Hepatocellular carcinoma (HCC) is a highly prevalent malignant tumor of digestive system, characterized by elevated morbidity and mortality rates worldwide1,2. Surgical resection is often not feasible upon diagnosis, resulting in a poor prognosis.
肝细胞癌(HCC)是一种常见的消化系统恶性肿瘤,其发病率和死亡率在世界范围内均较高。诊断后手术切除往往不可行,导致预后不良。
Within the cancer immune cycle, the immune checkpoint PD1 and its ligand PDL1 play pivotal roles in tumor evasion of immune-mediated apoptosis, suppression of T and B cell activation, and facilitation of tumor metastasis and progression. Elevated PDL1 expression correlates with these phenomena3-5. Targeted therapy for the PD1/PDL1 axis effectively suppresses tumor activity and restores immune cell functionality, indicating PDL1 expression as a potential biomarker guiding subsequent treatment strategies.
在癌症免疫周期内,免疫检查点PD 1及其配体PDL 1在肿瘤逃避免疫介导的细胞凋亡、抑制T和B细胞活化以及促进肿瘤转移和进展中发挥关键作用。PDL 1表达升高与这些现象相关3-5。针对PD 1/PDL 1轴的靶向治疗有效抑制肿瘤活性并恢复免疫细胞功能,表明PDL 1表达是指导后续治疗策略的潜在生物标志物。
Vessels Encapsulating Tumor Clusters (VETC) refers to tumor lesions surrounded by capillary endothelial cells that facilitate hematogenous metastasis, resulting in an unfavorable prognosis post-radical surgical resection. Consequently, VETC serves as a robust adverse prognostic indicator.6-8 Meanwhile, the expression of VETC provides guidance in selecting subsequent treatment strategies, such as agents like sorafenib demonstrate a notable increase in overall survival for VETC-positive patients, the do not extend survival in VETC-negative patients.9
血管包裹肿瘤簇(VETC)是指被毛细血管内皮细胞包围的肿瘤病变,其促进血行转移,导致根治性手术切除后预后不良。因此,VETC作为一个强大的不良预后指标。6-8同时,VETC的表达为后续治疗策略的选择提供了指导,例如索拉非尼等药物显示出VETC阳性患者的总生存期显著增加,但不能延长VETC阴性患者的生存期。9
A definitive pathological diagnosis is not currently required for the diagnosis of hepatocellular carcinoma. Many patients ineligible for surgical resection who also decline biopsy procedures lack pathological evidence to guide subsequent treatment decisions. Importantly, assessing the expression of PDL1 and VETC require histopathological evaluation. Therefore, there is a need for a non-invasive, convenient method to determine PDL1 and VETC expression in patients without histopathological data. Artificial intelligence (AI), specifically deep learning technologies applied to medical imaging, supports the need for non-invasive and precise medical interventions. AI has already shown promise in analyzing the cytoplasmic matrix and microenvironment of hepatocellular cancer cells, distinguishing between liver cancer types and predicting therapeutic responses to refine treatment strategies. 10AI-driven image analysis has the potential to predict PDL1 and VETC expression, thereby informing the choice of follow-up treatments hepatocellular carcinoma.
肝细胞癌的诊断目前不需要明确的病理诊断。许多不适合手术切除的患者也拒绝活检程序,缺乏病理学证据来指导随后的治疗决策。重要的是,评估PDL1和VETC的表达需要组织病理学评价。因此,需要一种非侵入性的、方便的方法来确定没有组织病理学数据的患者中的PDL 1和VETC表达。人工智能(AI),特别是应用于医学成像的深度学习技术,支持对非侵入性和精确医疗干预的需求。 人工智能已经在分析肝细胞癌细胞的细胞质基质和微环境,区分肝癌类型和预测治疗反应以完善治疗策略方面显示出前景。10人工智能驱动的图像分析有可能预测PDL 1和VETC表达,从而为肝细胞癌后续治疗的选择提供信息。
Materials and Methods
材料和方法
In this retrospective single-center study, we reviewed patients diagnosed with hepatocellular carcinoma at our institution between January 2015 and December 2017, who had undergone pathological examination and abdominal CT enhancement scans. The hospital’s ethics committee approved the study and waived the requirement for informed consent. After screening, 119 patients were enrolled.
