Cancer imaging : the official publication of the International Cancer Imaging Society · 2026
Habitat-based imaging and peritumoral radiomics on ultrasound images for predicting lymphovascular invasion in breast invasive ductal carcinoma: a two-center study.
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.
Liu Taixia, Zhao Guojia, Wei Wei, Zhang Qingling, Wu Jing, Chen Xuying, Liu Dong, Zhu Xiangming
Аннотация
PURPOSE: This study aimed to evaluate the feasibility of employing habitat-based radiomic distributions in ultrasound (US) images to quantitatively characterize intratumoral heterogeneity. It also explored the potential of this approach to predict lymphovascular invasion (LVI) in breast invasive ductal carcinoma (IDC) patients and to identify the optimal extent of multiple peritumoral regions.
METHODS: A total of 408 women diagnosed with IDC from January 2020 to October 2023 were enrolled in this retrospective cohort study from two medical centers. Intratumoral areas were partitioned into four distinct habitat areas using K-means cluster analysis, while peritumoral regions were expanded at increments of 2, 4, and 6 mm. Radiomic features were independently extracted from the intra- and peri-tumoral areas, and habitat subregions for developing predictive models. These models incorporated three machine learning classifiers: Random Forest, Extreme Gradient Boosting (XGBoost), and Support Vector Machine (SVM), respectively. We subsequently established an integrated model encompassing intra- and peri-tumoral areas, and habitat radiomic features, and clinicopathological factors. The model performance was assessed through receiver operating characteristic (ROC), calibration curves, and decision curve analysis (DCA). Finally, SHapley Additive exPlanations (SHAP) and nonogram were applied to enhance model interpretability.
RESULTS: The Random Forest model exhibited superior performance in terms of the area under the curve (AUC) values of 0.861 (95% CI: 0.809-0.912), 0.832 (95% CI: 0.746-0.919), and 0.810 (95% CI: 0.684-0.935) for the training, validation, and test sets, separately. Additionally, the peri-2 mm model surpassed the performance of the other models (peri-4 mm, peri-6 mm) in LVI prediction. The integrated model, encompassing peri-2 mm features, clinicopathological factors, and habitat models, achieved robust predictive performance with AUC values of 0.940 (95% CI: 0.907-0.973), 0.924 (95% CI: 0.875-0.973), and 0.852 (95% CI: 0.732-0.972) for each respective set.
CONCLUSION: The integrated model yields the improved predictive performance in predicting LVI status, and the model offer a reliable and feasible preoperative prediction method to enhance the clinical management and therapeutic planning for IDC patients.
METHODS: A total of 408 women diagnosed with IDC from January 2020 to October 2023 were enrolled in this retrospective cohort study from two medical centers. Intratumoral areas were partitioned into four distinct habitat areas using K-means cluster analysis, while peritumoral regions were expanded at increments of 2, 4, and 6 mm. Radiomic features were independently extracted from the intra- and peri-tumoral areas, and habitat subregions for developing predictive models. These models incorporated three machine learning classifiers: Random Forest, Extreme Gradient Boosting (XGBoost), and Support Vector Machine (SVM), respectively. We subsequently established an integrated model encompassing intra- and peri-tumoral areas, and habitat radiomic features, and clinicopathological factors. The model performance was assessed through receiver operating characteristic (ROC), calibration curves, and decision curve analysis (DCA). Finally, SHapley Additive exPlanations (SHAP) and nonogram were applied to enhance model interpretability.
RESULTS: The Random Forest model exhibited superior performance in terms of the area under the curve (AUC) values of 0.861 (95% CI: 0.809-0.912), 0.832 (95% CI: 0.746-0.919), and 0.810 (95% CI: 0.684-0.935) for the training, validation, and test sets, separately. Additionally, the peri-2 mm model surpassed the performance of the other models (peri-4 mm, peri-6 mm) in LVI prediction. The integrated model, encompassing peri-2 mm features, clinicopathological factors, and habitat models, achieved robust predictive performance with AUC values of 0.940 (95% CI: 0.907-0.973), 0.924 (95% CI: 0.875-0.973), and 0.852 (95% CI: 0.732-0.972) for each respective set.
CONCLUSION: The integrated model yields the improved predictive performance in predicting LVI status, and the model offer a reliable and feasible preoperative prediction method to enhance the clinical management and therapeutic planning for IDC patients.