第一部分：基于MRI盆腔脂肪体积分析对于局部进展期直肠癌新辅助放化疗后远处转移的预测价值

中文摘要

目的：本研究旨在探讨基于MRI盆腔脂肪体积分析对于局部进展期直肠癌（Locally advanced rectal cancer, LARC）新辅助放化疗（Neoadjuvant chemoradiotherapy, nCRT）后远处转移的预测价值，并构建LARC患者nCRT后无远处转移生存期（Distant metastasis-free survival, DMFS）的预测模型。

材料与方法：回顾性分析了2012年3月至2017年6月在我院接受nCRT和手术的236例LARC患者，中位随访时间为85个月（四分位距：73–100个月）。在轴位T1加权成像图像上分别测量直肠系膜脂肪体积（Mesorectal fat volume, MFV）、内脏脂肪体积（Visceral fat volume, VFV）、皮下脂肪体积 （Subcutaneous fat volume, SFV）、直肠系膜脂肪体积与内脏脂肪体积之比（Ratio of mesorectal fat volume to visceral fat volume, MFV/VFV）及内脏脂肪体积与皮下脂肪体积之比（Ratio of visceral fat volume to subcutaneous fat volume, VFV/SFV）等数据。采用Maximally selected rank statistics方法确定上述参数的最佳阈值。采用Kaplan-Meier方法及Log-rank检验评价不同组间的DMFS差异。采用单因素及多因素Cox回归分析、风险比（Hazard Ratio, HR）及95%置信区间（Confidence interval, CI）确定与DMFS相关的独立预测因子并构建列线图预测模型。通过一致性指数（Concordance index, C-index）、时间依赖性受试者工作特征曲线（Receiver operating characteristic curve, ROC）及曲线下面积（Area under the curve, AUC）评估列线图对于DMFS的预测性能。

结果： MFV ≤ 90.29 cm3及MFV/VFV ≤ 0.176的LARC患者具有更差的DMFS，经多因素Cox回归分析，MFV/VFV是DMFS的独立预测因子。MFV/VFV在男性及女性间具有显著差异（p < 0.001）。在性别亚组分析中，男性及女性MFV/VFV的最佳阈值分别为0.173及0.125，无论是男性还是女性，低MFV/VFV相较于高MFV/VFV的患者都具有更差的DMFS。联合MFV/VFV（HR = 0.394 [95% CI, 0.213–0.727], p = 0.003）、糖类抗原19-9（Carbohydrate antigen, CA19-9）（HR = 1.856 [95% CI, 1.044–3.301], p = 0.035）、临床N分期（HR = 1.900 [95% CI, 1.074–3.364], p = 0.028）及MRI识别的壁外静脉侵犯（MRI identified extramural venous invasion, mr-EMVI）（HR = 2.050 [95% CI, 1.241–3.387], p = 0.005）等特征构建了预测LARC患者nCRT后DMFS的列线图，该列线图的C-index为0.727, 预测1年、3年及5年DMFS的时间依赖性ROC的AUC分别为0.737，0.749，0.730。

结论：LARC患者的MFV及MFV/VFV越小预示着DMFS越差。MFV/VFV、CA19-9、临床N分期及mr-EMVI是DMFS的独立预测因子，联合上述4个特征构建了预测LARC患者nCRT后DMFS的列线图，该列线图可以有效地预测患者发生远处转移的可能性。

第二部分：基于MRI直肠系膜多区域影像组学分析对于局部进展期直肠癌新辅助放化疗后远处转移的预测价值

中文摘要

目的：本研究旨在探讨基于磁共振成像（Magnetic resonance imaging, MRI）直肠系膜影像组学特征对于局部进展期直肠癌（Locally advanced rectal cancer, LARC）新辅助放化疗（Neoadjuvant chemoradiotherapy, nCRT）后远处转移的预测价值，并构建LARC患者nCRT后无远处转移生存期（Distant metastasis-free survival, DMFS）的预测模型。

