Share：
Share this content in WeChat
X
[Chinese] [PDF] 29 1
• Clinical Article •
Development of a neonatal bilirubin encephalopathy prediction model using deep learning radiomics and mediation analysis
XU Shuang  LI Xiaoyan  QIAN Jingyu  YU Liang  JIANG Wei  DONG Xian 

Cite this article as XU S, LI X Y, QIAN J Y, et al. Development of a neonatal bilirubin encephalopathy prediction model using deep learning radiomics and mediation analysis[J]. Chin J Magn Reson Imaging, 2026, 17(6): 18-27. DOI:10.12015/issn.1674-8034.2026.06.003.


[Abstract] Objective To develop and validate a deep learning-radiomics (DL Radiomics) model for predicting the occurrence of bilirubin encephalopathy (BE) in neonates with severe hyperbilirubinemia, compare its performance with conventional clinical models, and explore the potential relationships among biochemical markers, radiomics features, and BE using mediation analysis.Materials and Methods This retrospective study included 173 neonates with severe hyperbilirubinemia admitted to the East and West Campuses of the Anhui Provincial Women and Children's Medical Center between January 2022 and September 2025. Clinical data and MRI scans were collected, and divided the dataset into two groups using a 7∶3 ratio. Deep learning features and handcrafted radiomics features were extracted from T1-weighted images using DenseNet-121 and PyRadiomics, respectively. After feature selection based on mutual information, a DL Radiomics model (Rad-score) was constructed. Clinical variables were selected using LASSO and logistic regression to build a nomogram model. Model performance was evaluated in the training and testing sets, with internal validation performed using leave-one-out, 10-fold cross-validation, and bootstrapping. Mediation analysis was conducted to assess the relationships among biochemical markers, Rad-score, and BE.Results Among the 173 neonates, 65 (37.57%) developed BE. The final Rad-score was derived from two deep learning features and three radiomics features. In the clinical model, weight, gestational age, mode of delivery, premature rupture of membranes, isoimmune hemolysis, total bilirubin, hemoglobin, and platelet count were identified as independent predictors (OR range: 0.993 to 27.935). The DL Radiomics model achieved area under the curves (AUCs) ranging from 0.762 to 0.796 across datasets, while the clinical model yielded AUCs of 0.767 to 0.843. The nomogram model demonstrated improved performance, with AUCs of 0.873 to 0.916, and the nomogram excluding biochemical variables achieved AUCs of 0.857 to 0.892. Mediation analysis indicated that the association between hemoglobin with total bilirubin and BE was partially mediated by the Radscore, with a mediation proportion of 20.70% and 43.10% (P = 0.006, P = 0.092).Conclusions The DL Radiomics model, when combined with clinical variables excluding biochemical markers, demonstrates favorable discriminative performance in predicting BE. In addition, the Radscore is found to be associated with peripheral hemoglobin and total bilirubin levels.
[Keywords] deep learning；radiomics；magnetic resonance imaging；bilirubin encephalopathy；predictive model

XU Shuang1   LI Xiaoyan2*   QIAN Jingyu3   YU Liang1   JIANG Wei1   DONG Xian2  

1 Department of Radiology, Anhui Women and Children's Medical Center (Hefei Maternal and Child Health Hospital), Hefei 230001, China

2 Department of Neonatal, Anhui Women and Children's Medical Center (Hefei Maternal and Child Health Hospital), Hefei 230001, China

3 Department of Obstetrics and Gynecology, Anhui Women and Children's Medical Center (Hefei Maternal and Child Health Hospital), Hefei 230001, China

Corresponding author: LI X Y, E-mail: 1801736240@qq.com

Conflicts of interest   None.

Received  2026-02-11
Accepted  2026-05-13
DOI: 10.12015/issn.1674-8034.2026.06.003
Cite this article as XU S, LI X Y, QIAN J Y, et al. Development of a neonatal bilirubin encephalopathy prediction model using deep learning radiomics and mediation analysis[J]. Chin J Magn Reson Imaging, 2026, 17(6): 18-27. DOI:10.12015/issn.1674-8034.2026.06.003.

