Abstract
BACKGROUND: Necrotizing enterocolitis (NEC) is a severe gastrointestinal condition which is mainly diagnosed by abdominal radiographs. Early diagnosis of NEC remains challenging due to its nonspecific clinical symptoms and the variability in radiographic findings. Radiomics can enhance diagnostic accuracy by extracting quantitative features from medical images. This study aimed to evaluate the value of radiomics as an assistant tool in cases of missed diagnosis by radiologists. METHODS: In this retrospective study, abdominal radiographs from 484 patients were collected, comprising 262 NEC patients and 222 non-NEC patients from January 2016 to December 2022 in Children's Hospital of Fudan University. The dataset was divided into a training set (n=246), test set (n=105), and a temporal validation set (n=133). Feature selection was performed consecutively using the minimum redundancy maximum relevance (mRMR) and the least absolute shrinkage and selection operator (LASSO) algorithms. A radiomics diagnostic model was constructed using logistic regression. Model performance was evaluated using the area under the curve (AUC). In the temporal validation set, we conducted a parallel test diagnosis using radiomics and the diagnostic results of radiologists, and performed a Chi-squared test against the diagnosis of radiologists. RESULTS: The radiomics diagnostic model which has included 18 features achieved AUCs of 0.82, 0.74, and 0.71 for the training set, test set, and temporal validation set, respectively. In the temporal validation set, the diagnostic results of the parallel test were more sensitive than those of the radiologists (P=0.003). CONCLUSIONS: The radiomics model showed certain diagnostic value and offers a unique perspective compared to radiologists, focusing on quantitative features that can assist in early diagnosis and treatment of NEC. This demonstrates the potential of the model in recognizing challenging cases that might be overlooked by naked eyes of radiologists.