材料与方法：回顾性分析了2011年6月至2017年6月在我院接受nCRT和手术的230例LARC患者，以7:3的比例随机分为训练组（161例）及验证组（69例）。中位随访时间为86个月（四分位距：73–101个月）。在T2加权成像（T2-weighted imaging, T2WI）图像上手动分割肿瘤总体积（Gross tumor volume, GTV），自GTV向周围的直肠系膜扩张并分割瘤周体积（Peritumoral volume，PTV），向外扩张距离分别为4 mm、6 mm和 8 mm（PTV₄、PTV₆ 和 PTV₈），图像分割后提取GTV、PTV₄、PTV₆及PTV₈的影像组学特征。采用最小绝对缩小和选择算子（Least absolute shrinkage and selection operator, LASSO）-Cox分析进行影像组学特征选择及影像组学模型构建。采用单因素及多因素Cox回归分析、风险比（Hazard Ratio, HR）及95%置信区间（Confidence interval, CI）确定与DMFS相关的独立预测因子并构建列线图预测模型。通过验证组数据对各个模型的预测性能进行验证。采用一致性指数（Concordance index，C-index）、时间依赖性受试者操作特征曲线（Receiver operating characteristic curve，ROC）及曲线下面积（Area under the curve，AUC）作为每个模型对于DMFS的预测性能的评价指标。

结果：与PTV₆和PTV₈影像组学模型相比，PTV₄影像组学模型表现出更好的预测性能，PTV₄影像组学模型在训练组和验证组中的C-index分别为0.750和0.703。通过多因素Cox回归分析，联合GTV影像组学模型（HR = 2.061 [95% CI, 1.487–2.855], p < 0.001）、PTV₄影像组学模型（HR = 2.206 [95% CI, 1.566–3.107], p < 0.001）、临床N分期（HR = 2.379 [95% CI, 1.164–4.864], p = 0.018）及MRI识别的壁外静脉侵犯（mr-EMVI）（HR = 1.930 [95% CI, 1.086–3.432], p = 0.025）共同构建了一个融合模型，并将该模型可视化为一个列线图。在训练组和验证组中，列线图的C-index分别为0.833和0.756，5年DMFS的AUC分别为0.873和0.809。以1.613作为列线图最佳阈值进行风险分层，可以将患者分为远处转移的高风险组和低风险组。Kaplan-Meier生存分析显示，训练组及验证组的高风险组患者的DMFS均显著差于低风险组患者。

结论：直肠系膜影像组学特征包含了与LARC患者远处转移相关的重要预后信息。临床N分期、mr-EMVI、GTV影像组学模型及PTV₄影像组学模型是DMFS的独立预测因子。上述4个特征构建的列线图可以准确预测LARC患者nCRT后的DMFS，能够帮助临床医生在治疗前筛选高风险患者并制定个体化的新辅助治疗方案。

Part 1: MRI-based Volume Analysis of Pelvic Fat for Predicting Distant Metastasis After Neoadjuvant Chemoradiotherapy in Patients with Locally Advanced Rectal Cancer

ABSTRACT

Objective: This study aimed to investigate the potential of MRI-based pelvic fat volume for predicting distant metastasis and construct a risk prediction model for distant metastasis-free survival (DMFS) after neoadjuvant chemoradiotherapy (nCRT) in patients with locally advanced rectal cancer (LARC).

Materials and methods: A total of 236 LARC patients who underwent nCRT and surgery at our institution from March 2012 to June 2017 were retrospectively analyzed, with a median follow-up of 85 months (interquartile range: 73–100 months). Mesorectal fat volume (MFV), visceral fat volume (VFV), subcutaneous fat volume (SFV), the ratio of mesorectal fat volume to visceral fat volume (MFV/VFV), and the ratio of visceral fat volume to subcutaneous fat volume (VFV/SFV) were measured on axial T1-weighted imaging images. Maximally selected rank statistics were used to determine the cut-off values for the above features. The Kaplan-Meier method and Log-rank test were used to evaluate the differences in DMFS between different groups. Univariate and multivariate Cox regression analyses, hazard ratio (HR), and 95% confidence interval (CI) were used to determine the independent predictors associated with DMFS and to construct a nomogram. The predictive performance of DMFS was evaluated by the concordance index (C-index), time-dependent Receiver operating characteristic curve (ROC), and area under the curve (AUC).