在这项回顾性单中心研究中,我们回顾了2015年1月至2017年12月期间在我们机构诊断为肝细胞癌的患者,这些患者接受了病理检查和腹部CT增强扫描。医院的伦理委员会批准了这项研究,并放弃了知情同意的要求。筛选后,119例患者入组。
Inclusion criteria included: 1) histologically confirmed hepatocellular carcinoma; 2) confirmed expression of PDL1 and VETC via histopathology; 3) complete abdominal enhanced CT imaging data available, with an interval of less than one week between pathology and imaging examinations; 4) no prior surgical, interventional, or systemic treatment for liver cancer before the imaging examination; 5) complete clinical and laboratory data available.
入选标准包括:1)组织学证实的肝细胞癌;2)通过组织病理学证实PDL 1和VETC表达;3)完整的腹部增强CT成像数据可用,病理学和影像学检查之间的间隔时间小于1周;4)在影像学检查之前,既往未接受过肝癌的手术、介入或全身治疗;5)完整的临床和实验室数据可用。
Exclusion criteria included: 1) absence of abdominal enhanced contrast CT imaging data; 2) lack of pathological examination; 3) history of surgical, interventional, chemotherapy, targeted therapy, or immunotherapy for cancer; 4) significant motion artifacts; 5) concurrent other malignancies; 6) incomplete medical information.
排除标准包括:1)缺乏腹部增强CT成像数据; 2)缺乏病理学检查; 3)癌症手术、介入、化疗、靶向治疗或免疫治疗史; 4)显著运动伪影; 5)并发其他恶性肿瘤; 6)医学信息不完整。
To ensure an appropriate sample distribution, the dataset was randomly divided into a training set and a test set using an 8:2 ratio. The screening and grouping process is illustrated in Figure 1.
为了确保适当的样本分布,使用8:2的比例将数据集随机分为训练集和测试集。筛选和分组过程如图1所示。
Figure 1 Flowchart of the study population.
图1研究人群的流程图。
Assessment of PDL1 and VETC Expression
PDL 1和VETC表达评估
In this study, the evaluation of PDL1 and VETC expression was independently conducted by two experienced pathologists. In cases of disagreement, a third pathologist was consulted to discuss and reach a consensus on the final conclusion. All three pathologists were blinded to the patients' clinical, laboratory, and imaging findings prior to reaching the final decision.
在本研究中,由两名经验丰富的病理学家独立进行PDL 1和VETC表达的评价。如果意见不一致,则咨询第三位病理学家进行讨论,并就最终结论达成共识。在做出最终决定之前,所有三位病理学家都对患者的临床、实验室和影像学检查结果不知情。
VETC is defined as the occurrence of sinusoid-like vessels that form web-like networks and encapsulate individual tumor clusters either partially or entirely within the tumor, as observed in CD34 immunostaining. The extent of VETC was semi-quantitatively assessed in increments of 5% from 0% to 100%. A VETC index of 5% or higher classifies the tumor as VETC positive (VETC+); below this threshold, it is classified as VETC negative (VETC-)11.
VETC定义为出现窦状血管,形成网状网络,并将单个肿瘤簇部分或全部包裹在肿瘤内,如CD 34免疫染色所示。VETC的程度以5%的增量从0%至100%进行半定量评估。VETC指数为5%或更高时,将肿瘤分类为VETC阳性(VETC+);低于此阈值时,将其分类为VETC阴性(VETC-)11。
The intensity of staining in the tissue microarray was rated on a scale from 0 (Negative) to 3 (Dark brown). The proportion of positively stained cells was scored as follows: 0% (0 points), 1-25% (1 point), 26-50% (2 points), 51-75% (3 points), and 76-100% (4 points). The total HC staining score was calculated by multiplying the intensity score by the proportion score12,13. A total score exceeding 4 points indicated high expression (PDL1+), whereas scores of 4 or less indicated low expression (PDL1-).
组织微阵列中的染色强度以0(阴性)至3(深棕色)的等级评定。阳性染色细胞的比例评分如下:0%(0分)、1-25%(1分)、26-50%(2分)、51-75%(3分)和76-100%(4分)。HC染色总评分通过强度评分乘以比例评分12,13计算。总分超过4分表示高表达(PDL 1+),而4分或更低表示低表达(PDL 1-)。
CT acquisition
CT采集
Patients underwent contrast-enhanced liver CT within 2 weeks before treatment initiation.
