Pediatric radiology · 2026
Diagnostic value of machine-learning using conventional magnetic resonance imaging markers for pediatric idiopathic intracranial hypertension: a retrospective study.
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.
Cuce Ferhat, Tulum Gokalp, Işık Muhammet Ikbal, Baştemur Mehmet, Arslan Mutluay, Ünay Bülent, Osman Onur
Аннотация
BACKGROUND: Pediatric idiopathic intracranial hypertension can be challenging to diagnose; magnetic resonance imaging (MRI) signs are considered supportive, while lumbar puncture remains the diagnostic reference.
OBJECTIVE: We evaluated whether a compact set of conventional magnetic resonance imaging markers, analyzed with supervised machine-learning, can assist in the diagnosis of idiopathic intracranial hypertension in children presenting with headache.
MATERIALS AND METHODS: In this retrospective single-center study, 62 pediatric patients with idiopathic intracranial hypertension and 62 headache controls without papilledema or structural pathology were analyzed. All lumbar puncture procedures in the idiopathic intracranial hypertension group were performed under benzodiazepine sedation. Twenty-four demographic and quantitative MRI features (optic nerve sheath and globe metrics, pituitary measurements, venous sinus findings, and posterior fossa measures) were extracted. Six classifiers (random forest, support vector machine, multilayer perceptron, XGBoost, k-nearest neighbors, and Bagging) were trained using repeated nested cross-validation with Bayesian hyperparameter optimization; performance was summarized on held-out folds.
RESULTS: Across models, accuracies were approximately 0.70-0.74, sensitivities approximately 0.64-0.72, and specificities approximately 0.69-0.84; among non-support vector machine models, the area under the receiver operating characteristic curve was approximately 0.80-0.82. Feature selection consistently retained anatomically plausible markers, led by optic nerve sheath measurements, posterior globe flattening, pituitary/sella configuration, and venous sinus abnormalities. The resulting models generate uncalibrated predicted probabilities for decision support rather than binary diagnostic labels. Performance was estimated across repeated validation splits.
CONCLUSION: Machine-learning applied to routinely obtainable MRI markers yields moderate diagnostic performance for pediatric idiopathic intracranial hypertension. Given that lumbar puncture remains the gold standard, these models are intended as decision-support tools to complement clinical assessment.
OBJECTIVE: We evaluated whether a compact set of conventional magnetic resonance imaging markers, analyzed with supervised machine-learning, can assist in the diagnosis of idiopathic intracranial hypertension in children presenting with headache.
MATERIALS AND METHODS: In this retrospective single-center study, 62 pediatric patients with idiopathic intracranial hypertension and 62 headache controls without papilledema or structural pathology were analyzed. All lumbar puncture procedures in the idiopathic intracranial hypertension group were performed under benzodiazepine sedation. Twenty-four demographic and quantitative MRI features (optic nerve sheath and globe metrics, pituitary measurements, venous sinus findings, and posterior fossa measures) were extracted. Six classifiers (random forest, support vector machine, multilayer perceptron, XGBoost, k-nearest neighbors, and Bagging) were trained using repeated nested cross-validation with Bayesian hyperparameter optimization; performance was summarized on held-out folds.
RESULTS: Across models, accuracies were approximately 0.70-0.74, sensitivities approximately 0.64-0.72, and specificities approximately 0.69-0.84; among non-support vector machine models, the area under the receiver operating characteristic curve was approximately 0.80-0.82. Feature selection consistently retained anatomically plausible markers, led by optic nerve sheath measurements, posterior globe flattening, pituitary/sella configuration, and venous sinus abnormalities. The resulting models generate uncalibrated predicted probabilities for decision support rather than binary diagnostic labels. Performance was estimated across repeated validation splits.
CONCLUSION: Machine-learning applied to routinely obtainable MRI markers yields moderate diagnostic performance for pediatric idiopathic intracranial hypertension. Given that lumbar puncture remains the gold standard, these models are intended as decision-support tools to complement clinical assessment.