Neuroradiology · 2025
Development of a machine learning model to predict changes in neuroimaging profiles among acute ischemic stroke patients following delayed transfer for endovascular thrombectomy.
СтатьяНазвание и аннотация: Перевод пока не подготовлен
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.
Chen Huanwen, Skorseth Paige, Rewinkel Scott, Kim Daniel, Amin Sonesh, Shakal Scott, Priest Ryan, Nesbit Gary, Clark Wayne, Colasurdo Marco
Аннотация
INTRODUCTION: Endovascular thrombectomy (EVT) patient selection depends on neuroimaging. However, interhospital transfer delays can lead to neuroimaging changes, whether and when repeat imaging is necessary are unclear. Herein, we develop a machine learning model (MLM) to predict vessel recanalization, ischemia progression, and imaging stability for EVT candidates who experience delayed interhospital transfer.
METHODS: This retrospective study included EVT candidates with internal carotid or middle cerebral artery occlusion stroke transferred 1.5-6.0 h after initial imaging. Clinical and radiographic data were collected. A gradient-boosted tree-based MLM (XGBoost) was trained and optimized on 66% of the cohort (randomly selected) using 10-fold cross-validation, and the MLM was independently validated on the remaining, untouched 33% of the study cohort. Model performance was assessed using areas under the receiver operating characteristics curve (AUC) for discrimination, F1 scores for precision/recall, and Brier scores for calibration.
RESULTS: Among 317 patients, 69.4% had stable imaging, 14.5% showed ischemia progression (ASPECTS drop ≥ 2), and 16.1% had vessel recanalization. The MLM was developed and optimized in the training cohort (n = 212). NIH stroke scale improvement, onset-to-imaging time, intravenous thrombolysis, initial ASPECTS, and collateral score were important features. In the validation cohort (n = 105), the MLM achieved AUCs of 0.81 (95%CI 0.72-0.90) for imaging stability, 0.82 (95%CI 0.72-0.91) for ischemia progression, and 0.89 (95%CI 0.77-1.00) for vessel recanalization. F1 scores were 0.87 and 0.95 for stability and no recanalization, with Brier scores of 0.17 and 0.08, respectively.
CONCLUSION: Our MLM accurately predicts imaging changes among EVT candidates who experienced transfer delays.
METHODS: This retrospective study included EVT candidates with internal carotid or middle cerebral artery occlusion stroke transferred 1.5-6.0 h after initial imaging. Clinical and radiographic data were collected. A gradient-boosted tree-based MLM (XGBoost) was trained and optimized on 66% of the cohort (randomly selected) using 10-fold cross-validation, and the MLM was independently validated on the remaining, untouched 33% of the study cohort. Model performance was assessed using areas under the receiver operating characteristics curve (AUC) for discrimination, F1 scores for precision/recall, and Brier scores for calibration.
RESULTS: Among 317 patients, 69.4% had stable imaging, 14.5% showed ischemia progression (ASPECTS drop ≥ 2), and 16.1% had vessel recanalization. The MLM was developed and optimized in the training cohort (n = 212). NIH stroke scale improvement, onset-to-imaging time, intravenous thrombolysis, initial ASPECTS, and collateral score were important features. In the validation cohort (n = 105), the MLM achieved AUCs of 0.81 (95%CI 0.72-0.90) for imaging stability, 0.82 (95%CI 0.72-0.91) for ischemia progression, and 0.89 (95%CI 0.77-1.00) for vessel recanalization. F1 scores were 0.87 and 0.95 for stability and no recanalization, with Brier scores of 0.17 and 0.08, respectively.
CONCLUSION: Our MLM accurately predicts imaging changes among EVT candidates who experienced transfer delays.