患者在治疗开始前2周内接受了对比增强肝脏CT。
CT scans were performed in the axial plane with 1.25mm-thick sections using a 64-row multi-detector row CT scanner (Discovery CT750hd; GE Healthcare, Waukesha, Wis). A liver protocol includes a non-contrast CT followed by subsequent contrast-enhanced scan encompassing arterial phase, portal phase, and delayed phase. Scan to capture contrast-enhanced images is conducted at 25 seconds, 65 seconds, and 180 seconds after the injection of Ioversol (injection rate of 3 mL/s, dose of 1.0 mL/kg of body weight). The scan parameters are as follows: tube current, 250-350 mA; tube voltage, 120 kVp; rotation time of the tube, 0.8 seconds; pitch, 1.375:1; slice thickness, 1.25 mm; interval thickness, 1.25 mm.
使用64排多探测器排CT扫描仪(Discovery CT 750 hd; GE Healthcare,沃基沙,威斯康星州)在轴向平面中用1.25 mm厚的切片进行CT扫描。肝脏方案包括非造影CT,随后进行包括动脉期、门脉期和延迟期的造影增强扫描。在注射碘佛醇(注射速率为3 mL/s,剂量为1.0 mL/kg体重)后25秒、65秒和180秒进行扫描以捕获造影增强图像。扫描参数如下:管电流,250-350 mA;管电压,120kVp;管旋转时间,0.8秒;螺距,1.375:1;层厚,1.25 mm;间隔厚度,1.25 mm。
Image segmentation and radiomic feature extraction
图像分割与放射组学特征提取
Two physicians, blinded to patients' clinical data and pathology results, manually delineated regions of interest (ROIs) on arterial phase images using the uAI Research Portal (United Imaging Intelligence) to segment tumors, which were then fused to generate the volumes of interest (VOIs). Two physicians independently segmented all images and compared the results. In cases of significant discrepancies, a third, more senior physician evaluated the ROIs. Consensus was reached following discussions among the three physicians.
两名医生对患者的临床数据和病理结果不知情,使用uAI Research Portal(United Imaging Intelligence)在动脉期图像上手动描绘感兴趣区域(ROI)以分割肿瘤,然后将其融合以生成感兴趣体积(VOI)。两名医生独立分割所有图像并比较结果。在存在显著差异的情况下,由第三位更资深的医生评估ROI。三位医生讨论后达成了共识。
Feature extraction was performed through the uAI Research Portal's radiomics module, incorporating first-order statistics features, shape features, and gray-level related features (such as gray-level co-occurrence matrix, among others). Features were extracted from both raw images and those processed with various filters such as box mean, Additive Gaussian Noise, and Binomial Blur Image.
通过uAIResearch Portal的放射组学模块进行特征提取,结合一阶统计特征、形状特征和灰度相关特征(例如灰度共生矩阵等)。从原始图像和用各种滤波器(例如箱均值、加性高斯噪声和BinomialBlurImage)处理的图像两者中提取特征。
Construction and validation of radiomics models
放射组学模型的构建和验证
Subsequently, features were selected sequentially using correlation coefficients, univariate logistic regression, and Least Absolute Shrinkage and Selection Operator regression (Lasso), followed by the calculation of a radiomics score (RAD score) with coefficients used for weighting. Based on the radiomics score, a radiomic model was constructed using machine learning algorithms such as AdaBoost, Bagging Decision Tree (BDT), Decision Tree (DT), Logistic regression (LR), Random Forest (RF), Support Vector Machine (SVM) and eXtreme Gradient Boosting (XGBoost). Receiver Operating Characteristic (ROC) curves were plotted. The area under the curve (AUC) was employed to evaluate the model's predictive accuracy for PDL1 and VETC. Decision Curve Analysis (DCA) was utilized to calculate the net benefit for patients at various threshold probabilities, assessing the clinical utility of each model.
随后,使用相关系数、单变量逻辑回归和最小绝对收缩和选择算子回归(Lasso)依次选择特征,然后使用用于加权的系数计算放射组学s评分(RAD评分)。 基于放射组学评分,采用AdaBoost、Bagging Decision Tree(BDT)、Decision Tree(DT)、Logistic regression(LR)、Random Forest(RF)、支持向量机(SVM)和极限梯度提升(XGBoost)等机器学习算法构建放射组学模型。绘制受试者工作特征(ROC)曲线。采用曲线下面积(AUC)评价模型对PDL 1和VETC的预测准确性。 决策曲线分析(DCA)用于计算患者在不同阈值概率下的净获益,评估每个模型的临床效用。
Statistical analysis
统计分析
Statistical analysis was conducted using SPSS 26 and R software 4.4.1. Normally distributed continuous variables were described using mean ± SD; non-normally distributed continuous variables were described using median and interquartile range (IQR), and categorical variables were expressed as numbers and percentages. Continuous clinical variables were analyzed using the Student's t-test or Mann-Whitney U test, and categorical variables using Pearson's chi-square test. A p-value < 0.05 was considered statistically significant. The Wilcoxon test compared RAD scores’ evaluation efficacy between the training and validation groups for PDL1 and VETC expression. Brier score and Hosmer-Lemeshow test assessed model fit, with p > 0.05 indicating good fit. The DeLong test evaluated the sensitivity, specificity, and AUC of the ROC curves to assess the effectiveness of the classification system.
