Purpose To identify hemangioma(HE)and Kaposiform hemangioendothelioma(KHE)by constructing two ultrasonography-based radiomics models to evaluate the application value of two models in distinguishing HE from KHE,and to compare the diagnostic efficiency of two models.Materials and Methods A total of 90 lesions of subcutaneous HE or KHE confirmed clinically or pathologically from Henan Provincial People's Hospital from August 2020 to May 2022,were retrospectively analyzed.Imaging features were extracted by using Pyradiomics and screened out by the least absolute shrinkage and selection operator algorithm.Support vector machine and random forest were used to construct the radiomics models.Then the diagnostic efficacy of different models was compared.Results Based on the selected 10 radiomics features,the area under the curve,accuracy,sensitivity,specificity,positive and negative prediction the training group and validation group of the support vector machine model were 0.902(95%CI 0.887-0.917),92.1%,85.0%,92.3%,90.9%,93.5%and 0.827(95%CI 0.787-0.856),85.2%,70.0%,94.1%,90.9%,85.0%,respectively;and those in the training group and validation group of the random forest model were 0.960(95%CI 0.938-0.983),98.4%,96.4%,97.8%,98.1%,97.2%and 0.742(95%CI 0.699-0.785),77.8%,57.1%,82.3%,79.6%,62.5%,respectively.The area under the curve between two models in the training group and validation group was statistically significant(Z=-3.306,-2.009;P<0.05).Conclusion Ultrasonography-based radiomics can distinguish HE from KHE,support vector machine model shows more stable diagnostic performance in small sample data.
NETL
NSTL
万方数据
