Prediction of recurrence risk in soft tissue sarcomas by MRI and digital pathology based omics nomogram
王童语 1王鹤翔 1赵心迪 1侯峰 2杨江飞 3侯明妤 4万光耀 1岳斌 5郝大鹏 1胡凌
扫码查看
点击上方二维码区域，可以放大扫码查看
作者信息
1. 青岛大学附属医院放射科，青岛 266003
2. 青岛大学附属医院病理科，青岛 266003
3. 山东第一医科大学附属省立医院医学影像科，济南 250021
4. 青岛大学附属妇女儿童医院病理科，青岛 266034
5. 青岛大学附属医院骨肿瘤科，青岛 266003
折叠
摘要
目的 探讨基于MRI影像及数字病理图像的组学列线图预测软组织肉瘤（STS）术后复发风险的价值。 方法 本研究为回顾性队列研究，回顾性收集2016年1月至2021年3月青岛大学附属医院经手术病理证实的192例STS患者，其中于崂山院区就诊的患者作为训练集（112例），市南院区就诊的患者作为验证集（80例）。对患者进行随访，分为复发组（87例）和未复发组（105例）。收集患者的临床和影像学特征，提取病灶脂肪抑制T2WI图像的影像组学和数字病理图像的病理组学特征，采用多因素Cox回归建立临床模型、影像组学模型、病理组学模型和联合组学模型，并结合最优组学模型和临床模型，构建组学列线图。采用一致性指数（C index）和时间依赖受试者操作特征曲线下面积（t-AUC）评价各模型预测STS术后复发风险的效能，采用DeLong检验比较t-AUC间的差异。采用X-tile软件确定组学列线图的截断值，将患者分为低风险（106例）、中风险（64例）及高风险（22例）组，采用Kaplan-Meier生存曲线和log-rank检验计算并比较3个复发风险组的累积无复发生存（RFS）率。 结果 联合组学模型的性能优于单一影像组学或病理组学模型，在验证集中的C index为0.727（95%CI 0.632~0.823）、中位t-AUC为0.737（95%CI0.584~0.891）。结合临床模型和联合组学模型构建组学列线图，在验证集中的C index为0.763（95%CI 0.685~0.842），中位t-AUC为0.783（95%CI0.639~0.927）。在验证集中，组学列线图的t-AUC值高于临床模型、TNM模型、影像组学模型及病理组学模型，差异有统计学意义（Z=3.33、2.18、2.08、2.72，P=0.001、0.029、0.037、0.007）；组学列线图与联合组学模型的t-AUC值差异无统计学意义（Z=0.70，P=0.487）。在验证集中，低、中、高复发风险组STS患者术后1年RFS率为92.0%（95%CI 81.5%~100%）、55.9%（95%CI 40.8%~76.6%）、37.5%（95%CI 15.3%~91.7%）。在训练集和验证集中，低、中、高复发风险组STS患者的术后累积RFS率差异有统计学意义（训练集χ²=73.90，P<0.001；验证集χ²=18.70，P<0.001）。 结论 基于MRI影像和数字病理图像的组学列线图对STS术后复发风险具有较好的预测性能。 Objective To investigate the value of an MRI and digital pathology images based omics nomogram for the prediction of recurrence risk in soft tissue sarcoma (STS). Methods This was a retrospective cohort study. From January 2016 to March 2021, 192 patients with STS confirmed by pathology in the Affiliated Hospital of Qingdao University were enrolled, among which 112 patients in the Laoshan campus were enrolled as training set, and 80 patients in the Shinan campus were enrolled as validation set. The patients were divided into recurrence group (n=87) and no recurrence group (n=105) during follow-up. The clinical and MRI features of patients were collected. The radiomics features based on fat saturated T2WI images and pathomics features based on digital pathology images of the lesions were extracted respectively. The clinical model, radiomics model, pathomics model, radiomics-pathomics combined model, and omics nomogram which combined the optimal prediction model and the clinical model were established by multivariate Cox regression analysis. The concordance index (C index) and time-dependent area under the receiver operating characteristic curve (t-AUC) were used to evaluate the performance of each model in predicting STS postoperative recurrence. The DeLong test was used for comparison of t-AUC between every two models. The X-tile software was used to determine the cut-off value of the omics nomogram, then the patients were divided into low risk (n=106), medium risk (n=64), and high risk (n=22) groups. Three groups′ cumulative recurrence-free survival (RFS) rates were calculated and compared by the Kaplan-Meier survival curve and log-rank test. Results The performance of the radiomics-pathomics combined model was superior to the radiomics model and pathomics model, with C index of 0.727 (95%CI 0.632-0.823) and medium t-AUC value of 0.737 (95%CI0.584-0.891) in the validation set. The omics nomogram was established by combining the clinical model and the radiomics-pathomics combined model, with C index of 0.763 (95%CI 0.685-0.842) and medium t-AUC value of 0.783 (95%CI0.639-0.927) in the validation set. The t-AUC value of omics nomogram was significantly higher than that of clinical model, TNM model, radiomics model, and pathomics model in the validation set (Z=3.33, 2.18, 2.08, 2.72, P=0.001, 0.029, 0.037, 0.007). There