使用SPSS 26和R软件4.4.1进行统计分析。正态分布的连续变量使用平均值± SD描述;非正态分布的连续变量使用中位数和四分位距(IQR)描述,分类变量表示为数量和百分比。使用Studentt检验或Mann-Whitney U检验分析连续临床变量,使用Pearson卡方检验分析分类变量。认为p值< 0.05具有统计学显著性。Wilcoxon检验比较了训练组和验证组之间RAD评分对PDL 1和VETC表达的评价功效。Brier评分和Hosmer-Lemeshow检验评估了模型拟合,p % 3E 0.05表明拟合良好。 DeLong检验评估了ROC曲线的灵敏度、特异性和AUC,以评估分类系统的有效性。
Result
结果
Patient characteristics
患者特征
This study included 119 patients (male, n=102; mean age 53.82±12.29), among whom 53 patients (44.5%) were PDL1 positive and 60 patients (50.4%) were VETC positive. Patients were randomly allocated in an 8:2 ratio to training sets (PDL1+, n=43, 45.7%; VETC+, n=49, 52.1%) and test sets (PDL1+, n=10, 40%; VETC+, n=11, 44%). There were no significant differences in clinical characteristics between the training and test sets. (Table1)
本研究包括119例患者(男性,n=102;平均年龄53.82±12.29),其中53例(44.5%)PDL 1阳性,60例(50.4%)VETC阳性。将患者以8:2的比例随机分配到训练集(PDL 1+,n=43,45.7%; VETC+,n=49,52.1%)和测试集(PDL 1+,n=10,40%; VETC+,n=11,44%)。训练集和测试集之间的临床特征没有显着差异。(附表一)
Table 1 Characteristic baseline of patients.
表1患者特征基线。
Variables | Total set(n=119) | Training set(n=93) | Test set(n=25) | P value | |
Age(years) | 53.82±11.74 | 53.94±12.29 | 53.36±9.561 | 0.828 | |
Sex(%) | 1 | ||||
Male | 102(85.7) | 81(86.2) | 21(84) | ||
Female | 17(14.3) | 12(13.8) | 4(16) | ||
PDL1(%) | 0.607 | ||||
Negative | 66(55.5) | 51(54.3) | 15(60) | ||
Positive | 53(44.5) | 43(45.7) | 10(40) | ||
VETC(%) | 0.47 | ||||
Negative | 59(49.6) | 45(47.9) | 14(56) | ||
Positive | 60(50.4) | 49(52.1) | 11(44) | ||
BNLC(%) | 0.608 | ||||
0 | 5(4.2) | 4(4.3) | 1(4) | ||
A | 52(43.7) | 40(42.6) | 12(48) | ||
B | 18(15.1) | 14(14.9) | 4(16) | ||
C | 44(37) | 36(38.3) | 8(32) | ||
CNLC(%) | 0.583 | ||||
A | 57(47.9) | 44(46.8) | 13(52) | ||
B | 18(15.1) | 14(14.9) | 4(16) | ||
C | 44(37) | 36(38.3) | 8(32) | ||
ECOG(%) | 0.4 | ||||
0 | 81(68.1) | 62(66) | 19(76) | ||
1 | 23(19.3) | 20(21.3) | 3(12) | ||
2 | 15(12.6) | 12(12.8) | 3(12) | ||
Child-pugh grade(%) | 0.775 | ||||
A | 111(93.3) | 88(93.6) | 23(92) | ||
B | 8(6.7) | 6(6.4) | 2(8) | ||
Tumor number(%) | 0.63 | ||||
1 | 76(63.9) | 59(62.8) | 17(68) | ||
≥2 | 43(36.1) | 35(37.2) | 8(32) | ||
Main tumor diameter (mm) | 65(38.75,95.25) | 64(40.5,88.5) | 0.961 | ||
Tumor metastasis(%) | 0.252 | ||||
No | 95(79.8) | 73(77.7) | 22(88) | ||
Yes | 24(20.2) | 21(22.3) | 3(12) | ||
Venous invasion (%) | 0.875 | ||||
No | 89(74.8) | 70(74.5) | 19(76) | ||
Yes | 30(25.2) | 24(25.5) | 6(24) | ||
Liver cirrhosis (%) | 0.912 | ||||
No | 82(68.9) | 65(69.1) | 17(68) | ||
Yes | 37(31.1) | 29(30.9) | 8(32) | ||
Hepatitis (%) | 0.57 | ||||
No | 16(13.4) | 14(14.9) | 2(8) | ||
