68Ga-FAPI PET影像组学和机器学习在肝细胞癌病理分化程度术前预测评估中的应用价值研究

The application value of 68Ga-FAPI PET radiomics and machine learning in preoperative prediction of hepatocellular carcinoma differentiation

    摘要
  • 摘要:
    目的  探究68Ga-成纤维细胞活化蛋白抑制剂（FAPI）PET影像组学模型在肝细胞癌（HCC）病理分化程度术前评估中的应用价值。
    方法  回顾性队列研究分析2021年6月至2023年6月于重庆大学附属肿瘤医院行68Ga-FAPI PET显像的90例HCC患者的临床资料，其中男性80例、女性10例，年龄（59.4±9.9）岁。采用简单随机抽样法将所有患者分为训练集（64例）和测试集（26例）。在68Ga-FAPI PET图像上勾画病灶感兴趣区体积（VOI）后进行特征提取。利用4种分类器逻辑回归（LR）、朴素贝叶斯（NB）、K-邻近算法（KNN）和随机森林（RF）进行机器学习，构建PET影像组学模型，并采用受试者工作特征（ROC）曲线进行评估，计算曲线下面积（AUC）及其他效能参数。训练集与测试集的组间差异比较采用独立样本t检验、Wilcoxon检验和卡方检验。
    结果  训练集与测试集的一般临床资料和病理分化结果的组间比较，差异均无统计学意义（χ2=0.002、0.433，t=−0.138~0.067，Z=1.019，均P>0.05）。在训练集中，LR模型的AUC最高0.882（95%CI：0.788~0.796），灵敏度亦最高（0.957）；KNN模型的特异度最高（0.938）。在测试集中，LR模型的AUC最高0.878（95%CI：0.751~1.000）；LR模型和RF模型的灵敏度并列最高（0.933）；NB模型和KNN模型的特异度并列最高（0.833）。
    结论  通过机器学习建立的基于68Ga-FAPI PET图像的影像组学模型在HCC病理分化程度的术前预测评估中的应用价值高，可以帮助HCC患者进行术前个体化预测。

     

    Abstract:
    Objective To explore the application value of 68Ga-fibroblast activation protein inhibitors (FAPI) PET radiomics models in the preoperative assessment of the degree of pathological differentiation of hepatocellular carcinoma (HCC).
    Methods A retrospective cohort study analysis was performed with 90 HCC patients (80 males, 10 females;aged (59.4±9.9) years) who underwent 68Ga-FAPI PET from June 2021 to June 2023, the Chongqing University Cancer Hospital. Patients were randomly divided into a training set (64 cases) and a test set (26 cases) by simple random sampling. Feature extraction was performed after drawing the volume of interest (VOI) of the lesion on 68Ga-FAPI PET images. Four classifiers (logistic regression (LR), naive Bayes (NB), K-nearest neighbors (KNN), randorn forest (RF)) were employed for machine learning to construct PET radiomics models, which were evaluated using the receiver operating characteristic (ROC) curves. Area under the curve (AUC) and other performance parameters were then calculated. Intergroup differences between the training and test sets were compared using independent-sample t-test, Wilcoxon rank-sum test, and Chi-square test.
    Results No significant differences were observed in general clinical data and pathological differentiation outcomes between the training and test sets (χ2=0.002, 0.433, t=−0.138–0.067, Z=1.019, all P>0.05). In the training set, the LR model achieved the highest AUC (0.882 (95%CI: 0.788–0.796)) and sensitivity (0.957), and the KNN model demonstrated the highest specificity (0.938). In the test set, the LR model exhibited the highest AUC (0.878 (95%CI: 0.751–1.000), and the LR and RF models achieved the highest sensitivity (0.933). The NB and KNN models showed the highest specificity (0.833).
    Conclusions The radiomics model based on 68Ga-FAPI PET images was established via machine learning. The model holds significant value for preoperative prediction of HCC pathological differentiation degree and may facilitate personalized preoperative assessment for patients with HCC.

     

    • HTML全文
    • 参考文献(28)
    • 相关文章
    • 施引文献
  • 资源附件(0)

/

返回文章
返回