TOPLINE

A smartphone-based three-dimensional reconstruction system called Digital Rulers of Ptosis (DRP) produced measurements similar to those made by oculoplastic specialists when assessing blepharoptosis, and correctly identified the clinical severity and surgical approach. In a 3-month postoperative subgroup of 248 eyes where treatment was guided by specialist examination, ptosis recurred in about 7%, compared with previously reported recurrence rates of about 14% to 63%.

METHODOLOGY

• A total of 952 patients (1904 eyes) were enrolled for the study across five hospitals in China throughout 2025; mean age was 27.9 years (range, 3-86 years) and included 493 females and 459 males (51.8% and 48.2%, respectively); same-day oculoplastic specialist assessment was the reference standard.
• Each patient underwent a 10-second smartphone video capture analyzed by DRP, which reconstructed 3 key frames — forward gaze, downgaze, and upgaze — to automatically calculate marginal reflex distance 1 and levator muscle function  in real-world units using iris-diameter-based unit conversion and real-time quality control.
• Primary outcomes included mean absolute error (MAE) and intraclass correlation coefficient (ICC) to measure how closely the system's measurements matched those of the specialists; area under the receiver operating characteristic curve (AUC), sensitivity, and specificity were used to assess how accurately the system diagnosed ptosis and identified the appropriate surgical approach.
• Among 206 participants (248 eyes) who underwent ptosis surgery, specialists remeasured marginal reflex distance 1 and levator muscle function at 3 months postoperatively to assess recurrence, and DRP's classification performance was also compared against a two-dimensional   deep-learning baseline model trained on the same video dataset.

TAKEAWAY

• Measurements from DRP differed from specialist measurements by a mean absolute error of 0.52 mm for marginal reflex distance 1  (95% CI, 0.50-0.54) and 0.95 mm for  levator muscle function (95% CI, 0.91 - 0.98), with ICCs of 0.93 and 0.95, respectively. 
• Compared with a 2D deep-learning baseline model, DRP had a higher AUC for ptosis diagnosis (0.97 vs 0.89; AUC difference, 0.07; 95% CI, 0.06 - 0.09) and surgical-approach classification (0.98 vs 0.89; AUC difference, 0.09; 95% CI, 0.08-0.11).
• Across subgroups,  the measurement error was within 1 mm in 85.9% of eyes for MRD1 and 63.1% of eyes for LF, with maximum intergroup differences of 9.7% and 10.0%, respectively, across sex, age, institution, and etiology categories.

IN PRACTICE

"Current evidence supports DRP as an adjunct for standardized quantitative assessment in primary care, community screening, home monitoring after ptosis diagnosis, remote follow-up, and triage settings lacking specialist measurements," the researchers reported. 

SOURCE

The study was led by Haotian Lin, MD, PhD, of the Zhongshan Ophthalmic Center at Sun Yat-sen University in Guangzhou, China, and Feng Xu, PhD, of Tsinghua University in Beijing. It was published online September 24 in JAMA Ophthalmology.

LIMITATIONS

Postoperative recurrence rates were evaluated at 3 months without a randomized comparison to standard care and were pooled across diverse ptosis subtypes. Specialist reference measurements carried interobserver variability. Video capture was harder in children and older adults, and limited ethnic diversity in the cohort constrains generalizability.

DISCLOSURES

The study authors reported a patent pending for a method, system, device, and medium for processing eye image data based on 3D reconstruction. No other disclosures were reported.

This article was created using several editorial tools, including AI, as part of the process. Human editors reviewed this content before publication.