Deep Learning Model for Predicting Neurodevelopmental Outcome in Very Preterm Infants Using Cerebral Ultrasound

Objective: To develop deep learning (DL) models applied to neonatal cranial ultrasound (CUS) and clinical variables to predict neurodevelopmental impairment (NDI) in very preterm infants (VPIs) at 3 years of corrected age. Patients and Methods: This is a retrospective study of a cohort of VPI (220-3...

Full description

Saved in:
Bibliographic Details
Main Authors: Tahani M. Ahmad, MD, ABR, Alessandro Guida, PhD, Sam Stewart, PhD, Noah Barrett, MSc, Michael J. Vincer, MD, Jehier K. Afifi, MD, MSc
Format: Article
Language:English
Published: Elsevier 2024-12-01
Series:Mayo Clinic Proceedings: Digital Health
Online Access:http://www.sciencedirect.com/science/article/pii/S2949761224001007
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1850285010554191872
author Tahani M. Ahmad, MD, ABR
Alessandro Guida, PhD
Sam Stewart, PhD
Noah Barrett, MSc
Michael J. Vincer, MD
Jehier K. Afifi, MD, MSc
author_facet Tahani M. Ahmad, MD, ABR
Alessandro Guida, PhD
Sam Stewart, PhD
Noah Barrett, MSc
Michael J. Vincer, MD
Jehier K. Afifi, MD, MSc
author_sort Tahani M. Ahmad, MD, ABR
collection DOAJ
description Objective: To develop deep learning (DL) models applied to neonatal cranial ultrasound (CUS) and clinical variables to predict neurodevelopmental impairment (NDI) in very preterm infants (VPIs) at 3 years of corrected age. Patients and Methods: This is a retrospective study of a cohort of VPI (220-306 weeks’ gestation) born between 2004 and 2016 in Nova Scotia, Canada. Clinical data at hospital discharge and CUS images at 3 time points were used to develop DL models using elastic net (EN) and convolutional neural network (CNN). The models’ performances were compared using precision recall area under the curve (PR-AUC) and area under the receiver operation characteristic curve (ROC-AUC) with their 95% ci. Results: Of 665 eligible VPIs, 619 (93%) infants with 4184 CUS images were included. The CNN model combining CUS and clinical variables reported better performance (PR-AUC, 0.75; 95% CI, 072-0.79; ROC-AUC, 0.71; 95% CI, 0.67-0.74) in the prediction of positive NDI outcome compared with the traditional models based solely on clinical predictors (PR-AUC, 0.60; 95% CI, 0.52-0.68; ROC-AUC, 0.72; 95% CI, 0.68-0.75). When analyzed by the CUS plane and acquisition time point, the model using the anterior coronal plane at 6 weeks of age provided the highest predictive accuracy (PR-AUC, 0.81; 95% CI, 0.77-0.91; ROC-AUC, 0.78; 95% CI, 0.66-0.87). Conclusion: We developed and internally validated a DL prognostic model using CUS and clinical predictors to predict NDI in VPIs at 3 years of age. Early and accurate identification of infants at risk for NDI enables referral to targeted interventions, which improves functional outcomes.
format Article
id doaj-art-ce11ff2789174a4b8d05d7529a6a88be
institution OA Journals
issn 2949-7612
language English
publishDate 2024-12-01
publisher Elsevier
record_format Article
series Mayo Clinic Proceedings: Digital Health
spelling doaj-art-ce11ff2789174a4b8d05d7529a6a88be2025-08-20T01:47:25ZengElsevierMayo Clinic Proceedings: Digital Health2949-76122024-12-012459660510.1016/j.mcpdig.2024.09.003Deep Learning Model for Predicting Neurodevelopmental Outcome in Very Preterm Infants Using Cerebral UltrasoundTahani M. Ahmad, MD, ABR0Alessandro Guida, PhD1Sam Stewart, PhD2Noah Barrett, MSc3Michael J. Vincer, MD4Jehier K. Afifi, MD, MSc5Department of Pediatric Radiology, IWK Health, Halifax, Nova Scotia, Canada; Department of Diagnostic Radiology, Dalhousie University, IWK Health, Nova Scotia, Canada; Department of Radiology, University of Jordan, Amman, Jordan; Correspondence: Tahani Ahmad, MD, IWK Health, Department of Diagnostic Imaging, 5850/5980 University Avenue, PO Box 9700, Halifax, Nova Scotia B3K6R8, Canada.Biomedical Translational Imaging Centre (BIOTIC), Department of Diagnostic Radiology, Dalhousie University, Halifax, Nova Scotia, Canada; Department of Diagnostic Imaging, Nova Scotia