Deep learning radiomics based on contrast enhanced computed tomography predicts microvascular invasion and survival outcome in early stage hepatocellular carcinoma

ObjectiveTo evaluate the performance of a deep learning (DL)-based radiomics strategy on contrast-enhanced computed tomography (CT) to predict microvascular invasion (MVI) status and clinical outcomes, recurrence-free survival (RFS) and overall survival (OS) in patients with early stage hepatocellular carcinoma (HCC) receiving surgical resection.MethodsAll 283 eligible patients were included retrospectively between January 2008 and December 2015, and assigned into the training cohort (n = 198) and the testing cohort (n = 85). We extracted radiomics features via handcrafted radiomics analysis manually and DL analysis of pretrained convolutional neural networks via transfer learning automatically. Support vector machine was adopted as the classifier. A clinical-radiological model for MVI status integrated significant clinical features and the radiological signature generated from the radiological model with the optimal area under the receiver operating characteristics curve (AUC) in the testing cohort. Otherwise, DL-based prognostic models were constructed in prediction of recurrence and mortality via Cox proportional hazard analysis.ResultsThe clinical-radiological model for MVI represented an AUC of 0.909, accuracy of 96.47%, sensitivity of 90.91%, specificity of 97.30%, positive predictive value of 83.33%, and negative predictive value of 98.63% in the testing cohort. The clinical-radiological models for identification of RFS and OS outperformed prediction performance of the clinical model or the DL signature alone. The DL-based integrated model for prognostication showed great predictive value with significant classification and discrimination abilities after validation.ConclusionsThe integrated DL-based radiomics models achieved accurate preoperative prediction of MVI status, and might facilitate predicting tumor recurrence and mortality in order to optimize clinical decisions for patients with early stage HCC.     

Comparing Radiomics features of tumour and healthy liver tissue in a limited CT dataset: A machine learning study

IntroductionLiver cancer lesions on Computed Tomography (CT) withholds a great amount of data, which is not visible to the radiologists and radiographer. Radiomics features can be extracted from the lesions and used to train Machine Learning (ML) algorithms to predict between tumour and liver tissue. The purpose of this study was to investigate and classify Radiomics features extracted from liver tumours and normal liver tissue in a limited CT dataset.MethodsThe Liver Tumour Segmentation Benchmark (LiTS) dataset consisting of 131 CT scans of the liver with segmentations of tumour tissue and healthy liver was used to extract Radiomic features. Extracted Radiomic features included size, shape, and location extracted with morphological and statistical techniques according to the International Symposium on Biomedical Imaging manual. Relevant features was selected with chi2 correlation and principal component analysis (PCA) with tumour and healthy liver tissue as outcome according to a consensus between three experienced radiologists. Logistic regression, random forest and support vector machine was used to train and validate the dataset with a 10-fold cross-validation method and the Grid Search as hyper-parameter tuning. Performance was evaluated with sensitivity, specificity and accuracy.ResultsThe performance of the ML algorithms achieved sensitivities, specificities and accuracy ranging from 96.30% (95% CI: 81.03%–99.91%) to 100.00% (95% CI: 86.77%–100.00%), 91.30% (95% CI: 71.96%–98.93%) to 100.00% (95% CI: 83.89%–100.00%)and 94.00% (95% CI: 83.45%–98.75%) to 100.00% (95% CI: 92.45%–100.00%), respectively.ConclusionML algorithms classifies Radiomics features extracted from healthy liver and tumour tissue with perfect accuracy. The Radiomics signature allows for a prognostic biomarker for hepatic tumour screening on liver CT.Implications for practiceDifferentiation between tumour and liver tissue with Radiomics ML algorithms have the potential to increase the diagnostic accuracy, assist in the decision-making of supplementary multiphasic enhanced medical imaging, as well as for developing novel prognostic biomarkers for liver cancer patients.     

