International journal of surgery (London, England) · 2025
Interpretable machine learning model for predicting early recurrence of pancreatic cancer: integrating intratumoral and peritumoral radiomics with body composition.
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.
Wu Linxia, Cen Chunyuan, Ouyang Die, Zhang Licai, Li Xin, Wu Heshui, He Ming, Han Ping, Tan Wei, Chen Lei, Zheng Chuansheng
Аннотация
BACKGROUND: Pancreatic ductal adenocarcinoma (PDAC) is associated with a high rate of early recurrence (ER) after radical resection, which significantly affects long-term survival. Currently, no reliable system exists for predicting ER in these patients. This study aimed to develop a machine learning (ML) model combining intratumoral and peritumoral radiomic features with body composition to predict the ER risk in patients with PDAC following radical resection.
MATERIALS AND METHODS: This study included patients with PDAC who underwent upfront surgery at four hospitals between June 2014 and December 2023. Preoperative clinical information, computed tomography (CT) images, and postoperative pathological data were collected. CT-quantified body composition was measured; radiomic features were extracted from the intratumoral and peritumoral regions. Six ML algorithms were used to develop predictive models, including radiomics, clinical, clinical-radiomics, and clinicopathological-radiomics models. The SHapley Additive exPlanations (SHAP) method was applied for model interpretability.
RESULTS: A total of 589 patients were evaluated, including 320 patients (mean age: 60.4 ± 8.3 years; 191 men) in the training cohort, 138 patients (mean age: 60.7 ± 8.9 years; 84 men) in the internal validation cohort, and 131 patients (mean age: 61.7 ± 10.9 years; 76 men) in the external validation cohort. The intra-peri-radiomics model, based on the random forest algorithm, achieved the best performance, with AUCs of 0.865, 0.849, and 0.839 in the training, internal validation, and external validation cohorts, respectively. Incorporating clinicopathological factors, the combined model showed superior performance, with AUCs of 0.936, 0.899, and 0.884 in the training, internal validation, and external validation cohorts, respectively. SHAP analysis revealed that radiomic features, adjuvant therapy, CA199, lymphovascular invasion, platelet-lymphocyte ratio, visceral fat index, CA125, visceral-subcutaneous fat tissue ratio, tumor size, and TNM stage significantly contributed to the prediction of ER.
CONCLUSION: The developed ML model, integrating radiomic features and clinicopathological factors, offered superior predictive accuracy for ER in patients with PDAC post-surgery. SHAP visualization enhanced the model's interpretability and facilitated clinical applications.
MATERIALS AND METHODS: This study included patients with PDAC who underwent upfront surgery at four hospitals between June 2014 and December 2023. Preoperative clinical information, computed tomography (CT) images, and postoperative pathological data were collected. CT-quantified body composition was measured; radiomic features were extracted from the intratumoral and peritumoral regions. Six ML algorithms were used to develop predictive models, including radiomics, clinical, clinical-radiomics, and clinicopathological-radiomics models. The SHapley Additive exPlanations (SHAP) method was applied for model interpretability.
RESULTS: A total of 589 patients were evaluated, including 320 patients (mean age: 60.4 ± 8.3 years; 191 men) in the training cohort, 138 patients (mean age: 60.7 ± 8.9 years; 84 men) in the internal validation cohort, and 131 patients (mean age: 61.7 ± 10.9 years; 76 men) in the external validation cohort. The intra-peri-radiomics model, based on the random forest algorithm, achieved the best performance, with AUCs of 0.865, 0.849, and 0.839 in the training, internal validation, and external validation cohorts, respectively. Incorporating clinicopathological factors, the combined model showed superior performance, with AUCs of 0.936, 0.899, and 0.884 in the training, internal validation, and external validation cohorts, respectively. SHAP analysis revealed that radiomic features, adjuvant therapy, CA199, lymphovascular invasion, platelet-lymphocyte ratio, visceral fat index, CA125, visceral-subcutaneous fat tissue ratio, tumor size, and TNM stage significantly contributed to the prediction of ER.
CONCLUSION: The developed ML model, integrating radiomic features and clinicopathological factors, offered superior predictive accuracy for ER in patients with PDAC post-surgery. SHAP visualization enhanced the model's interpretability and facilitated clinical applications.