Comparison of 5 protocols based on their abilities to use data extracted from digitized clinical radiographs to discriminate between patients with gingivitis and periodontitis

Background: This study was undertaken to compare 5 digital analytic protoco ls for their abilities to extract data from digital clinical radiographs an d discriminate between patients with gingivitis and periodontitis. Methods: Five digital-image analysis protocols were compared for their abil ities to discriminate between two groups of 24 patients each. One group was diagnosed with healthy gingiva (or gingivitis) and the second with periodo ntitis. These groups were previously evaluated in published studies that us ed fractal and morphologic analyses. Pre-existing clinical radiographs for each patient were digitized and regions of interest (ROIs) were placed on i nterdental bone in mandibular posterior quadrants. The 5 protocols used wer e: 1) MGB: a median filtration to remove high-frequency noise, a Gaussian f iltration to remove low-frequency noise, binarization of the resulting imag e, and quantification of the black pixels; 2) MGBS: the same protocol as MG B except for a skeletonization of the binary image and a quantification of the skeleton's pixels; 3) GBS: Gaussian filtration, binarization (threshold ing on the mean pixel value) of the resulting image, skeletonization, and q uantification of the pixels of the skeleton; 4) NS: normalization, skeleton ization, and quantification of the skeleton's pixels; and 5) S: a variation of NS, except normalization was not used. The resulting values for the 2 p atient groups were compared with Mann-Whitney U tests and effect likelihood -ratio test. Results: For digitized radiographs, the mean gray-scale value (+/- standard deviation) for gingivitis patients was 183.22 +/- 18.53 and for periodonti tis patients 181.26 +/- 17.20. Mann-Whitney U tests resulted in the followi ng P values for these protocols: MGBS <0.01; S <0.01; GBS <0.01; NS <0.01; and MGB <0.83. Effect likelihood-ratio tests indicated that only MGBS and S significantly contributed to models containing the other factors. Conclusions: Small variations to protocols affected the strength of the dis crimination between the gingivitis and periodontitis groups. While there is potential for morphologic analysis to be used to discriminate between pati ents with gingivitis and periodontitis, a robust technique was not identifi ed.