Three-dimensional markerless surface topography approach with convolutional neural networks for adolescent idiopathic scoliosis screening
Bibliographic record
Abstract
Adolescent idiopathic scoliosis (AIS) is a three-dimensional lateral and torsional deformity of the spine, affecting up to 5% of the population. Traditional scoliosis screening methods exhibit limited accuracy, leading to unnecessary referrals and exposure to ionizing radiation from x-ray examinations. The 3D markerless surface topography (ST) technique quantifies trunk asymmetry and can be a potential scoliosis screening tool. However, differences in trunk asymmetry between individuals with scoliosis and those with a typically developing spine have yet to be thoroughly studied. Using the ST method, this study aims to distinguish adolescents with AIS from those with typically a developing spine. Participants aged 10 to 18 years, comprising of 285 individuals with confirmed AIS and 273 with typically developing spines, were included in the study (total scans including follow-ups: 693 for the AIS group and 298 for the control group). The positive for AIS group was identified through radiographic exams, specifically with curves ranging from 10° to 45°, while the negative (control) group qualified if their scoliometer test measured less than 7° and they had no known scoliosis diagnosis. The dataset comprised of surface torso scans captured either using stationary Minolta cameras or with the Structure sensor. ST analysis involved the reflection of the 3D geometry of the torso, aligning it with the original torso by minimizing the distance between corresponding points. Deviations between the original and reflected torso over the back surface and torso surface depth were mapped onto 102 × 102 grids. A convolutional neural network (CNN) was developed using deviations and depth (distance between the back surface and frontal plane) maps as inputs to classify the torso surface of typically developing adolescents and those with AIS. 10-fold cross-validation was applied during model development. 20% of the data was used as a holdout for final testing. Classification results of the proposed model were compared to the ground truth. The average training and validation accuracy across the ten folds was 100% and 94%, respectively. The classifications from the testing sets using the best performing model from the 10-fold cross-validation obtained accuracy, sensitivity, and specificity of 95%, 97%, and 90%, respectively. The positive likelihood ratio (PLR) of the testing set was 9.7. Likewise, a negative likelihood ratio (NLR) of 0.032 was also attained. The model sensitivity for detecting curves with Cobb greater than 25° was 99%. The sensitivity for detecting mild cases (Cobb < 25°) was 96%. The proposed CNN predictive model to detect AIS using ST showed excellent classification results. Markerless surface topography can serve as a dependable and non-invasive method for screening AIS.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Full frame machine prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. The Gemma side is a direct model label for every work in the frame, read from the title-only record. The Codex side is a classifier learned from the 10,348 direct Codex labels and calibrated to design-weighted sample rates; fields without enough sample support carry no Codex call. Candidate is the union of the two sides; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels.
Distilled classifier scores by category (both heads)
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.000 | 0.001 |
| Meta-epidemiology (narrow) | 0.001 | 0.000 |
| Meta-epidemiology (broad) | 0.001 | 0.001 |
| Bibliometrics | 0.001 | 0.001 |
| Science and technology studies | 0.000 | 0.000 |
| Scholarly communication | 0.001 | 0.001 |
| Open science | 0.001 | 0.001 |
| Research integrity | 0.001 | 0.001 |
| Insufficient payload (model declined to judge) | 0.002 | 0.000 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedMachine predicted; a candidate call from one source (direct Gemma or distilled Codex), not a consensus.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".