基于多序列MRI的影像组学模型在腮腺肿瘤良恶性诊断中的应用

  目的  探究腮腺肿瘤良恶性的诊断中应用基于多序列MRI的影像组学模型的临床价值。  方法  选取2021年1月1日—2022年5月30日于台州市中心医院就诊经病理检查证实的腮腺肿瘤患者共97例，其中良性肿瘤64例，恶性肿瘤33例，提取患者的临床资料及MRI图像。通过ITK-SNAP软件分割感兴趣区域(ROI)，使用3D-slicer软件的PyRadiomics插件在T₁加权对比增强成像序列(T₁Wce)、T₂加权序列(T₂WI)及基于弥散加权成像(DWI)序列构建的表观弥散系数(ADC)图像中提取120个影像特征，使用Lasso回归进行特征降维，最后使用筛选的影像特征构建支持向量机(SVM)模型。然后绘制ROC曲线，评估模型的诊断效能。  结果  2组患者年龄、性别差异无统计学意义。基于患者T₁Wce、T₂WI和ADC序列以及3个序列联合构建4个影像组学模型用于腮腺肿瘤良恶性诊断，AUC分别为T₁Wce模型0.752，T₂WI模型0.776，ADC模型0.810，T₁Wce+T₂WI+ADC联合模型0.897，三序列联合模型的AUC显著高于单个序列模型。  结论  基于MRI影像数据构建的影像组学模型能够有效用于辅助腮腺肿瘤的良恶性诊断，其中联合T₁Wce、T₂WI和ADC序列构建的模型具有最佳的诊断效能。 

Multi-sequence MRI image data-based machine learning model in the diagnosis of benign and malignant parotid gland tumours

  Objective  The clinical value of multi-sequence MRI image data-based radiomics in the diagnosed of benign and malignant parotid gland tumors.  Methods  A total of 97 patients with parotid gland tumours diagnosed by pathological examination in Taizhou Central Hospital from January 1, 2021 to May 30, 2022 were selected, including 64 benign tumours and 33 malignant tumours. Meantime, patients' clinical data and MRI images were extracted. ITK-SNAP was used to segment regions of interest. In addition, PyRadiomics plugin of 3D-Slicer was used to extract 120 image features from T₁-weighted contrast enhancement images (T₁Wce), T₂-weighted images (T₂WI) and apparent diffusion coefficient (ADC) images that were constructed on the basis of diffusion-weighted imaging (DWI) sequence. Then, Lasso regression was used for the reduction of image features. Finally, a support vector machine model (SVM) was constructed using the selected image features. Then, the ROC curve was drawn to evaluate the diagnostic efficiency of each model.  Results  No significant differences in age and gender ratio was observed between the two groups. Four imaging models were constructed, including the models based on T₁Wce, T₂WI, ADC images and the combination of the three sequences. The AUC was as follows: T₁Wce model 0.752, T₂WI model 0.776, ADC model 0.810, T₁Wce+T₂WI+ADC model 0.897. The AUC of the three-sequence combination model was significantly higher than that of the single-sequence model.  Conclusion  The radiomics model based on MRI image data can be effectively used for assisting the diagnosis of benign and malignant parotid gland tumours. Moreover, the model based on the combination of T₁Wce, T₂WI and ADC images has the best diagnostic efficiency.