Quantitative imaging in medicine and surgery · 2025
Machine learning models for predicting transstenotic pressure gradient based on computed tomography angiography features.
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.
Huang Yan, Han Xu, Qiu Xiaoyu, Dai Chihang, Cai Linkun, An Zhiyuan, Zhang Hui, Wang Guo-Peng, Gong Shusheng, Jin Long, Zhang Xue, Sui Binbin, Yang Zhenghan, Zhao Pengfei, Wang Zhenchang, Ding Heyu
Аннотация
BACKGROUND: The transstenotic pressure gradient (TPG) is closely associated with the pathogenesis, diagnosis, and treatment strategies of conditions such as idiopathic intracranial hypertension and pulsatile tinnitus. Currently, TPG assessment relies on invasive, complex, and costly digital subtraction angiography (DSA), and an efficient, noninvasive evaluation method remains lacking. We thus aimed to develop a machine learning model to predict the transverse sinus TPG using features derived from computed tomography angiography (CTA).
METHODS: We included 139 patients who underwent DSA for transverse sinus TPG measurement. Six feature indicators were extracted from their CTA data. Using TPG as the dependent variable, we applied six machine learning algorithms-logistic regression, adaptive boosting, k-nearest neighbor, naïve Bayes, light gradient boosting machine, and support vector machine-to construct classification models based on TPG thresholds of 4 and 8 mmHg via fivefold cross-validation. Model performance was assessed through use of accuracy, sensitivity, specificity, F1-score, Matthews correlation coefficient, and area under the curve (AUC). Shapley additive explanations analysis was used to interpret feature importance. Ultimately, 19 patients were included as an external validation dataset to assess model accuracy.
RESULTS: The six selected CTA features were residual area ratio, stenosis length, stenosis type, Labbé vein location, degree of drainage dominance, and contralateral stenosis. Logistic regression demonstrated the best performance at both thresholds, with AUC of 0.83 for 4 mmHg and 0.83 for 8 mmHg. Shapley additive explanations analysis revealed a positive correlation between the location of the Labbé vein and TPG, whereas the residual area ratio and stenosis type were negatively correlated. External validation showed good accuracy, reaching 0.89 for both thresholds.
CONCLUSIONS: The developed machine learning model shows promising potential for the noninvasive prediction of transverse sinus TPG based on CTA-derived features.
METHODS: We included 139 patients who underwent DSA for transverse sinus TPG measurement. Six feature indicators were extracted from their CTA data. Using TPG as the dependent variable, we applied six machine learning algorithms-logistic regression, adaptive boosting, k-nearest neighbor, naïve Bayes, light gradient boosting machine, and support vector machine-to construct classification models based on TPG thresholds of 4 and 8 mmHg via fivefold cross-validation. Model performance was assessed through use of accuracy, sensitivity, specificity, F1-score, Matthews correlation coefficient, and area under the curve (AUC). Shapley additive explanations analysis was used to interpret feature importance. Ultimately, 19 patients were included as an external validation dataset to assess model accuracy.
RESULTS: The six selected CTA features were residual area ratio, stenosis length, stenosis type, Labbé vein location, degree of drainage dominance, and contralateral stenosis. Logistic regression demonstrated the best performance at both thresholds, with AUC of 0.83 for 4 mmHg and 0.83 for 8 mmHg. Shapley additive explanations analysis revealed a positive correlation between the location of the Labbé vein and TPG, whereas the residual area ratio and stenosis type were negatively correlated. External validation showed good accuracy, reaching 0.89 for both thresholds.
CONCLUSIONS: The developed machine learning model shows promising potential for the noninvasive prediction of transverse sinus TPG based on CTA-derived features.