Yes | 103(86.6) | 80(85.1) | 23(92) | ||
AFP | 251.8(9.12,3352) | 180.75(9.2475,2996.25) | 323.1(6.815,55312.5) | 0.62 | |
AST | 43(31,64) | 43(31,64.5) | 45(30,66.5) | 0.99 | |
ALT | 36(23,61) | 35(22,61.25) | 38(28.5,62.5) | 0.385 | |
Total bilirubin | 14.9(11.1,19.9) | 14.45(11.075,19.75) | 15.4(11.5,21.95) | 0.577 | |
GGT | 77(40,142) | 80(40,142) | 67(36,142) | 0.538 | |
Albumin | 40.1(36.5,43.7) | 40.8(36.75,43.7) | 39.2(34.3,44.5) | 0.708 | |
PT | 12.8(12.2,13.7) | 12.95(12.2,13.85) | 12.6(11.65,13.2) | 0.092 | |
Hepatic encephalopathy (%) | 1 | ||||
No | 119(100) | 94(100) | 25(100) | ||
Yes | 0(0) | 0(0) | 0(0) | ||
Ascites (%) | 1 | ||||
No | 113(95) | 89(94.7) | 24(96) | ||
Yes | 6(5) | 5(5.3) | 1(4) |
This study extracted a total of 2,286 features, which were then selected using correlation coefficients, logistic regression, and LASSO algorithms. Ultimately, we identified 7 features strongly associated with PDL1 and 10 features closely related to VETC. Radscore was calculated.
这项研究总共提取了2,286个特征,然后使用相关系数,逻辑回归和LASSO算法进行选择。最终,我们确定了7个与PDL 1密切相关的特征和10个与VETC密切相关的特征。计算Radscore。
Figure 2 Features strongly associated with PDL1 and VETC.
图2与PDL 1和VETC密切相关的特征。
PDL1 Radscore = 0.11055924 * NGI + -0.005726956 * LFLK + -0.185813025 * BFK+ -0.324425757 * WGWLL + -0.365141183 * NGG + -0.4365912 * WFWLK + -0.6000737*BNS + 0.7483548
PDL 1Radscore= 0.11055924 *NGI+ -0.005726956 *LFLK+ -0.185813025 *BFK+ -0.324425757 *WGWLL+ -0.365141183 *NGG+ -0.4365912 *WFWLK+ -0.6000737*BNS+ 0.7483548
VETC Radscore = 0.186430633 * WGWL + 0.120599955 * BFE+ 0.105206117 * WFW + 0.09074153 * BGG + 0.0647860542 * LFL + -0.011508585 * WGWH + -0.0331391133 * LGL + -0.03635304 * LBL + -0.04580407 * LGLLR + -0.136617273*RGL + 0.445031732
VETCRadscore= 0.186430633 *WGWL+ 0.120599955 *BFE+ 0.105206117 *WFW+ 0.09074153 *BGG+ 0.0647860542 *LFL+ -0.011508585 *WGWH+ -0.0331391133 *LGL+ -0.03635304 *LBL+ -0.04580407 *LGLLR+ -0.136617273*RGL+ 0.445031732
Various radiomic models for predicting PDL1 and VETC were constructed based on Radscore, and the performance of each model was evaluated (table 2-3 and figure 3). Random Forest (RF) demonstrated the best performance, with an AUC of 0.834 (95% CI: 0.752-0.915) for PDL1 prediction in the training set and an AUC of 0.740 (95% CI: 0.541-0.939) in the test set. For VETC prediction, RF achieved an AUC of 0.883 (95% CI: 0.818-0.949) in the training set and an AUC of 0.705 (95% CI: 0.488-0.922) in the test set. Calibration curves are shown in the figure 4. The Hosmer-Lemeshow test (P > 0.05) and Brier Score indicate good model fit for the RF model.
