瘤内及瘤周表观弥散系数影像组学特征预测髓母细胞瘤患儿预后

目的 观察瘤内及瘤周表观弥散系数（ADC）影像组学特征预测髓母细胞瘤（MB）患儿预后的价值。方法 回顾性分析74例MB患儿资料，根据术后2年随访结果将其分为进展组（n=29）及无进展组（n=45），并按6：4比例分为训练集（n=44）或验证集（n=30）。基于ADC图提取并筛选瘤内及瘤周影像组学特征，分别建立瘤内、瘤周及瘤内+瘤周影像组学模型，并以之结合临床及常规影像学特征建立联合模型；比较各模型预测MB患儿预后的效能。结果 训练集中，临床-常规影像-瘤周影像组学模型、临床-常规影像-瘤内+瘤周影像组学模型曲线下面积（AUC）均大于单一瘤周影像组学模型（P均<0.05）；验证集中，临床-常规影像学-瘤内+瘤周影像组学模型的AUC最大，但与其他模型差异均无统计学意义（P均>0.05）。结论 瘤内及瘤周ADC影像组学特征可用于预测MB患儿预后；联合临床及常规影像学特征或有助于提高预测效能。

Intratumoral and peritumoral radiomics features of apparent diffusion coefficient for predicting prognosis of children with medulloblastoma

Objective To observe the value of intratumoral and peritumoral radiomics features of apparent diffusion coefficient (ADC) for predicting prognosis of children with medulloblastoma (MB). Methods Data of 74 children with MB were retrospectively analyzed. The children were divided into progression group (n=29) or non-progression group (n=45) according to results of 2-year follow-up, also into training set (n=44) or validation set (n=30) at the ratio of 6:4. The intratumoral and peritumoral radiomics features were extracted and screened based on ADC images. The intratumoral, peritumoral as well as intratumoral+peritumoral radiomics models were established, and 3 combination models were constructed combining with clinical and conventional imaging features. The predictive efficiency were compared between each 2 combination models. Results In training set, the area under the curve (AUC) of clinical-conventional imaging-peritumoral radiomics model and clinical-conventional imaging-intratumoral+peritumoral radiomics model were both larger than that of peritumoral radiomics models (both P<0.05). In validation set, AUC of clinical-conventional imaging-intratumoral+peritumoral radiomics model was the largest, but no significant difference of AUC was found among 3 combination models (all P>0.05). Conclusion The intratumoral and peritumoral ADC radiomics features could be used to predict prognosis of MB children. Combining with clinical and conventional imaging features might improve the efficiency of prediction.