MRI Radiomic Analysis of IMRT-Induced Bladder Wall Changes in Prostate Cancer Patients: A Relationship with Radiation Dose and Toxicity

BackgroundThe main purpose of this study was to assess the structural changes in the bladder wall of prostate cancer patients treated with intensity-modulated radiation therapy using magnetic resonance imaging texture features analysis and to correlate image texture changes with radiation dose and urinary toxicity.MethodsEthical clearance was granted to enroll 33 patients into this study who were treated with intensity-modulated radiation therapy for prostate cancer. All patients underwent two magnetic resonance imagings before and after radiation therapy (RT). A total of 274 radiomic features were extracted from MR-T2W–weighted images. Wilcoxon singed rank-test was performed to assess significance of the change in mean radiomic features post-RT relative to pre-RT values. The relationship between radiation dose and feature changes was assessed and depicted. Cystitis was recorded as urinary toxicity. Area under receiver operating characteristic curve of a logistic regression–based classifier was used to find correlation between radiomic features with significant changes and radiation toxicity.ResultsThirty-three bladder walls were analyzed, with 11 patients developing grade ≥2 urinary toxicity. We showed that radiomic features may predict radiation toxicity and features including S5.0SumVarnc, S2.2SumVarnc, S1.0AngScMom, S0.4SumAverg, and S5. _5InvDfMom with area under receiver operating characteristic curve 0.75, 0.69, 0.65, 0.63, and 0.62 had highest correlation with toxicity, respectively. The results showed that most of the radiomic features were changed with radiation dose.ConclusionFeature changes have a good correlation with radiation dose and radiation-induced urinary toxicity. These radiomic features can be identified as being potentially important imaging biomarkers and also assessing mechanisms of radiation-induced bladder injuries.     

CBCT影像组学联合构建Nomogram模型预测食管癌放疗患者放射性肺炎

目的 通过对放疗疗程中不同时段CBCT图像的影像组学分析,寻找早期定量预测食管癌放疗放射性肺炎(RP)的参数,结合临床特征和肺剂量体积参数建立联合Nomogram模型并探讨这一模型对食管癌RP的预测价值。方法 回顾分析2017—2019年间临床资料、剂量学参数、CBCT图像资料完整的 96例胸中段食管鳞癌调强放疗患者资料,每例患者均分别获取放疗期间3个不同时段的肺CBCT图像。全组病例随机分成训练集(67例)和验证集(29例),以CBCT上双肺实质作为感兴趣区,运用3D-Slicer软件进行图像分割和特征提取,经LASSO-Logistics回归分析方法进行特征参数筛选并构建影像组学标签(Rad-score)。从3个不同时段建立的RP预测模型中选择最优模型联合经回归分析获得的最佳临床及剂量学参数,建立联合Nomogram模型,并进行受试者工作特征曲线分析,基于曲线下的面积(AUC)验证其诊断效能。结果 第一时段的影像组学预测模型优于其他两个时段,在训练集中的AUC值为0.700(95%CI为 0.568~0.832),敏感性和特异性分别为61.5%、75.0%;在验证集中的AUC值为0.765(95%CI为 0.588~0.941),敏感性和特异性分别为84.6%、64.7%。影像组学联合临床及剂量学构建的Nomogram模型在训练集中的AUC值为0.836(95%CI为 0.700~0.918),敏感性和特异性分别为96.0%、54.8%;在验证集中的AUC值为0.905(95%CI为 0.799~1.000),敏感性和特异性分别为92.9%、73.3%。联合Nomogram模型诊断效能最佳。结论 基于放疗早期肺CBCT影像组学特征构建的模型,对于食管癌RP具有一定的预测效能,Rad-score联合 肺V5Gy、肺 Dmean、肿瘤分期建立的Nomogram模型具有更好的预测准确性,可作为一种定量预测模型用于RP的预测。     