[1]
WICKREMASINGHE A C, KUZNIEWICZ M W. Neonatal hyperbilirubinemia[J]. Pediatr Clin N Am, 2025, 72(4): 605-622. DOI: 10.1016/j.pcl.2025.04.003.
[2]
LEE B, PIERSANTE T, CALKINS K L. Neonatal hyperbilirubinemia[J/OL]. Pediatr Ann, 2022, 51(6) [2026-02-11]. https://journals.healio.com/doi/10.3928/19382359-20220407-02. DOI: 10.3928/19382359-20220407-02.
[3]
ISKANDER I, GAMALELDIN R, HOUCHI S EL, et al. Serum bilirubin and bilirubin/albumin ratio as predictors of bilirubin encephalopathy[J/OL]. Pediatrics, 2014, 134(5): e1330-e1339 [2026-02-11]. https://publications.aap.org/pediatrics/article-abstract/134/5/e1330/75894/Serum-Bilirubin-and-Bilirubin-Albumin-Ratio-asredirectedFrom=fulltext. DOI: 10.1542/peds.2013-1764.
[4]
WANG S S, WANG C, ZHENG S Q, et al. Easy-to-use nomogram to predict neonatal hyperbilirubinemia[J/OL]. PeerJ, 2025, 13: e20017 [2026-02-11]. https://pubmed.ncbi.nlm.nih.gov/40936765/. DOI: 10.7717/peerj.20017.
[5]
XIA X X, CHEN L M, YANG W, et al. Analysis of risk factors and construction of a risk prediction model for neonatal acute bilirubin encephalopathy[J]. J Ningxia Med Univ, 2025, 47(8): 802-807. DOI: 10.16050/j.cnki.issn1674-6309.2025.08.008.
[6]
QIU X, DONG Y B, CAO Y Q. Changes of serum CRP, AFP and IL-6 levels in neonatal jaundice and their predictive value for acute bilirubin encephalopathy[J]. J Mol Diagn Ther, 2025, 17(4): 606-609. DOI: 10.3969/j.issn.1674-6929.2025.04.005.
[7]
Expert consensus on the diagnosis and treatment of neonatal hyperbilirubinemia. Expert consensus on diagnosis and treatment of neonatal hyperbilirubinemia[J]. Chin J Pediatr, 2014, 52(10): 745-748. DOI: 10.3760/cma.j.issn.0578-1310.2014.10.006.
[8]
XU X F, CHEN M J, JI J S. Progress in the application of radiomics in hepatocellular carcinoma[J]. J Hepatopancreatobiliary Surg, 2024, 36(1): 59-64. DOI: 10.11952/j.issn.1007-1954.2024.01.013.
[9]
TANG J, YI T, LIU Y Q, et al. Prognostic prediction of hypoxic-ischemic encephalopathy in full-term neonates based on multimodal MR radiomics[J]. Chin Comput Med Imaging, 2025, 31(5): 726-731. DOI: 10.3969/j.issn.1006-5741.2025.05.021.
[10]
XIA Y Q, YANG M S, QIAN T Y, et al. Correlation analysis between MRI-based radiomics features and clinical grading of neonatal hypoxic-ischemic encephalopathy[J]. Chin J Evid Based Pediatr, 2025, 20(4): 259-264. DOI: 10.3969/j.issn.1673-5501.2025.04.004.
[11]
YOU Q J, WAN C. Medical image segmentation methods based on deep learning[J]. Chin J New Clin Med, 2020, 13(2): 115-118. DOI: 10.16180/j.cnki.issn1007-7820.2024.01.011.
[12]
Chinese Society of Pediatrics, Neonatology Group, and Editorial Board of the Chinese Journal of Pediatrics. Guidelines for the diagnosis and management of neonatal hyperbilirubinemia (2025)[J]. Chin J Pediatr, 2025, 63(4): 338-350. DOI: 10.3760/cma.j.cn112140-20241231-00944.
[13]
MA J L, KONG D X, WU F, et al. Densely connected convolutional networks for ultrasound image based lesion segmentation[J/OL]. Comput Biol Med, 2024, 168: 107725 [2026-02-11]. https://www.sciencedirect.com/science/article/pii/S0010482523011903via%3Dihub. DOI: 10.1016/j.compbiomed.2023.107725.
[14]
