Frontiers in medicine · 2026
Topological feature-driven TabPFN model for prediction of enlarged hemorrhage and edema after tumor resection in meningiomas.
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.
Han Wenjing, Tan Guirong, Li Lijia, Feng Zhenyang, Zhou Chen, Liu Xiang, Hu Lingjing
Аннотация
INTRODUCTION: Enlarged hemorrhage and edema after tumor resection (EHETR) is a serious postoperative complication in meningiomas. The peritumoral edema (PE) region was reported to be associated with EHETR. Topological data analysis (TDA) has recently emerged as a novel approach providing a multiscale characterization of structural organization. This study aims to assess the feasibility of using topological features extracted from the PE region to predict EHETR.
METHODS: We retrospectively enrolled 161 patients with meningiomas, of whom 79 (49.1%) developed EHETR. Multiscale topological features were extracted from the PE regions on preoperative MRI sequences, including contrast-enhanced T1-weighted imaging (T1CE), T2-weighted imaging (T2WI), and apparent diffusion coefficient (ADC) maps, using cubical persistent homology. Feature selection was performed using the least absolute shrinkage and selection operator (LASSO) within a nested five-fold cross-validation framework (5-fold outer loop and 3-fold inner loop). Subsequently, predictive models were constructed using the Tabular Prior-data Fitted Network (TabPFN). Model performance was evaluated using receiver operating characteristic (ROC) analysis, calibration curves, and decision curve analysis. Model interpretability was further assessed using SHapley Additive exPlanations (SHAP) to quantify feature contributions.
RESULTS: The ADC-based topological model exhibited superior discriminative performance, achieving a mean area under the receiver operating characteristic curve (AUC) of 0.80 (95% CI: 0.71-0.89) in the validation set, compared with models based on T1CE (AUC: 0.74) and T2WI (AUC: 0.70). DeLong tests further confirmed that the ADC-based model significantly outperformed models based on T1CE and T2WI (DeLong, P < 0.001). SHAP analysis highlighted persistence landscape features, particularly those from the H1 (loops) and H2 (cavities) homology dimensions as the primary drivers of prediction, suggesting that higher-order topological features from the PE region may be key contributors to EHETR.
CONCLUSION: The topological features derived from the PE region can predict EHETR in patients with meningiomas as a novel computational imaging framework.
METHODS: We retrospectively enrolled 161 patients with meningiomas, of whom 79 (49.1%) developed EHETR. Multiscale topological features were extracted from the PE regions on preoperative MRI sequences, including contrast-enhanced T1-weighted imaging (T1CE), T2-weighted imaging (T2WI), and apparent diffusion coefficient (ADC) maps, using cubical persistent homology. Feature selection was performed using the least absolute shrinkage and selection operator (LASSO) within a nested five-fold cross-validation framework (5-fold outer loop and 3-fold inner loop). Subsequently, predictive models were constructed using the Tabular Prior-data Fitted Network (TabPFN). Model performance was evaluated using receiver operating characteristic (ROC) analysis, calibration curves, and decision curve analysis. Model interpretability was further assessed using SHapley Additive exPlanations (SHAP) to quantify feature contributions.
RESULTS: The ADC-based topological model exhibited superior discriminative performance, achieving a mean area under the receiver operating characteristic curve (AUC) of 0.80 (95% CI: 0.71-0.89) in the validation set, compared with models based on T1CE (AUC: 0.74) and T2WI (AUC: 0.70). DeLong tests further confirmed that the ADC-based model significantly outperformed models based on T1CE and T2WI (DeLong, P < 0.001). SHAP analysis highlighted persistence landscape features, particularly those from the H1 (loops) and H2 (cavities) homology dimensions as the primary drivers of prediction, suggesting that higher-order topological features from the PE region may be key contributors to EHETR.
CONCLUSION: The topological features derived from the PE region can predict EHETR in patients with meningiomas as a novel computational imaging framework.