Machine learning models based on MRI radiomics have high diagnostic value for neonatal acute bilirubin encephalopathy

DONG Shuangjun; SUN Bin; WU Miao; JIA Wenxiao

doi:10.12122/j.issn.1674-4500.2025.10.04

Abstract：Objective To develop an efficient and robust machine learning model for predicting neonatal acute bilirubin encephalopathy based on T1WI radiomics features using six algorithms:Support Vector Machine(SVM),Logistic Regression,Random Forest,K-Nearest Neighbors,Naive Bayes,and Multilayer Perceptron.Methods A retrospective analysis was conducted involving 54 neonates clinically diagnosed with ABE admitted to the First Affiliated Hospital of Xinjiang Medical University from January 2019 to August 2023,with a mean gestational age of 37+2 to 40+1(38.03±2.57)weeks.Additionally,47 healthy neonates were selected as controls,with a mean gestational age of 37+4 to 40+5(38.05±2.61)weeks.High-throughput radiomics features were extracted from T1WI images using Python and Pyradiomics software.Feature selection was performed using Pearson correlation coefficients and least absolute shrinkage and selection operator(LASSO)regression.Subsequently,machine learning models were constructed based on the selected radiomics features,and the classification performance of each algorithm was compared.Results After feature extraction and selection,eight representative radiomics features were identified to construct the ABE radiomics prediction model.Among the algorithms tested,SVM achieved the highest accuracy of 0.739,surpassing the performance of the other five methods.Conclusion Machine learning models based on MRI radiomics show significant clinical potential for diagnosing neonatal ABE.Particularly,the SVM algorithm demonstrates superior classification performance and model stability,offering a novel approach to early ABE diagnosis with promising clinical application prospect.