Radiomics-based predictive model of Growth Hormone Status in Children with Short Stature on Contrast-enhanced T1WI Images

Abstract

Abstract Background A predictive model for growth hormone status in children were constructed, by extracting radiomic features based on pituitary post-contrast T1WI images, and its evaluation efficacy was assessed. Methods Biochemical examination data and MRI data of short stature were collected. All patients were treated in Shenzhen Children's Hospital between October 2019 and December 2019. Patients were divided into two groups, growth hormone normal (GHN) and growth hormone deficient group (GHD), according to growth hormone (GH) peak value. GHN, GH VPeak≥10 ng/ml; GHD, GH VPeak <10 ng/ml. Independent Samples t-tests were employing to evaluate the differences in age between the two groups. The differences in age between the two groups were compared using independent-samples t-tests. Then all patients were randomly divided into training and validation groups by the ratio of 7:3. The ROIs were set as the whole pituitary gland, by drawing the outline of whole gland. Radiomics features were extracted using PyRadiomics package. There are 4 steps in radiomics feature selection: Intra-group Correlation, Independent Sample t-Test, least absolute shrinkage and selection operator (LASSO), and Spearman Correlation Analysis. Building the model with Support Vector Machine, Using Receiver Operating Characteristic (ROC) Curves and calculating the Area Under the Curve (AUC) to evaluate the efficiency of the model. RESULTS There were 300 cases of short stature, 136 cases (45.3%) in the GHN, and 164 cases (54.7%) in the GHD. Total 1316 radiomics features were extracted from the images. After 4 steps screening, remaining 13 radiomics features were used for model construction. AUC was used to assessed the predictive accuracy of GH status model; its value for training group and validation group were 0.78 and 0.66 separately. CONCLUSION This study builds a practicable and efficiency GH status model for predicting growth hormone status of short stature patients. It provides a novel and non-invasive approach for growth hormone status evaluation, which would be very helpful for clinic treatment strategy decision.