Results: LARC patients with MFV ≤ 90.29 cm3 and MFV/VFV ≤ 0.176 had worse DMFS, and MFV/VFV was an independent predictor of DMFS after multivariate Cox regression analysis. MFV/VFV was significantly different between men and women (p < 0.001). In the gender subgroup analysis, the cut-off values for MFV/VFV were 0.173 and 0.125 for males and females, respectively, and both males and females had worse DMFS with low MFV/VFV compared with those with high MFV/VFV. The nomogram integrating MFV/VFV (HR = 0.394 [95% CI, 0.213–0.727], p = 0.003), carbohydrate antigen 19-9 (CA19-9) (HR = 1.856 [95% CI, 1.044–3.301], p = 0.035), clinical N stage (HR = 1.900 [95% CI, 1.0074–3.364] p = 0.028), and MRI identified extramural venous invasion (mr-EMVI) (HR = 2.050 [95% CI 1.241–3.387], p = 0.005) was constructed to predict DMFS after nCRT in LARC. The nomogram achieved a C-index of 0.727, with corresponding AUCs of 0.737, 0.749, and 0.730 for predicting 1-year, 3-year, and 5-year DMFS, respectively.

Conclusion: Smaller MFV and MFV/VFV implied worse DMFS in patients with LARC. CA19-9, clinical N-stage, mr-EMVI, and MFV/VFV were independent predictors of DMFS. The nomogram combined the above four features showed good performance for predicting DMFS after nCRT in patients with LARC.

Part 2: MRI-based Multiregional Radiomics in Mesorectum for Predicting Distant Metastasis After Neoadjuvant Chemoradiotherapy in Patients with Locally Advanced Rectal Cancer

ABSTRACT

Objective: This study aimed to construct an MRI-based mesorectal radiomics model to predict distant metastasis-free survival (DMFS) after neoadjuvant chemoradiotherapy (nCRT) in locally advanced rectal cancer (LARC).

Materials and Methods: 230 LARC patients who underwent nCRT and surgery at our institution from June 2011 to June 2017 were retrospectively analyzed and randomly divided into a training group (161 patients) and a validation group (69 patients) in a 7:3 ratio. The median follow-up time was 86 months (interquartile range: 73–101 months). The gross tumor volume (GTV) was manually segmented on T2-weighted imaging (T2WI) images, and the peritumoral volume (PTV) was segmented by dilating the GTV toward the mesorectum by 4 mm, 6 mm, and 8 mm (PTV₄, PTV₆ , and PTV₈). Least absolute shrinkage and selection operator (LASSO)-Cox analysis was used for the radiomics features selection and radiomics model construction. Univariate and multivariate Cox regression analyses, hazard ratio (HR) and 95% confidence interval (CI) were used to determine the independent predictors associated with DMFS and to construct a nomogram. The predictive performance of each model was validated by the validation group. The predictive performance of each model for DMFS was assessed by the concordance index (C-index), time-dependent receiver operating characteristic curve (ROC), and area under the curve (AUC).

Results: The PTV₄ radiomics model demonstrated superior performance compared to the PTV₆ and PTV₈ radiomics models, with C-indexes of 0.750 and 0.703 in the training and validation cohorts, respectively. The nomogram integrating GTV radiomics model (HR = 2.061 [95% CI, 1.487–2.855], p < 0.001), PTV₄ radiomics model (HR = 2.206 [95% CI, 1.566–3.107], p < 0.001), clinical N stage (HR = 2.379 [95% CI, 1.164–4.864], p = 0.01), and MRI identified extramural venous invasion (mr-EMVI) (HR = 1.930 [95% CI, 1.086–3.432], p = 0.025) was constructed to predict DMFS after nCRT in LARC. The nomogram achieved C-indexes of 0.833 and 0.756, with corresponding AUCs of 0.873 and 0.809 for 5-year DMFS in the training and validation cohorts, respectively. The cut-off value of 1.613 could stratify patients into high- and low-risk groups for DM. Kaplan-Meier analysis revealed the DMFS outcomes of patients in the high-risk group were significantly worse than those in the low-risk group in the training and validation cohort.

Conclusion: Mesorectal radiomics features contain important prognostic information related to DMFS in LARC patients. Clinical N-stage, mr-EMVI, GTV radiomics model, and PTV₄ radiomics model were independent predictors of DMFS. The nomogram combined the above four features showed good performance for predicting DMFS after nCRT in patients with LARC, which can help clinicians to identify high-risk patients before treatment and facilitate personalized neoadjuvant treatment plans.