Computer Vision-Based Artificial Intelligence Tool for Direct Bilirubin Jaundice Measurement in Newborns: A Pilot...

Area of Science:

• Neonatal Medicine
• Medical Imaging
• Artificial Intelligence

Background:

• Conjugated hyperbilirubinemia, indicated by elevated direct bilirubin (DB), can signal serious hepatobiliary disorders like biliary atresia in infants.
• Accurate and timely measurement of DB is crucial for diagnosing neonatal jaundice and guiding treatment.
• Current diagnostic methods for DB levels can be invasive and require laboratory analysis.

Purpose of the Study:

• To develop and validate an artificial intelligence (AI)-based computer vision tool for accurately measuring direct bilirubin (DB) levels in newborn infants.
• To assess the precision and correlation of AI-derived DB measurements compared to traditional laboratory blood tests.

Main Methods:

• A computer vision tool employing data processing, color transformations, and contrast enhancement was utilized.
• A convolutional neural network (CNN) was developed to predict DB levels from infant photographs.
• Data from 97 infants (80 for training/validation, 17 for retesting) were analyzed, with five photographs taken per infant under standardized lighting conditions.

Main Results:

• The AI tool demonstrated a 5.24% margin of error when compared to laboratory-based DB measurements.
• A positive correlation was found between the AI system's calculated mean values and the infants' actual blood DB levels.
• The study confirmed the AI tool's capability for precise DB level determination.

Conclusions:

• AI-powered computer vision offers a precise and accurate method for measuring direct bilirubin levels in newborns under appropriate conditions.
• This non-invasive approach shows promise for early detection of jaundice and related hepatobiliary conditions.
• Further research is recommended to extend this AI methodology to total bilirubin measurements.