普美显增强MRI预测肝细胞癌消融术后早期复发的影像组学模型研究

Study on the Radiomics Model to Predict Early Recurrence after Hepatocellular Carcinoma Ablation

目的肝癌是我国导致肿瘤相关死亡的第二大主要原因。针对早期肝细胞癌(hepatocellular carcinoma,HCC),局部热消融术的疗效可与外科手术相当,且微创、可重复、并发症发生率低。然而早期复发(2年)是HCC消融术后的主要死亡原因,但目前临床上尚缺乏准确、可靠的早期复发预测模型。因此本研究拟利用普美显增强MRI资料结合影像组学技术,构建HCC消融术后早期复发的预测模型,并对其预测效能进行评价和验证。旨在探讨消融术前普美显增强MRI影像组学在HCC患者预后评估上的应用价值,为临床治疗决策提供可靠的数据和理论依据。方法对120例行消融手术且术前1月内有普美显增强MRI检查的HCC患者进行回顾性分析。利用术前普美显增强MRI的T2加权序列(T2WI)图像,每位患者共提取出1318个影像组学特征。经过特征筛选后共选用6种机器学习算法进行建模和比较,最终运用Logistic回归分析分别建立临床-影像学模型、影像组学模型和综合预测模型并构建Nomogram图,评估模型的区分度、精准度和临床获益价值。结果最终选取5个与早期复发相关的影像组学特征建立普美显增强MRI-T2WI影像组学模型,该模型具有良好的预测性能,在验证组中AUC值为0.80。纳入与早期复发相关的2个临床-影像学危险因素包括肿瘤数量和肝胆期瘤周低信号,进一步建立临床-影像学-影像组学综合预测模型(Clinical-Radiological-Radiomics Model,CRRM模型)。CRRM模型的预测性能更为优越,在验证组中AUC值高达0.92,模型各项指标和整体表现均优于单纯影像组学模型。校正曲线结果显示Nomogram图对预测早期复发风险有较好的一致性。临床决策曲线分析进一步证实CRRM模型的临床应用价值。结论术前普美显增强MRI-T2WI图像特征的影像组学模型对预测早期复发有一定价值。相比之下,CRRM模型拥有更为全面且准确的预测复发性能,能提前预测HCC消融术后复发风险,根据不同的风险联合恰当的靶向或免疫等有效的术前治疗,制定更加切合患者个体情况的围手术期治疗策略,从而提高HCC患者的预后和长期生存率。

Objective Liver cancer is the second leading cause of tumor-related death.The efficacy of local thermal ablation is comparable to surgical resection for the early hepatocellular carcinoma(HCC),and the ablation technique is minimally invasive,repeatable,and has a low complication rate.However,early recurrence(2 years)is the main cause of death after HCC ablation,but there is still a lack of accurate and reliable prediction models for early recurrence.Therefore,this survey intended to construct prediction models for early recurrence of HCC after ablation by using preoperative gadoxetic acid disodium-enhanced magnetic resonance(MR)images data combined with radiomics methods,evaluate and verify their predictive efficacy.To explore the application value of contrast-enhanced MRI imaging before ablation in the prognosis assessment of HCC patients,and to provide reliable data and theoretical basis for clinical treatment decisions.Methods A retrospective study was performed on 120 patients with HCC who underwent ablation and all the patients were underwent contrast-enhanced MRI examination within 1 month.A total of 1318 radiomic features were extracted from each patient by using preoperative T2-weighted sequence(T2WI)images of contrast-enhanced MRI.After feature selection,six machine learning algorithms would be used for construction of models and comparison.Finally,Logistic regression analysis was used to establish a clinical model,a radiomics model and a combined model which included the above risk factors and radiologic features.The nomogram was constructed based the combined model to evaluate the differentiation,accuracy and clinical benefit.Results Five radiomic features most closely related to early recurrence were identified and selected for model construction.The radiomic model had effective predictive performance,with AUC of 0.80 in the training sets.Two clinical risk factors associated with early recurrence,including tumor number and peritumoral hypodensity on the hepatobiliary phase,were selected to established a clinical-radiological-radiomics(CRRM)model,with AUC as high as 0.92 in the validation sets.The nomogram of CRRM model was constructed and the calibration curves indicated the goodness of fit.Decision curve analysis further confirmed the clinical usefulness of CRRM model.Conclusion The radiomics model of preoperatively contrast-enhanced MRI-T2WI image features was identified be effective to predict HCC early recurrence.In contrast,the CRRM model could be used as a more comprehensive and superior tool to predict individual probability of early recurrence.Patients at high risk of early recurrence could be identified and the appropriate and effective preoperative treatments could also be taken,to improve the prognosis and long-term survival rate of HCC patients the individualized treatment strategies should be formulated.