Diagnosing acute bilirubin encephalopathy in neonates using MRI-based deep learning model

Area of Science:

• Medical Imaging
• Artificial Intelligence in Medicine
• Neonatal Health

Background:

• Neonatal acute bilirubin encephalopathy (ABE) poses a significant threat to infant health.
• Nonspecific early symptoms of ABE often result in delayed or missed diagnoses.
• Accurate and timely diagnosis of ABE is crucial for effective intervention.

Purpose of the Study:

• To develop a computer-assisted integrated model for the clinical assessment and diagnosis of ABE.
• To leverage deep learning for improved ABE detection.
• To create a tool that assists physicians in managing ABE.

Main Methods:

• Retrospective analysis of diagnostic data from ABE and non-ABE patient groups.
• Development and testing of a deep learning (DL) model using multimodal, self-supervised, and multi-instance learning.
• Validation of the DL model against radiologists using independent test cohorts.

Main Results:

• The DL model achieved high accuracy (86.3% and 91.1%) in diagnosing ABE across two test cohorts.
• Area under the curve values demonstrated strong diagnostic performance (91.2% and 89.3%).
• The DL model's performance approached that of senior radiologists.

Conclusions:

• Integrated DL models using clinical and radiological data can accurately diagnose ABE.
• The developed model offers a convenient and cost-effective tool for clinical decision-making.
• This approach enhances the early detection and management of neonatal acute bilirubin encephalopathy.