应用MR影像组学量化放疗前后脑白质变化的研究

目的 研究脑肿瘤放疗前后脑白质MR影像组学特征变化,分析影像组学特征变化与放疗剂量的关系,为放射性脑白质损伤的早期预测及监测提供参考方法。方法 选取2018年9月至2020年7月在山东省肿瘤医院接受放疗的脑肿瘤患者70例,分别获取患者模拟定位CT及MR图像（T1平扫、T1增强以及T1增强到T1平扫的剪影图像T1剪影）,患者放疗23～50 Gy后再次获取MR图像。根据患者实际照射剂量,将接受0～5 Gy、5～10 Gy、10～15 Gy、15～20 Gy、20～30 Gy、30～40 Gy及 > 40 Gy不同剂量梯度的脑白质定义为感兴趣区域（ROI）。提取不同ROI在T1平扫、T1增强以及T1剪影三个序列图像中的影像组学特征,比较放疗前后每个ROI MR影像组学特征的差异,分析其与放射剂量变化的关系。结果 每个ROI在每套图像上提取93个影像组学特征。在T1平扫、T1增强以及T1剪影图像中放疗前后差异具有统计学意义的特征数分别为,0～5 Gy：52、52、7。5～10 Gy：1、1、9。10～15 Gy：0、16、28。15～20 Gy：15、8、2。20～30 Gy：1、77、25。30～40 Gy：38、64、29。> 40 Gy：32、47、6。放疗前后变化率超过±50%的特征中,在0～5 Gy、5～10 Gy、10～15 Gy、15～20 Gy、20～30 Gy、30～40 Gy、> 40 Gy剂量梯度下不同MR序列影像组学特征最大变化率分别达到：T1平扫164.06%、1.39%、无、35.76%、7.4%、156.45%、657.25%;T1增强126.88%、2.7%、198.7%、192.92%、128%、149.19%、531.96%;T1剪影 -605.04%、-656.93%、739.06%、-325.36%、1919.53%、4967.44%、6081.3%。T1剪影影像组学特征变化显著高于T1平扫、T1增强。结论 脑白质MR影像组学特征在放疗前后不同剂量梯度下的变化显著,较常规大体影像可更好的揭示脑白质微观变化,为脑白质放射性损伤的早期预测及检测提供了可行的客观方法。     

影像组学在肺癌中的研究进展

影像组学(radiomics)技术在可视化、精准定量化以及人工智能技术的推动下,高通量地从CT、MRI、PET等方面提取并分析大量高级的定量影像学数据,最终通过肿瘤异质性对肿瘤诊断及鉴别诊断、分型及分期、转移、基因表达、疗效评估以及预后等方面展现出巨大的价值,加快了肿瘤学的临床和转化研究。本文从影像组学协助肺癌诊断、评估治疗反应及预测患者预后三方面进行论述,旨在提高对肺癌影像组学的认识。     

多参数MRI的影像组学融合模型在术前预测宫颈癌淋巴结转移的应用价值

目的 探讨基于多参数MRI及临床特征的融合模型在术前预测宫颈癌患者淋巴结转移的价值。方法 回顾性分析山西省肿瘤医院2016年6月-2019年3月经病理证实为宫颈鳞癌并于术前行MRI检查的168例患者的资料。按照7∶3的比例,采用完全随机法将所有患者分为训练组115例和验证组53例。由两名影像科医师在MRI图像上手动勾画三维容积感兴趣区(VOI),并进行一致性分析。根据临床手术病理结果将所有患者分为淋巴结转移阴性(LN-)和阳性(LN+),临床及影像资料也对应分组。分别基于每例患者的T₂WI、表观扩散系数(ADC)和增强T₁WI(cT₁WI)序列图像上均提取3 111个影像组学特征,然后对训练组采用以最大相关最小冗余(MRMR)和最小绝对收缩与选择(LASSO)回归为主的四步法进行特征选择和影像组学标签的构建,并进行分层分析。通过多变量逻辑回归筛选独立临床危险因素并联合影像组学标签构建影像组学融合模型,并制作列线图。采用ROC曲线、校正曲线、决策分析曲线(DCA)评估列线图的预测性能及临床效益。结果 训练组和验证组患者基线资料差异均无统计学意义(P值均>0.05)。基于T₂WI、ADC和cT₁WI合并特征降维后共得到6个影像组学特征(P值均＜0.05),其中包括3个小波类特征参数和3个LoG类特征参数,均与淋巴结转移显著相关。单序列影像组学标签在训练组中ROC曲线下面积(AUC)值为0.763和0.829,显示具有良好的预测效能,合并上述序列构建的影像组学标签对应的AUC值0.859,其诊断效能优于其中任意单一序列,并在验证组得到验证。联合影像组学标签和MRI评价淋巴结状态构建的列线图在训练组和验证组中均显示出良好的鉴别能力和校正性能,对应的AUC分别为 0.865和0.861。在独立验证组中的决策曲线示,当风险阈值＞10%时,采用影像组学方法预测LN+的净收益优于将所有患者都看作LN+或LN-,也优于MRI评价淋巴结状态。结论 通过联合基于多参数MRI的影像组学标签和MRI评价淋巴结状态建立的融合模型可作为术前评估宫颈癌淋巴结转移的一种辅助方法。     

