Wang Bo, Han Xiaomeng, Zhang Zaixian, Song Hongzheng, Song Yaolin, Liu Ruiqing, Li Zhiming, Liu Shunli
Department of Radiology, The Affiliated Hospital of Qingdao University, Qingdao, Shandong Province, China (B.W., X.H., Z.Z., Z.L., S.L.).
Department of Radiology, Qingdao Municipal Hospital, Shandong Province, Qingdao, Shandong Province, China (H.S.).
Acad Radiol. 2025 May;32(5):2618-2629. doi: 10.1016/j.acra.2024.11.068. Epub 2024 Dec 27.
To develop and validate a radiomics signature, utilizing baseline and restaging CT, for preoperatively predicting progression-free survival (PFS) after neoadjuvant chemotherapy (NAC) in locally advanced gastric cancer (LAGC).
A total of 316 patients with LAGC who received NAC followed by gastrectomy were retrospectively included in this single-center study; these patients were split into two cohorts, one for training (n = 243) and the other for validation (n = 73), based on the different districts of our hospital. A total of 1316 radiomics features were extracted from the volume of interest of the gastric-cancer lesion on venous phase CT images. Four radiomics signatures were built for predicting PFS based on baseline CT (Pre-Rad), restaging CT (Post-Rad), delta radiomics (Delta-Rad) and multi-time radiomics (PrePost-Rad), respectively. Then the PrePost-Rad was combined with clinical factors to establish a nomogram (Rad-clinical model). Kaplan-Meier survival curves with log-rank tests were used to assess the prognostic usefulness of the Rad-clinical model.
All baseline characteristics were not statistically different between the two cohorts. The PrePost-Rad achieved improved predictive value by a C-index of 0.724 (95% CI: 0.639-0.809) in the validation cohort [Pre-Rad: 0.715 (0.632-0.798); Post-Rad: 0.632 (0.538-0.725), Delta-Rad: 0.549 (0.447-0.651)]. In terms of clinical benefit, calibration capability, and prediction efficacy, the Rad-clinical model performed well for PFS prediction, with a C-index of 0.754 (95% CI: 0.707-0.800) and 0.719 (95% CI: 0.639-0.800) in the training and validation cohorts, respectively, superior to the clinical model (cN stage and CA199) but comparable to the PrePost-Rad. Moreover, the Rad-clinical model could accurately classify gastric-cancer patients after NAC into three PFS risk groups in both training and validation cohorts. The risk stratification also performed well in most subgroups (good responders, poor responders, ypTNM Ⅱ, and ypTNM Ⅲ/Ⅳ).
The Rad-clinical model integrating longitudinal radiomics score and clinical factors performed well in preoperatively predicting PFS of LAGC patients after NAC and surgery.
利用基线和再分期CT开发并验证一种影像组学特征，用于术前预测局部晚期胃癌（LAGC）新辅助化疗（NAC）后的无进展生存期（PFS）。
本单中心研究回顾性纳入了316例接受NAC后行胃切除术的LAGC患者；根据我院不同院区，将这些患者分为两个队列，一个用于训练（n = 243），另一个用于验证（n = 73）。从静脉期CT图像上胃癌病灶的感兴趣区提取了总共1316个影像组学特征。分别基于基线CT（Pre-Rad）、再分期CT（Post-Rad）、影像组学差值（Delta-Rad）和多期影像组学（PrePost-Rad）构建了四个用于预测PFS的影像组学特征。然后将PrePost-Rad与临床因素相结合，建立了列线图（Rad临床模型）。采用Kaplan-Meier生存曲线和对数秩检验来评估Rad临床模型的预后价值。
两个队列之间所有基线特征均无统计学差异。在验证队列中，PrePost-Rad的C指数为0.724（95%CI：0.639 - 0.809），预测价值有所提高[Pre-Rad：0.715（0.632 - 0.798）；Post-Rad：0.632（0.538 - 0.725），Delta-Rad：0.549（0.447 - 0.651）]。在临床获益、校准能力和预测效能方面，Rad临床模型在PFS预测方面表现良好，在训练队列和验证队列中的C指数分别为0.754（95%CI：0.707 - 0.800）和0.719（95%CI：0.639 - 0.800），优于临床模型（cN分期和CA199），但与PrePost-Rad相当。此外，Rad临床模型在训练队列和验证队列中均可将NAC后的胃癌患者准确分为三个PFS风险组。风险分层在大多数亚组（良好反应者、不良反应者、ypTNMⅡ和ypTNMⅢ/Ⅳ）中也表现良好。
整合纵向影像组学评分和临床因素的Rad临床模型在术前预测LAGC患者NAC及手术后的PFS方面表现良好。
