Necrotizing Enterocolitis Detection in Premature Infants Using Broadband Optical Spectroscopy

Area of Science:

• Biomedical Optics
• Medical Informatics
• Neonatal Medicine

Background:

• Necrotizing enterocolitis (NEC) is a severe gastrointestinal disease impacting premature infants.
• Early detection of NEC is crucial for improving patient outcomes.
• Noninvasive diagnostic methods are needed for neonatal care.

Purpose of the Study:

• To develop and validate a machine learning approach for NEC classification using broadband optical spectroscopy (BOS).
• To identify optimal spectral features for noninvasive NEC detection in premature infants.

Main Methods:

• Iterative principal component analysis (iPCA) was employed to select informative wavelengths from BOS data.
• Neural network models were trained using selected wavelengths for NEC classification.
• Data were collected in vivo from neonatal intensive care unit (NICU) patients.

Main Results:

• A reduced-feature neural network model achieved 88% accuracy, 89% sensitivity, and 88% specificity for NEC classification.
• Whole-spectrum models showed higher accuracy and specificity, but the reduced-feature model offered superior sensitivity.
• The optimized model requires as few as seven wavelengths for potential NEC screening.

Conclusions:

• Broadband optical spectroscopy (BOS) combined with machine learning shows promise for noninvasive NEC detection.
• The developed iPCA method effectively reduces spectral data while maintaining high classification performance.
• This approach could lead to a medical device for early NEC screening in high-risk infants.