Department of Nuclear Medicine, The Affiliated Hospital of Qingdao University, No. 59 Hair Road, Qingdao, Shandong, China.
Department of Gastrointestinal Surgery, The Affiliated Hospital of Qingdao University, Qingdao, Shandong, China.
Abdom Radiol (NY). 2022 Apr;47(4):1244-1254. doi: 10.1007/s00261-022-03453-0. Epub 2022 Feb 26.
Perineural invasion (PNI) has been recognized as an important prognosis factor in patients with colorectal cancer (CRC). The purpose of this retrospective study was to investigate the value of F-FDG PET/CT-based radiomics integrating clinical information, PET/CT features, and metabolic parameters for preoperatively predicting PNI and outcome in non-metastatic CRC and establish an easy-to-use nomogram.
A total of 131 patients with non-metastatic CRC who undergo PET/CT scan were retrospectively enrolled. Univariate analysis was used to compare the differences between PNI-present and PNI-absent groups. Multivariate logistic regression was performed to select the independent predictors for PNI status. Akaike information criterion (AIC) was used to select the best prediction models for PNI status. CT radiomics signatures (RSs) and PET-RSs were selected by maximum relevance minimum redundancy (mRMR) and the least absolute shrinkage and selection operator algorithm (LASSO) regression and radiomics scores (Rad-scores) were calculated for each patient. The prediction models with or without Rad-score were established. According to the nomogram, nomogram scores (Nomo-scores) were calculated for each patient. The performance of different models was assessed with the area under the curve (AUC), specificity, and sensitivity. The clinical usefulness was assessed by decision curve (DCA). Multivariate Cox regression was used to selected independent predictors of progression-free survival (PFS).
Among all the clinical information, PET/CT features, and metabolic parameters, CEA, lymph node metastatic on PET/CT (N stage), and total lesion glycolysis (TLG) were independent predictors for PNI (p < 0.05). Six CT-RSs and 12 PET-RSs were selected as the most valuable factors to predict PNI. The Rad-score calculated with these RSs was significantly different between PNI-present and PNI-absent groups (p < 0.001). The AUC of the constructed model was 0.90 (95%CI: 0.83-0.97) in the training cohort and 0.80 (95%CI: 0.65-0.95) in the test cohort. The nomogram's predicting sensitivity was 0.84 and the specificity was 0.83 in the training cohort. The clinical model's predicting sensitivity and specificity were 0.66 and 0.85 in the training cohort, respectively. Besides, DCA showed that patients with non-metastatic CRC could get more benefit with our model. The results also indicated that N stage, PNI status, and the Nomo-score were independent predictors of PFS in patients with non-metastatic CRC.
The nomogram, integrating clinical data, PET/CT features, metabolic parameters, and radiomics, performs well in predicting PNI status and is associated with the outcome in patients with non-metastatic CRC.
神经周围侵犯（PNI）已被认为是结直肠癌（CRC）患者的一个重要预后因素。本回顾性研究的目的是探讨基于 F-FDG PET/CT 的放射组学结合临床信息、PET/CT 特征和代谢参数，能否在术前预测非转移性 CRC 患者的 PNI 状态和预后，并建立一个易于使用的列线图。
共纳入 131 例接受 PET/CT 扫描的非转移性 CRC 患者。采用单因素分析比较 PNI 阳性组和 PNI 阴性组之间的差异。采用多变量逻辑回归分析筛选 PNI 状态的独立预测因子。采用赤池信息量准则（AIC）选择用于预测 PNI 状态的最佳预测模型。采用最大相关性最小冗余（mRMR）和最小绝对收缩和选择算子算法（LASSO）回归选择 CT 放射组学特征（RSs）和 PET-RSs，并计算每位患者的放射组学评分（Rad-score）。建立有无 Rad-score 的预测模型。根据列线图，为每位患者计算列线图评分（Nomo-score）。采用曲线下面积（AUC）、特异性和敏感性评估不同模型的性能。通过决策曲线（DCA）评估临床实用性。采用多变量 Cox 回归分析筛选无进展生存（PFS）的独立预测因子。
在所有临床信息、PET/CT 特征和代谢参数中，CEA、PET/CT 上的淋巴结转移（N 期）和总病变糖酵解（TLG）是 PNI 的独立预测因子（p<0.05）。选择 6 个 CT-RSs 和 12 个 PET-RSs 作为预测 PNI 的最有价值因素。这些 RSs 计算的 Rad-score 在 PNI 阳性组和 PNI 阴性组之间差异有统计学意义（p<0.001）。在训练队列中，构建模型的 AUC 为 0.90（95%CI：0.83-0.97），在测试队列中为 0.80（95%CI：0.65-0.95）。在训练队列中，列线图的预测敏感性为 0.84，特异性为 0.83。在训练队列中，临床模型的预测敏感性和特异性分别为 0.66 和 0.85。此外，DCA 表明，非转移性 CRC 患者可从我们的模型中获得更多获益。结果还表明，N 期、PNI 状态和 Nomo-score 是非转移性 CRC 患者 PFS 的独立预测因子。
该列线图整合了临床数据、PET/CT 特征、代谢参数和放射组学，可较好地预测 PNI 状态，并与非转移性 CRC 患者的预后相关。
