This prospective diagnostic accuracy study was conducted in accordance with the Declaration of Helsinki and approved by the institutional review board of the First Affiliated Hospital of Anhui Medical University (approval PJ 2025-16-05). Written informed consent was obtained from all participants and their legal guardians prior to enrollment.

The inclusion criteria were as follows: (1) no prior history of scoliosis diagnosis; (2) suspected scoliosis based on positive physical examination findings, parental concern, or referral from school or community screening programs; and (3) ability to maintain a standing posture during imaging examinations.

The exclusion criteria were as follows: (1) inability to tolerate standing examination due to severe orthostatic intolerance or balance disorders, (2) severe skin conditions or open wounds on the back that might affect imaging quality, or (3) any condition judged by the investigator as potentially compromising patient safety or data integrity.

The scoliosis screening system used a custom-developed millimeter-wave scanner operating in the K_a band (29-40 GHz). This frequency range was selected to balance penetration through common clothing materials with spatial resolution sufficient for detailed surface topography mapping [21]. The system integrated frequency-modulated continuous wave radar for precise range resolution [19] and synthetic aperture radar processing via the range migration algorithm [20,22] to generate high-resolution azimuthal imagery. The fusion of these techniques produced a detailed 3D scattering map of the back’s surface and underlying contours.

The reconstructed volume was characterized by a voxel matrix of 192 (width) × 448 (height) × 128 (depth) with spacings of 4.2 mm (horizontal), 3.6 mm (vertical), and 4.0 mm (depth). A maximum intensity projection algorithm was applied for initial visualization and qualitative analysis. The scanning procedure was completed within approximately 2 seconds, and calibration was verified using standardized test boards to ensure imaging accuracy. The image of the calibration test board is shown in Figure 1.

Each participant underwent millimeter-wave scanning in 5 standardized standing positions (as illustrated in Figure 2): natural standing position (posture A), palms externally rotated (posture B), hands behind the head (posture C), and single-arm elevation (postures D and E).

Considering reproducibility and stability, this study used only the natural standing position (posture A) for quantitative analysis, with the remaining 4 postures reserved for qualitative assessment.

Participants were scanned while wearing lightweight clothing (including athletic wear, single-layer garments, or T-shirts) without the need for disrobing, thereby substantially preserving patient privacy and dignity. Metallic items (such as underwire bras) that could potentially interfere with imaging quality were removed when present (Multimedia Appendix 3).

All participants underwent standard standing anteroposterior and lateral full-spine radiography using standardized imaging parameters (85 kV; 20-55 mAs; focus-film distance: 180 cm). Regarding examination sequence, participants first underwent millimeter-wave scanning followed by standard radiographic examination as the definitive diagnostic reference. Both examinations were completed on the same day to minimize potential changes in spinal alignment.

The blinding protocol required millimeter-wave image analysts to be blinded to radiographic results and radiographic assessors to be blinded to millimeter-wave results. Cobb angle measurements were performed independently by 2 experienced orthopedic surgeons, with assessors blinded to each other’s results and to millimeter-wave findings. All image identifiers were removed prior to analysis, and an independent data management system was used to prevent unblinding.

Millimeter-wave images were analyzed using a predefined protocol focusing on 4 clinically relevant morphological indicators [23-29]: shoulder height asymmetry, trunk lateral shift, waistline contour asymmetry, and lower-limb height discrepancy.

Diagnostic performance was assessed using sensitivity, specificity, positive predictive value, negative predictive value, and overall accuracy in accordance with the STARD (Standards for Reporting of Diagnostic Accuracy Studies) guidelines [30]. Continuous variables with normal distribution were expressed as means and SDs and compared between groups using the independent-sample t test (2-tailed), whereas nonnormally distributed data were analyzed using the Mann-Whitney U test. All statistical analyses were performed using SPSS (version 22.0; IBM Corp), with statistical significance defined as a 2-tailed P value of less than .05.

Between July 2024 and December 2025, a total of 165 consecutive patients presenting to the spinal surgery outpatient clinic with suspected scoliosis were screened for eligibility. Of these 165 patients, 19 (11.5%) were excluded for failing to meet the inclusion criteria (eg, not first-time presentation). Of the remaining 146 patients, 14 (9.6%) were subsequently excluded based on the exclusion criteria: 9 (64.3%) for inability to undergo radiographic examination (patient refusal or contraindications) and 5 (35.7%) for excessive perspiration on the back compromising image quality. Ultimately, 132 patients were enrolled and completed the study protocol, constituting the final analysis cohort.

Of the 132 patients enrolled in this study, 98 (74.2%) were confirmed to have scoliosis (Cobb angle≥10°) based on x-ray examination, whereas the remaining 34 (25.8%) had measurements of less than 10° and were classified as the nonscoliotic group. The baseline characteristics of the 2 groups are summarized in Table 1. There were no significant differences in age, height, or weight between the 2 groups (P>.05), ensuring the comparability of the study cohorts.

The comparative analysis between millimeter-wave assessments and the x-ray reference standard yielded the results shown in Table 2. The overall diagnostic accuracy of the millimeter-wave imaging protocol was 86.4% (114/132). The technology demonstrated a high sensitivity of 85.7% (84/98) and a specificity of 88.2% (30/34). The positive predictive value was 95.5%, and the negative predictive value was 68.2%.