影像组学在局限期小细胞肺癌脑预防照射优化中的价值

目的探讨治疗前胸部增强CT影像组学模型对局限期小细胞肺癌(LS-SCLC)脑转移的预测能力以及指导个体化预防性脑照射(PCI)的价值。方法回顾性分析2012年1月至2018年12月在山西省肿瘤医院经病理确诊为小细胞肺癌及影像学检查确定为局限期患者资料97例。基于最小绝对值收缩和选择算子(LASSO)Cox与相关性检验筛选与LS-SCLC脑转移显著相关的影像组学特征构建模型, 使用校正曲线、受试者操作特征曲线下面积(AUC)、内部5折交叉验证、决策曲线分析(DCA)与整合布莱尔评分(IBS)评估影像组学模型的预测效能与临床获益, 使用Kaplan-Merier曲线和log-rank检验绘制生存曲线和评估组间差异。结果提取出影像组学特征1272个, 使用LASSO Cox回归和相关性检验筛选特征, 最后通过8个与LS-SCLC患者脑转移发生相关的影像组学特征构建影像组学模型。影像组学模型预测LS-SCLC患者1年与2年脑转移的AUC分别为0.845(95%CI为0.746～0.943)和0.878(95%CI为0.774～0.983)。5折内部交叉验证、校正曲线、DCA以及IBS显示模型有...     

120例胶质瘤及瘤周水肿MRI影像组学在评估肿瘤复发中的价值

目的 探讨胶质瘤及瘤周水肿(PTE)MRI影像组学在评估肿瘤复发中的价值。 方法 选取山东大学齐鲁医院2013年1月至2020年12月经术后病理证实的胶质瘤患者120例,包括55例复发和65例无复发患者,根据术前T₂WI和T₁WI增强图像对肿瘤和PTE进行三维容积感兴趣区勾画,并按照8∶2的比例分为训练组和验证组,分析两者及联合的组学特征与肿瘤复发的关系。使用受试者工作特征(ROC)曲线下面积(AUC)与准确性矩阵,比较和评价不同影像组学模型的训练结果。 结果 对于PTE,K临近法(KNN)分类器预测效能最好:训练组AUC值、敏感度、特异度分别为0.910、0.84、0.88,验证组分别为0.916、0.82、0.93。对于肿瘤,逻辑回归(LR)分类器预测效能最好:训练组AUC值、敏感度和特异度分别为0.777、0.69、0.67,验证组分别为0.758、0.82、0.92。当肿瘤+PTE联合时,逻辑回归(LR)分类器预测效能最好:训练组AUC值、敏感度、特异度为0.977、0.88、0.89,验证组则为0.841、0.73、0.83。 结论 胶质瘤PTE和肿瘤影像组学特征在预测胶质瘤术后复发方面具有一定的价值,其中PTE的KNN组学模型效能最佳。