基于Radscore构建了用于预测PDL 1和VETC的各种放射组学模型,并对每个模型的性能进行了评价(表2-3和图3)。随机森林(RF)表现出最佳性能,训练集中PDL 1预测的AUC为0.834(95% CI:0.752-0.915),测试集中的AUC为0.740(95% CI:0.541-0.939)。对于VETC预测,RF在训练集中实现的AUC为0.883(95% CI:0.818-0.949),在测试集中实现的AUC为0.705(95% CI:0.488-0.922)。校准曲线如图4所示。Hosmer-Lemeshow检验(P <0.05)。05)和Brier评分表明RF模型拟合良好。
Table 2 Performance evaluation of radiomics models in predicting PDL1 expression.
表2放射组学模型预测PDL 1表达的性能评价
Set | Model | AUC(95%CI) | Sensitivity | Specificity | Accuracy | Precision |
Training | RF | 0.834 (0.752-0.915) | 0.818 | 0.731 | 0.771 | 0.72 |
AB | 1(1-1) | 1 | 1 | 1 | 1 | |
BDT | 0.618 (0.504-0.732) | 0.651 | 0.558 | 0.6 | 0.549 | |
DT | 0.858 (0.786-0.931) | 0.628 | 0.923 | 0.789 | 0.871 | |
LR | 0.644 (0.532-0.757) | 0.512 | 0.731 | 0.632 | 0.611 | |
SVM | 0.962 (0.917-1) | 0.953 | 0.865 | 0.905 | 0.854 | |
XGBOOST | 0.87 (0.794-0.946) | 0.721 | 0.885 | 0.811 | 0.838 | |
Test | RF | 0.74 (0.541-0.939) | 0.889 | 0.562 | 0.68 | 0.533 |
AB | 0.722 (0.52-0.924) | 0.556 | 0.688 | 0.64 | 0.5 | |
BDT | 0.576 (0.352-0.801) | 0.222 | 0.75 | 0.56 | 0.333 | |
DT | 0.691 (0.474-0.908) | 0.444 | 0.75 | 0.64 | 0.5 | |
LR | 0.493 (0.244-0.742) | 0.222 | 0.688 | 0.52 | 0.286 | |
SVM | 0.486 (0.203-0.769) | 0.333 | 0.812 | 0.64 | 0.5 | |
XGBOOST | 0.663 (0.406-0.921) | 0.444 | 0.938 | 0.76 | 0.8 |
Table 3 Performance evaluation of radiomics models in predicting VETC expression.
表3放射组学模型在预测VETC表达中的性能评估。
Set | Model | AUC(95%CI) | Sensitivity | Specificity | Accuracy | Precision |
Training | RF | 0.883 (0.818-0.949) | 0.787 | 0.837 | 0.812 | 0.822 |
AB | 1(1-1) | 1 | 1 | 1 | 1 | |
BDT | 0.872 (0.803-0.942) | 0.812 | 0.787 | 0.8 | 0.796 | |
DT | 0.976 (0.955-0.997) | 0.833 | 0.957 | 0.895 | 0.952 | |
LR | 0.794 (0.706-0.882) | 0.75 | 0.617 | 0.684 | 0.667 | |
SVM | 0.59 (0.475-0.705) | 0.292 | 0.872 | 0.579 | 0.7 | |
XGBOOST | 0.998 (0.994-1) | 0.875 | 1 | 0.937 | 1 | |
Test | RF | 0.705 (0.488-0.922) | 0.583 | 0.846 | 0.72 | 0.778 |
AB | 0.604 (0.369-0.839) | 0.636 | 0.5 | 0.56 | 0.5 | |
BDT | 0.588 (0.346-0.829) | 0.545 | 0.714 | 0.64 | 0.6 | |
DT | 0.536 (0.309-0.763) | 0.455 | 0.643 | 0.56 | 0.5 | |
LR | 0.643 (0.415-0.87) | 0.818 | 0.429 | 0.6 | 0.529 | |
SVM | 0.422 (0.168-0.676) | 0.182 | 0.857 | 0.56 | 0.5 | |
XGBOOST | 0.675 (0.44-0.91) | 0.364 | 0.857 | 0.64 | 0.667 |
Figure 3 Comparison of the ROC curves of each radiomic model.
图 3 各 辐射 组 模型 的 曲线 比较 。
A: ROC curve of training group about PDL1. B: ROC curve of test group about PDL1. C: ROC curve of training group about VETC. D: ROC curve of test group about VETC.
A:训练组关于PDL 1的ROC曲线。B:试验组PDL 1的ROC曲线。C:训练组VETC的ROC曲线。D:试验组VETC的ROC曲线。
Figure 4 The calibration curves and Brier scores of each radiomic model.
图4每个放射组学模型的校准曲线和Brier评分。
A: Calibration curve of training group about PDL1. B: Calibration curve of test group about PDL1. C: Calibration curve of training group about VETC. D: Calibration curve of test group about VETC.