ZHANG Y, LUAN H N, SONG P. Bilirubin metabolism and its application in disease prevention: mechanisms and research advances[J/OL]. Inflamm Res, 2025, 74: 81 [2026-02-11]. https://link.springer.com/article/10.1007/s00011-025-02049-w. DOI: 10.1007/s00011-025-02049-w.
[15]
GEORGE J, POOMALAI I, TERRENCE D H, et al. Cord blood bilirubin, albumin, and hemoglobin as predictors of significant hyperbilirubinemia in ABO-incompatible newborns: a prospective study in a tertiary neonatal center in India[J/OL]. Cureus, 2025, 17(10): e94975 [2026-02-11]. https://pubmed.ncbi.nlm.nih.gov/41132481/. DOI: 10.7759/cureus.94975.
[16]
LV X X, WANG S J, YANG F, et al. Histogram and texture analysis of conventional MRI sequences in the diagnosis of bilirubin encephalopathy[J]. J Clin Radiol, 2020, 39(12): 2503-2506. DOI: 10.13437/j.cnki.jcr.2020.12.032.
[17]
SUN Y, LIAO Y, JIA F L, et al. The differential value of radiomics based on traditional T1-weighted sequences in newborns with hyperbilirubinemia[J/OL]. BMC Med Imaging, 2023, 23(1): 112 [2026-02-11]. https://link.springer.com/article/10.1186/s12880-023-01075-6. DOI: 10.1186/s12880-023-01075-6.
[18]
CHEN F, ZHANG J B, REN H. Diagnostic value of serum neuron-specific enolase, serum total bilirubin combined with neurobehavioral score in neonatal bilirubin encephalopathy[J]. Chin J Clin, 2024, 52(6): 726-729. DOI: 10.3969/j.issn.2095-8552.2024.06.024.
[19]
YAN W L, DU N, ZHANG K, et al. Bilirubin-microbiota interaction: molecular mechanisms and therapeutic strategies in neonatal jaundice[J/OL]. Front Microbiol, 2025, 16: 1749152 [2026-02-11]. https://pubmed.ncbi.nlm.nih.gov/41608689/. DOI: 10.3389/fmicb.2025.1749152.
[20]
WU M, LIU Q, LAI C. Detecting acute bilirubin encephalopathy in neonates based on multimodal MRI images and non-image clinical data[J/OL]. BMC Pediatr, 2025, 25(1): 419 [2026-02-11]. https://pubmed.ncbi.nlm.nih.gov/40414877/. DOI: 10.1186/s12887-025-05411-3.
[21]
XU H, JIANG X N, LI J, et al. Analysis of influencing factors of acute bilirubin encephalopathy in neonates with severe hyperbilirubinemia and MRI image features[J]. J Clin Radiol, 2025, 44(9): 1744-1749. DOI: 10.13437/j.cnki.jcr.2025.09.016.
[22]
DONG S J, SUN B, WU M, et al. Machine learning models based on MRI radiomics have high diagnostic value for neonatal acute bilirubin encephalopathy[J]. J Mol Imaging, 2025, 48(10): 1213-1218. DOI: 10.12122/j.issn.1674-4500.2025.10.04.
[23]
SUN B, DONG S J, WU M. Automatic Diagnostic study of neonatal acute bilirubin encephalopathy based on multimodal magnetic resonance images and deep learning[J]. China Digit Med, 2025, 20(7): 64-72. DOI: 10.3969/j.issn.1673-7571.2025.07.011.
[24]
TAN W T, LUO Y, SUN W S, et al. Diagnosis of neonatal bilirubin encephalopathy based on quantitative analysis of multiparameter magnetic resonance imaging[J]. Chin J Magn Reson Imaging, 2023, 14(2): 27-32. DOI: 10.12015/issn.1674-8034.2023.02.005.
[25]
LI Y, ZHANG X Y, LIU L L, et al. Application value of 3.0 T MAGIC relaxation time quantitative technique in neonates with acute bilirubin encephalopathy[J]. Chin J Magn Reson Imaging, 2023, 14(10): 65-70, 97. DOI: 10.12015/issn.1674-8034.2023.10.012.
[26]
LINDQVIST D, HANSSON M, THOMAS M, et al. Bilirubin photoisomers in neonatal jaundice[J/OL]. Int J Mol Sci, 2025, 26(21): 10791 [2026-02-11]. https://pubmed.ncbi.nlm.nih.gov/41226827/. DOI: 10.3390/ijms262110791.
