 0.628 (S500).

Conclusions: Since differences were neither significant nor were they relevant with respect to their magnitude (maximum difference between mean Az: 0.015) we conclude, that calibrated off-the-shelf ink-jet printers are capable to reproduce digital radiographs in a quality sufficient enough to carry out even challenging diagnostic tasks such as approximal non-cavitated lesion detection.