Frontiers in artificial intelligence · 2026
MRI-based morphometric analysis of the patellofemoral joint: diagnostic modeling of knee pathologies in adolescents.
СтатьяНазвание и аннотация: Перевод в очереди
Обработка статьи · этапы, ошибки и управление
Загрузка состояния…
Ошибки и ограничения
История · было → стало
Состояние обновляется каждые 5 секунд, пока блок открыт. Процент показывается только там, где известен объём работы. Пауза останавливает продолжение этапа; текущий запрос к источнику может завершиться.
Spasic Dusan, Djuricic Goran, Kovac Jelena Djokic, Bukva Bojan, Radlovic Vladimir, Maletic Milos, Rajkovic Stanislav, Radulović Marko
Аннотация
INTRODUCTION: To evaluate whether routinely measured MRI-based patellofemoral joint morphometric parameters can support diagnostic modeling of selected adolescent knee pathologies and to compare a conventional multivariable logistic regression baseline with machine-learning approaches.
METHODS: This retrospective single-center pilot diganostic modeling study included 168 adolescents (97 girls, 71 boys, mean age 15.5 ± 1.7 years) who underwent knee MRI between January 2018 and December 2024 because of anterior knee pain or suspected patellofemoral structural abnormality. Thirteen patellofemoral morphometric parameters were measured by two radiologists. Three binary MRI endpoints were modeled: composite chondromalacia, a composite endpoint of ACL injury or patellar bone bruise, and patellar retinacular lesion. Baseline multivariable logistic regression was compared with machine-learning approaches using chronological training, validation, and test splits. Additional gradient-boosting comparators were evaluated, and uncertainty was quantified using bootstrap confidence intervals on the independent test set.
RESULTS: In multivariable logistic regression, no individual continuous morphometric predictor reached statistical significance for any endpoint. After correction of the model-selection procedure, predictive performance proved endpoint-specific rather than uniformly strong. For composite chondromalacia, discrimination remained weak. For the ACL injury/patellar bone bruise composite, machine-learning models showed only modest improvement in point estimates over logistic regression, but confidence intervals for differences crossed zero. The strongest reproducible signal was observed for patellar retinacular lesions. After correction of model selection and bootstrap-based uncertainty estimation, a morphometric-only CatBoost model achieved an AUC of 0.85 (95% CI 0.68-0.97) and balanced accuracy of 0.79 (95% CI 0.61-0.94), while a morphometric-only LightGBM model achieved an AUC of 0.84 (95% CI 0.64-0.97) and balanced accuracy of 0.76 (95% CI 0.59-0.88).
DISCUSSION: In adolescents, routine patellofemoral morphometrics do not provide uniformly strong diagnostic discrimination across all MRI-defined knee pathologies studied here. Their most convincing predictive value was observed for patellar retinacular lesions, whereas performance for composite chondromalacia and the ACL injury/patellar bone bruise composite remained limited. These findings support a narrower interpretation of clinical utility and justify further validation in larger external cohorts.
METHODS: This retrospective single-center pilot diganostic modeling study included 168 adolescents (97 girls, 71 boys, mean age 15.5 ± 1.7 years) who underwent knee MRI between January 2018 and December 2024 because of anterior knee pain or suspected patellofemoral structural abnormality. Thirteen patellofemoral morphometric parameters were measured by two radiologists. Three binary MRI endpoints were modeled: composite chondromalacia, a composite endpoint of ACL injury or patellar bone bruise, and patellar retinacular lesion. Baseline multivariable logistic regression was compared with machine-learning approaches using chronological training, validation, and test splits. Additional gradient-boosting comparators were evaluated, and uncertainty was quantified using bootstrap confidence intervals on the independent test set.
RESULTS: In multivariable logistic regression, no individual continuous morphometric predictor reached statistical significance for any endpoint. After correction of the model-selection procedure, predictive performance proved endpoint-specific rather than uniformly strong. For composite chondromalacia, discrimination remained weak. For the ACL injury/patellar bone bruise composite, machine-learning models showed only modest improvement in point estimates over logistic regression, but confidence intervals for differences crossed zero. The strongest reproducible signal was observed for patellar retinacular lesions. After correction of model selection and bootstrap-based uncertainty estimation, a morphometric-only CatBoost model achieved an AUC of 0.85 (95% CI 0.68-0.97) and balanced accuracy of 0.79 (95% CI 0.61-0.94), while a morphometric-only LightGBM model achieved an AUC of 0.84 (95% CI 0.64-0.97) and balanced accuracy of 0.76 (95% CI 0.59-0.88).
DISCUSSION: In adolescents, routine patellofemoral morphometrics do not provide uniformly strong diagnostic discrimination across all MRI-defined knee pathologies studied here. Their most convincing predictive value was observed for patellar retinacular lesions, whereas performance for composite chondromalacia and the ACL injury/patellar bone bruise composite remained limited. These findings support a narrower interpretation of clinical utility and justify further validation in larger external cohorts.