[27]
FALKE M. The basics of neonatal hyperbilirubinemia[J]. Neonatal Netw, 2025, 44(1): 61-67. DOI: 10.1891/nn-2024-0051.
[28]
LI Y L, LIU H Y, CHEN K, et al. Pathological significance and prognostic roles of indirect bilirubin/albumin ratio in hepatic encephalopathy[J/OL]. Front Med, 2021, 8: 706407 [2026-02-11]. https://pubmed.ncbi.nlm.nih.gov/34527681/. DOI: 10.3389/fmed.2021.706407.
[29]
GBUREK-AUGUSTAT J, SORGE I, STANGE M, et al. Acute and chronic kernicterus: MR imaging evolution of globus pallidus signal change during childhood[J]. AJNR Am J Neuroradiol, 2023, 44(9): 1090-1095. DOI: 10.3174/ajnr.a7948.
[30]
HANSEN T W R, WONG R J, STEVENSON D K. Molecular physiology and pathophysiology of bilirubin handling by the blood, liver, intestine, and brain in the newborn[J]. Physiol Rev, 2020, 100(3): 1291-1346. DOI: 10.1152/physrev.00004.2019.
[31]
VARGHESE S, PRAKASH S, MUKHERJEE S, et al. Hemolytic disease of the fetus and newborn due to minor blood group alloimmunization in a mother of sickle cell disease with multiple alloantibodies[J]. Asian J Transfus Sci, 2023, 17(2): 291-294. DOI: 10.4103/ajts.ajts_161_22.
[32]
TEWARI V V, KUMAR A, SINGHAL A, et al. Evaluation of Rh-hemolytic disease in neonates and management with early intensive phototherapy in the neonatal intensive care unit[J]. J Trop Pediatr, 2020, 66(1): 75-84. DOI: 10.1093/tropej/fmz033.
[33]
TESFIE T K, ANLAY D Z, ABIE B, et al. Nomogram to predict risk of neonatal mortality among preterm neonates admitted with sepsis at University of Gondar Comprehensive Specialized Hospital: risk prediction model development and validation[J/OL]. BMC Pregnancy Childbirth, 2024, 24(1): 139 [2026-02-11]. https://pubmed.ncbi.nlm.nih.gov/38360591/. DOI: 10.1186/s12884-024-06306-4.
[34]
ZHANG F H, CHEN L H, SHANG S Q, et al. A clinical prediction rule for acute bilirubin encephalopathy in neonates with extreme hyperbilirubinemia: a retrospective cohort study[J/OL]. Medicine, 2020, 99(9): e19364 [2026-02-11]. https://pubmed.ncbi.nlm.nih.gov/32118780/. DOI: 10.1097/MD.0000000000019364.
[35]
WIEDERMANN C J, ZABOLI A, TURCATO G. Synthesis of expert opinions on fluid management in severe sepsis: a contextual review of human albumin and crystalloids[J/OL]. Heart Lung, 2025, 70: 339-359 [2026-02-11]. https://www.heartandlung.org/article/S0147-9563(25)00012-3/abstract. DOI: 10.1016/j.hrtlng.2025.01.010.
[36]
BELAY H G, DEBEBE G A, AYELE A D, et al. Determinants of neonatal jaundice in Ethiopia: a systematic review and meta-analysis[J]. World J Pediatr, 2022, 18(11): 725-733. DOI: 10.1007/s12519-022-00597-3.
[37]
XIN S M, LIU M J, LAI H, et al. The severity of intrahepatic cholestasis of pregnancy and its association with pregnancy complications and neonatal asphyxia: a single-center case analysis and systematic review[J]. Biomol Biomed, 2024, 24(6): 1501-1516. DOI: 10.17305/bb.2024.10588.

PREV Correlation between brain functional network alterations and cognitive function in patients with type 1 diabetes mellitus
NEXT Changes of ALPS index and its clinical significance in patients with cerebral small vessel disease presenting with chronic dizziness
  


LINK


Tel & Fax: +8610-67113815    E-mail: editor@cjmri.cn