Non-AI image sharpening made landmark visibility on 60%-dose lateral cephalometric radiographs comparable to routine 100% dose, and markedly sharpened images were significantly superior at 60% and 40% dose

Under mild sharpening (sharpening strength 55%, contrast ratio 20%), 60%-dose images gave visibility comparable to or numerically better than the routine 100%-dose image for four bony landmarks: Point A (OR 1.12, P = .83), N (OR 2.26, P = .11), PNS (OR 2.35, P = .1), and U1 (OR 1.35, P = .53), none reaching statistical significance. Prior dose-reduction efforts in dental radiography centered on acquisition-side radioprotective measures such as beam collimation, additional filtration, fastest compatible receptor type, and thyroid shielding; this work evaluates non-AI image sharpening as a post-acquisition approach for cephalometric dose reduction. Six reviewers, including a board-certified oral and maxillofacial radiologist, a board-certified orthodontist, and trainees, rated all 10 pairs of the first five-image set; ordinal logistic regression included reviewer identity as a covariate; first-set average-measure ICC (2,6) was 0.819-0.888 and weighted Cohen's kappa medians were 0.403-0.538.

Under marked sharpening (sharpening strength 70%, contrast ratio 70%), visibility of all five landmarks was significantly superior to the 100%-dose image at both 60% and 40% dose, for example ANS OR 14.6 (P = 3.96 x 10-7) and U1 OR 8.06 (P = 3.56 x 10-5) at 60%, and ANS OR 19.4 (P = 8.19 x 10-8) and N OR 7.94 (P = 3.67 x 10-5) at 40%. Unlike mild sharpening, which showed only trends, marked sharpening reached statistically significant superiority even at the lower 40% dose, indicating that the choice of sharpening strength and contrast parameters directly determines how much visibility is compensated. All 15 pairs of the second six-image set were rated by the same six reviewers; ordinal logistic regression reported ORs with 95% CIs and Wald-test P values; second-set ICC (2,6) was 0.791-0.984 and weighted kappa medians were 0.392-0.898, though the authors note the second set was immediately recognizable and its reliability values are indicative only.

At 20% dose, marked sharpening still yielded significantly higher visibility than the 100%-dose image for ANS (OR 5.85, P = .00043), N (OR 3.94, P = .0044), PNS (OR 3.15, P = .011), U1 (OR 4.15, P = .0035), and soft tissue (OR 2040, P = 1.03 x 10-9), while Point A showed only a trend toward improvement (OR 1.19, P = .71). This pushes the dose level at which sharpening compensates from 60% under mild sharpening to 20% under marked sharpening for most landmarks, and shows that different anatomical landmarks differ in sensitivity to dose reduction and sharpening. Based on the same paired-comparison and ordinal logistic regression framework, with all ORs calculated relative to the 100%-dose routine clinical image, where OR above 1 indicates superiority over the 100%-dose image.

Soft-tissue visibility was significantly better than the 100%-dose image at all three dose levels under mild sharpening (ORs 9.92, 6.13, and 4.56 at 60%, 40%, and 20%), and rose to 2040-3530 under marked sharpening. The gain in soft-tissue visibility far exceeded that of bony landmarks, suggesting the algorithm's local dynamic range optimization is especially effective in low-contrast regions. Soft-tissue visibility was also scored by the six reviewers on the 5-point comparative scale, with ICC of 0.888 in the first set and 0.984 in the second, the highest agreement among the rated items.

The result addresses the post-acquisition stage of cephalometric dose reduction and applies to settings using a storage-phosphor system, fixed exposure at 80 kVp and 200 mA, and an adult skull phantom as the imaging subject. For radiology and orthodontics, it suggests that software sharpening can lower the routine dose to 60% (mild sharpening) or lower (marked sharpening) while maintaining or improving landmark visibility without changing hardware; the authors accordingly propose that sharpening may deliver a dose-reduction effect comparable to a flat panel detector at substantially lower cost. What comes next is prospective clinical research using patient images, particularly in pediatric populations, to determine whether this visibility gain translates into diagnostic accuracy, landmark identification error, and effects on orthodontic diagnosis and treatment planning.

Readers should still watch several points: the visibility assessment was based entirely on adult dry skull phantom images without patient clinical images, so it is unclear whether results hold under real anatomy and tissue superposition; the outcome was subjective visibility scoring and did not directly measure diagnostic accuracy, landmark identification error, or impact on orthodontic decision-making, and the authors themselves emphasize this is technical feasibility rather than clinical validity; the second marked-sharpening image set was immediately recognizable to reviewers, which the authors consider a possible source of evaluation bias, so its higher reliability values are indicative only; and under mild sharpening at 20% dose, Point A (OR 0.35, P = .04) and ANS (OR 0.28, P = .017) were in fact significantly worse than the 100%-dose image, showing that sharpening does not compensate for dose reduction at every dose and every landmark, leaving its applicable lower limit and parameter choice an open question.