A : 1 训练 组 的 曲线 。B : 测试 组 关于 1 的 校准 曲线 。C : 关于 VETC 的 培训 组 的 校准 曲线 。D : 测试 组 关于 VETC 的 校准 曲线 。
Figure 5 Decision curve analysis of each radiomic model.
图5每个放射组学模型的决策曲线分析。
A: Decision curve analysis of training group about PDL1. B: Decision curve analysis of test group about PDL1. C: Decision curve analysis of training group about VETC. D: Decision curve analysis of test group about VETC.
A:训练组关于PDL 1的决策曲线分析。B:试验组关于PDL 1的决策曲线分析。C:训练组关于VETC的决策曲线分析。D:VETC试验组决策曲线分析。
Discussion
In addition, HCC that concurrently exhibits both VETC and classical vascular structures (VETC±) has a worse prognosis than HCC solely expressing VETC- or VETC+. Patients with VETC± have shorter survival times and higher recurrence rates. Pathological evidence suggests that in VETC± HCC, liver metastasis is more likely to occur through an Epithelial-Mesenchymal Transition (EMT)-dependent mechanism associated with VETC- rather than via the VETC pathway14. This variation in expression influences the choice of treatment and prognosis. In this study, we did not separate the VETC± cases from the VETC+ cases. Future studies could isolate these cases for dedicated radiomic analysis.
此外,同时表现出VETC和经典血管结构(VETC±)的HCC比仅表达VETC-或VETC+的HCC具有更差的预后。VETC±患者的生存时间较短,复发率较高。病理学证据表明,在VETC± HCC中,肝转移更可能通过与VETC相关的上皮-间充质转化(EMT)依赖性机制发生,而不是通过VETC途径14。这种表达的变化影响治疗和预后的选择。在本研究中,我们没有将VETC±病例与VETC+病例分开。未来的研究可以将这些病例分离出来进行专门的放射组学分析。
1. Ganesan P, Kulik LM. Hepatocellular Carcinoma: New Developments. Clin Liver Dis. Feb 2023;27(1):85-102. doi:10.1016/j.cld.2022.08.004
1. Ganesan P,Kulik LM.肝细胞癌:新进展。临床肝病2023年2月;27(1):85-102。doi:10.1016/j.cld.2022.08.004
2. Bray F, Laversanne M, Sung H, et al. Global cancer statistics 2022: GLOBOCAN estimates of incidence and mortality worldwide for 36 cancers in 185 countries. CA Cancer J Clin. May-Jun 2024;74(3):229-263. doi:10.3322/caac.21834
2. Bray F,Laversanne M,Sung H,et al. Global Cancer Statistics 2022:GLOBOCAN estimates of incidence and mortality worldwide for 36 cancers in 185 countries. CA癌症临床杂志2024年5月至6月;74(3):229-263。doi:10.3322/caac.21834
3. Lei Q, Wang D, Sun K, Wang L, Zhang Y. Resistance Mechanisms of Anti-PD1/PDL1 Therapy in Solid Tumors. Front Cell Dev Biol. 2020;8:672. doi:10.3389/fcell.2020.00672
3. 雷青,王丹,孙凯,王丽,张艳。抗PD 1/PDL 1治疗实体瘤的耐药机制前池开发生物2020;8:672. doi:10.3389/fcell.2020.00672
4. Thibult ML, Mamessier E, Gertner-Dardenne J, et al. PD-1 is a novel regulator of human B-cell activation. Int Immunol. Feb 2013;25(2):129-37. doi:10.1093/intimm/dxs098
4. Thibult ML,Mamessier E,Gertner-Dardenne J,et al. PD-1 is a novel regulator of human B-cell activation. 国际免疫学2013年2月;25(2):129-37。doi:10.1093/intimm/dxs098
5. Pedoeem A, Azoulay-Alfaguter I, Strazza M, Silverman GJ, Mor A. Programmed death-1 pathway in cancer and autoimmunity. Clin Immunol. Jul 2014;153(1):145-52. doi:10.1016/j.clim.2014.04.010
5. Pedoeem A,Azoulay-Alfaguter I,Strazza M,Silverman GJ,莫尔A.癌症和自身免疫中的程序性死亡-1通路。临床免疫学2014年7月;153(1):145-52。doi:10.1016/j.clim.2014.04.010
6. Matsuda K, Kurebayashi Y, Masugi Y, et al. Immunovascular microenvironment in relation to prognostic heterogeneity of WNT/beta-catenin-activated hepatocellular carcinoma. Hepatol Res. Apr 2023;53(4):344-356. doi:10.1111/hepr.13869