Health, Nova Scotia, CanadaDepartment of Community Health and Epidemiology, Dalhousie University, Halifax, Nova Scotia, CanadaFaculty of Computer Science, Dalhousie University, Halifax, Nova Scotia, CanadaDivision of Neonatal-Perinatal Medicine, IWK Health, Halifax, Nova Scotia, Canada; Department of Pediatrics, Dalhousie University, Halifax, Nova Scotia, CanadaDivision of Neonatal-Perinatal Medicine, IWK Health, Halifax, Nova Scotia, Canada; Department of Pediatrics, Dalhousie University, Halifax, Nova Scotia, CanadaObjective: To develop deep learning (DL) models applied to neonatal cranial ultrasound (CUS) and clinical variables to predict neurodevelopmental impairment (NDI) in very preterm infants (VPIs) at 3 years of corrected age. Patients and Methods: This is a retrospective study of a cohort of VPI (220-306 weeks’ gestation) born between 2004 and 2016 in Nova Scotia, Canada. Clinical data at hospital discharge and CUS images at 3 time points were used to develop DL models using elastic net (EN) and convolutional neural network (CNN). The models’ performances were compared using precision recall area under the curve (PR-AUC) and area under the receiver operation characteristic curve (ROC-AUC) with their 95% ci. Results: Of 665 eligible VPIs, 619 (93%) infants with 4184 CUS images were included. The CNN model combining CUS and clinical variables reported better performance (PR-AUC, 0.75; 95% CI, 072-0.79; ROC-AUC, 0.71; 95% CI, 0.67-0.74) in the prediction of positive NDI outcome compared with the traditional models based solely on clinical predictors (PR-AUC, 0.60; 95% CI, 0.52-0.68; ROC-AUC, 0.72; 95% CI, 0.68-0.75). When analyzed by the CUS plane and acquisition time point, the model using the anterior coronal plane at 6 weeks of age provided the highest predictive accuracy (PR-AUC, 0.81; 95% CI, 0.77-0.91; ROC-AUC, 0.78; 95% CI, 0.66-0.87). Conclusion: We developed and internally validated a DL prognostic model using CUS and clinical predictors to predict NDI in VPIs at 3 years of age. Early and accurate identification of infants at risk for NDI enables referral to targeted interventions, which improves functional outcomes.http://www.sciencedirect.com/science/article/pii/S2949761224001007
spellingShingle Tahani M. Ahmad, MD, ABR
Alessandro Guida, PhD
Sam Stewart, PhD
Noah Barrett, MSc
Michael J. Vincer, MD
Jehier K. Afifi, MD, MSc
Deep Learning Model for Predicting Neurodevelopmental Outcome in Very Preterm Infants Using Cerebral Ultrasound
Mayo Clinic Proceedings: Digital Health
title Deep Learning Model for Predicting Neurodevelopmental Outcome in Very Preterm Infants Using Cerebral Ultrasound
title_full Deep Learning Model for Predicting Neurodevelopmental Outcome in Very Preterm Infants Using Cerebral Ultrasound
title_fullStr Deep Learning Model for Predicting Neurodevelopmental Outcome in Very Preterm Infants Using Cerebral Ultrasound
title_full_unstemmed Deep Learning Model for Predicting Neurodevelopmental Outcome in Very Preterm Infants Using Cerebral Ultrasound
title_short Deep Learning Model for Predicting Neurodevelopmental Outcome in Very Preterm Infants Using Cerebral Ultrasound
title_sort deep learning model for predicting neurodevelopmental outcome in very preterm infants using cerebral ultrasound
url http://www.sciencedirect.com/science/article/pii/S2949761224001007
work_keys_str_mv AT tahanimahmadmdabr deeplearningmodelforpredictingneurodevelopmentaloutcomeinverypreterminfantsusingcerebralultrasound
AT alessandroguidaphd deeplearningmodelforpredictingneurodevelopmentaloutcomeinverypreterminfantsusingcerebralultrasound
AT samstewartphd deeplearningmodelforpredictingneurodevelopmentaloutcomeinverypreterminfantsusingcerebralultrasound
AT noahbarrettmsc deeplearningmodelforpredictingneurodevelopmentaloutcomeinverypreterminfantsusingcerebralultrasound
AT michaeljvincermd deeplearningmodelforpredictingneurodevelopmentaloutcomeinverypreterminfantsusingcerebralultrasound
AT jehierkafifimdmsc deeplearningmodelforpredictingneurodevelopmentaloutcomeinverypreterminfantsusingcerebralultrasound