was no statistical difference in t-AUC between the omics nomogram and radiomics-pathomics combined model (Z=0.70, P=0.487). In the validation set, the 1-year RFS rates of STS patients in the low, medium, and high recurrence risk groups were 92.0% (95%CI 81.5%-100%), 55.9% (95%CI 40.8%-76.6%), and 37.5% (95%CI 15.3%-91.7%). In the training and validation sets, there were statistically significant in cumulative RFS rates among the low, medium, and high groups of STS patients (training set χ²=73.90, P<0.001 validation setχ²=18.70, P<0.001). Conclusion The omics nomogram based on MRI and digital pathology images has favorable performance for the prediction of STS recurrence risk.
Abstract
Objective To investigate the value of an MRI and digital pathology images based omics nomogram for the prediction of recurrence risk in soft tissue sarcoma (STS). Methods This was a retrospective cohort study. From January 2016 to March 2021, 192 patients with STS confirmed by pathology in the Affiliated Hospital of Qingdao University were enrolled, among which 112 patients in the Laoshan campus were enrolled as training set, and 80 patients in the Shinan campus were enrolled as validation set. The patients were divided into recurrence group (n=87) and no recurrence group (n=105) during follow-up. The clinical and MRI features of patients were collected. The radiomics features based on fat saturated T2WI images and pathomics features based on digital pathology images of the lesions were extracted respectively. The clinical model, radiomics model, pathomics model, radiomics-pathomics combined model, and omics nomogram which combined the optimal prediction model and the clinical model were established by multivariate Cox regression analysis. The concordance index (C index) and time-dependent area under the receiver operating characteristic curve (t-AUC) were used to evaluate the performance of each model in predicting STS postoperative recurrence. The DeLong test was used for comparison of t-AUC between every two models. The X-tile software was used to determine the cut-off value of the omics nomogram, then the patients were divided into low risk (n=106), medium risk (n=64), and high risk (n=22) groups. Three groups′ cumulative recurrence-free survival (RFS) rates were calculated and compared by the Kaplan-Meier survival curve and log-rank test. Results The performance of the radiomics-pathomics combined model was superior to the radiomics model and pathomics model, with C index of 0.727 (95%CI 0.632-0.823) and medium t-AUC value of 0.737 (95%CI0.584-0.891) in the validation set. The omics nomogram was established by combining the clinical model and the radiomics-pathomics combined model, with C index of 0.763 (95%CI 0.685-0.842) and medium t-AUC value of 0.783 (95%CI0.639-0.927) in the validation set. The t-AUC value of omics nomogram was significantly higher than that of clinical model, TNM model, radiomics model, and pathomics model in the validation set (Z=3.33, 2.18, 2.08, 2.72, P=0.001, 0.029, 0.037, 0.007). There was no statistical difference in t-AUC between the omics nomogram and radiomics-pathomics combined model (Z=0.70, P=0.487). In the validation set, the 1-year RFS rates of STS patients in the low, medium, and high recurrence risk groups were 92.0% (95%CI 81.5%-100%), 55.9% (95%CI 40.8%-76.6%), and 37.5% (95%CI 15.3%-91.7%). In the training and validation sets, there were statistically significant in cumulative RFS rates among the low, medium, and high groups of STS patients (training set χ²=73.90, P<0.001 validation setχ²=18.70, P<0.001). Conclusion The omics nomogram based on MRI and digital pathology images has favorable performance for the prediction of STS recurrence risk.
万方数据