Frontiers in oncology · 2026
Dual-region deep learning model integrating prostate and peri-prostatic adipose tissue MRI features for bone metastasis prediction in prostate cancer.
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.
Qin Peng, Liu Bohao, Su Huabin, Wang Ziqiao, Lin Zhengxu, Zhu Weian, Zou Chen, Cai Qian, Wu Jianjie, Zheng Jiayu, Zhao Xiao, Wen Xuan, Luo Yun
Аннотация
RATIONALE AND OBJECTIVES: Bone metastasis (BM) is pivotal in prostate cancer (PCa) management. This study developed a multimodal model integrating MRI-derived deep features from the prostate gland (PG) and periprostatic adipose tissue (PPAT) with clinical variables for BM risk assessment.
METHODS: Retrospectively, 464 patients were recruited and randomly divided into training and internal test cohorts at a ratio of 7:3. Deep features were extracted from PG and PPAT regions on T2-weighted MRI using a pretrained ResNet-50 as a fixed feature extractor. Clinical, PG, and PPAT component models were developed using patient-level cross-validation. Their out-of-fold probabilities were integrated by a logistic-regression meta-learner to construct the Deep feature-based PG-PPAT-Clinical (DPPC) model. Model performance was evaluated using ROC-AUC, average precision, calibration analysis, decision-curve analysis, and SHAP analysis.
RESULTS: The DPPC model achieved ROC-AUCs of 0.922 (95% CI, 0.887-0.954) in the training cohort and 0.928 (95% CI, 0.864-0.978) in the internal test cohort. Its ROC-AUC was significantly higher than that of the Clinical model in the training cohort and the PPAT model in the internal test cohort, whereas the remaining pairwise differences were not statistically significant. At a probability threshold of 0.5, the internal-test sensitivity and specificity were 65.1% and 95.9%, respectively. The observed BM rates were 87.5% in the high-risk group and 13.9% in the low-risk group.
CONCLUSION: By synergizing deep learning signatures from PG and PPAT with clinical factors, the DPPC model demonstrates promising performance for BM risk stratification, and external validation and further calibration assessment are required.
METHODS: Retrospectively, 464 patients were recruited and randomly divided into training and internal test cohorts at a ratio of 7:3. Deep features were extracted from PG and PPAT regions on T2-weighted MRI using a pretrained ResNet-50 as a fixed feature extractor. Clinical, PG, and PPAT component models were developed using patient-level cross-validation. Their out-of-fold probabilities were integrated by a logistic-regression meta-learner to construct the Deep feature-based PG-PPAT-Clinical (DPPC) model. Model performance was evaluated using ROC-AUC, average precision, calibration analysis, decision-curve analysis, and SHAP analysis.
RESULTS: The DPPC model achieved ROC-AUCs of 0.922 (95% CI, 0.887-0.954) in the training cohort and 0.928 (95% CI, 0.864-0.978) in the internal test cohort. Its ROC-AUC was significantly higher than that of the Clinical model in the training cohort and the PPAT model in the internal test cohort, whereas the remaining pairwise differences were not statistically significant. At a probability threshold of 0.5, the internal-test sensitivity and specificity were 65.1% and 95.9%, respectively. The observed BM rates were 87.5% in the high-risk group and 13.9% in the low-risk group.
CONCLUSION: By synergizing deep learning signatures from PG and PPAT with clinical factors, the DPPC model demonstrates promising performance for BM risk stratification, and external validation and further calibration assessment are required.