6. Matsuda K,Kurebayashi Y,Masugi Y,et al. Immunovascular microenvironment in relation to prognostic heterogeneity of WNT/beta-catenin-activated hepatocellular carcinoma. 肝醇水库2023年4月;53(4):344-356。doi:10.1111/hepr.13869
7. Fang JH, Zhou HC, Zhang C, et al. A novel vascular pattern promotes metastasis of hepatocellular carcinoma in an epithelial-mesenchymal transition-independent manner. Hepatology. Aug 2015;62(2):452-65. doi:10.1002/hep.27760
7. 方继华,周华春,张聪,等。一种新的血管模式以上皮-间质转化非依赖性方式促进肝细胞癌转移。肝病学。2015年8月;62(2):452-65。doi:10.1002/hep.27760
8. Sugino T, Yamaguchi T, Hoshi N, et al. Sinusoidal tumor angiogenesis is a key component in hepatocellular carcinoma metastasis. Clin Exp Metastasis. 2008;25(7):835-41. doi:10.1007/s10585-008-9199-6
8. Sugino T,Yamaguchi T,Hoshi N,et al.肝窦肿瘤血管生成是肝细胞癌转移的关键组成部分。临床实验转移。2008;25(7):835-41. doi:10.1007/s10585-008-9199-6
9. Fang JH, Xu L, Shang LR, et al. Vessels That Encapsulate Tumor Clusters (VETC) Pattern Is a Predictor of Sorafenib Benefit in Patients with Hepatocellular Carcinoma. Hepatology. Sep 2019;70(3):824-839. doi:10.1002/hep.30366
9. Fang JH,Xu L,Shang LR,et al. Vessels That Encapsulate Tumor Clusters(VETC)Pattern Is a Predictor of Sorafenib Benefit in Patients with Hepatocellular Carcinoma. 肝病学。2019年9月;70(3):824-839。doi:10.1002/hep.30366
10. Liu Z, Zhao P. Integrative analysis unveils ECM signatures and pathways driving hepatocellular carcinoma progression: A multi-omics approach and prognostic model development. J Cell Mol Med. Apr 2024;28(8):e18230. doi:10.1111/jcmm.18230
10. Liu Z,Zhao P. Integrative analysis unveils ECM signatures and pathways driving hepatocellular carcinoma progression:A multi-omics approach and prognostic model development. 细胞分子医学杂志2024年4月;28(8):e18230。doi:10.1111/jcmm.18230
11. Renne SL, Woo HY, Allegra S, et al. Vessels Encapsulating Tumor Clusters (VETC) Is a Powerful Predictor of Aggressive Hepatocellular Carcinoma. Hepatology. Jan 2020;71(1):183-195. doi:10.1002/hep.30814
11. Renne SL,Woo HY,Allegra S,et al. Vessels Encapsulating Tumor Clusters(VETC)Is a Powerful Predictor of Aggressive Hepatocellular Carcinoma. 肝病学。2020年1月;71(1):183-195。doi:10.1002/hep.30814
12. Meyerholz DK, Beck AP. Principles and approaches for reproducible scoring of tissue stains in research. Lab Invest. Jul 2018;98(7):844-855. doi:10.1038/s41374-018-0057-0
12. Meyerholz DK,Beck AP.研究中组织染色可重复评分的原则和方法。实验室投资2018年7月;98(7):844-855。doi:10.1038/s41374-018-0057-0
13. Gong XQ, Liu N, Tao YY, et al. Radiomics models based on multisequence MRI for predicting PD-1/PD-L1 expression in hepatocellular carcinoma. Sci Rep. May 12 2023;13(1):7710. doi:10.1038/s41598-023-34763-y
13. 龚晓勤,刘宁,陶永永,等.基于多序列MRI的放射组学模型预测肝细胞癌中PD-1/PD-L1表达. 科学代表。2023年5月12日;13(1):7710。doi:10.1038/s41598-023-34763-y
14. He C, Zhou Z, Jiang H, et al. Epithelial-Mesenchymal Transition is Superior to Vessels-Encapsulate Tumor Cluster in Promoting Metastasis of Hepatocellular Carcinoma: a Morphological Evidence. J Cancer. 2017;8(1):39-47. doi:10.7150/jca.16736
14. 何春,周忠,姜华,等.上皮-间质转化上级肝细胞癌转移的形态学证据。J Cancer. 2017;8(1):39-47。doi:10.7150/jca.16736