With the exception of lower-limb length discrepancy, the remaining 3 morphological parameters demonstrated higher abnormality rates in the scoliosis group than in the nonscoliosis group as detailed in Table 3. Notably, shoulder height asymmetry and waistline contour asymmetry also exhibited relatively elevated abnormality rates in the nonscoliosis group (13/34, 38.2% and 11/34, 32.4%, respectively).

To further evaluate the integrated diagnostic capability of millimeter-wave imaging, a multivariate logistic regression model was developed incorporating the 4 morphological parameters. The receiver operating characteristic curve analysis of this composite model demonstrated superior performance compared to the individual parameters. The multivariate model achieved an area under the curve of 0.862 (95% CI 0.802-0.922), indicating excellent discriminatory ability (as illustrated in Figure 7).

The detailed performance metrics of the model are shown in the confusion matrix (Figure 8). The model exhibited a sensitivity of 92.9% (91/98 true positives) and a specificity of 38.2% (13/34 true negatives), with an overall accuracy of 78.8% (104/132). This indicates that the methodology aligns with the primary goal of screening protocols, which is to prioritize sensitivity to avoid missed cases, thereby underscoring its value in clinical practice.

In a subgroup analysis of the 98 patients with scoliosis, the millimeter-wave imaging system demonstrated varying performance across different curve severity groups. The technology demonstrated high diagnostic accuracy in the moderate curve group (Cobb angle=20°-40°) and severe group (≥40°) compared to 75.4% (43/57) in the mild group (10°-20°).

This diagnostic accuracy study evaluated a custom-developed millimeter-wave imaging system in 132 pediatric outpatients with suspected scoliosis. The system demonstrated a sensitivity of 85.7% (84/98), specificity of 88.2% (30/34), and overall accuracy of 86.4% (114/132) for detecting scoliosis (Cobb angle≥10°). Each scan was completed in approximately 2 seconds, effectively minimizing motion artifacts. Notably, sensitivity was lower for mild curves (Cobb angle=10°-20°) than for moderate or severe curves.

As a non-ionizing modality, millimeter-wave imaging aligns with the “as low as reasonably achievable” principle and may reduce unnecessary radiation exposure in pediatric populations. The rapid acquisition time (approximately 2 seconds) and ability to scan through light clothing balanced efficiency with privacy preservation, supporting application in large-scale, high-throughput settings such as schools or communities. Moreover, the 3D surface data may enable future assessment of parameters such as vertebral rotation, offering morphological information beyond that provided by conventional 2D radiography.

In summary, this technology is suitable for initial scoliosis screening. Key advantages include (1) absence of ionizing radiation, avoiding cumulative cancer risks from repeated radiographic examinations and supporting longitudinal monitoring; (2) ability to image through everyday clothing, preserving patient dignity and addressing privacy concerns relevant to adolescents; (3) rapid scanning, facilitating efficient large-scale screening; and (4) 3D morphological data that complement planar radiographic findings. Current scoliosis screening techniques based on angular measurement primarily include the use of a scoliometer for assessing the angle of trunk rotation (ATR) and optical surface topography systems. Reported sensitivities for these methods typically range from approximately 70% to 86% [31,32]. In comparison, the millimeter-wave imaging system used in this study achieved a sensitivity of 85.7% (84/98), which is comparable to the performance of existing non–radiation‑based screening technologies. The device can be positioned as a triage tool to identify high-risk individuals requiring confirmatory radiographic evaluation. It should be emphasized that this system does not replace Cobb angle measurement, and definitive diagnosis still requires standard radiography.

This study has several limitations that should be acknowledged. First, diagnostic accuracy was significantly lower in the mild scoliosis group (Cobb angle=10°-20°) than in the moderate (20°-40°) and severe (≥40°) groups. This difference arose because, under the current diagnostic criteria, some imaging-confirmed mild scoliosis cases were classified as negative during initial screening when at least one quantitative parameter did not meet the predefined threshold, resulting in false-negative outcomes (Multimedia Appendix 5). Moreover, although multi-posture analysis often identifies compensatory asymmetries indicative of scoliosis (eg, scapular asymmetry), these features are not fully captured in single-posture or single-parameter assessments. To preserve the objectivity of this blinded diagnostic accuracy study, qualitative findings from multi-posture observations were not incorporated into the formal statistical analysis, nor were they used to retrospectively revise the initial binary (positive or negative) classification based on the predefined quantitative thresholds.

Second, this was a diagnostic accuracy study conducted in patients with suspected scoliosis referred for clinical evaluation, not a screening study in an asymptomatic population. Future multicenter studies with larger and more diverse populations are needed to further validate the clinical applicability of this technology.

Third, we did not include ATR measurements for direct comparison with existing clinical screening tools. In addition, we did not perform direct comparison between clothed and unclothed measurements or between millimeter-wave imaging and optical surface topography systems. Future studies should incorporate direct comparison among ATR measurements, optical surface topography systems, and millimeter-wave imaging.

Finally, diagnostic accuracy was lower for mild scoliosis (10°-20°) as subtle asymmetries may not always meet quantitative threshold criteria. Future integration of machine learning algorithms for automated analysis may improve detection of mild curves [32-35].

Millimeter-wave imaging represents a promising advancement in scoliosis screening. It provides a rapid, safe, and privacy-preserving approach, achieving a diagnostic performance comparable to initial clinical assessments while offering greater objectivity. As an effective preliminary screening step prior to radiographic evaluation, this technology has the potential to optimize health care resource use. Overall, it shows substantial promise for enabling large-scale, routine scoliosis screening in a safe and practical manner, consistent with recommendations from leading clinical organizations.