Image-free recognition of moderate ROP from mild with machine learning algorithm on plasma Raman spectrum

Area of Science:

• Biomedical Optics
• Ophthalmology
• Machine Learning

Background:

• Retinopathy of prematurity (ROP) poses significant risks to premature infants.
• Timely diagnosis of ROP is crucial but challenging due to workload on specialists.
• Current diagnostic methods require specialized expertise in retinal image analysis.

Purpose of the Study:

• To develop a non-invasive diagnostic method for ROP using plasma Raman spectroscopy.
• To classify ROP severity using machine learning models based on blood samples.
• To reduce the diagnostic burden on pediatric ophthalmologists.

Main Methods:

• Collected infrared Raman spectra from plasma samples of control and ROP groups.
• Utilized a support vector machine (SVM) model for classification.
• Analyzed spectral differences across different ROP severity levels.

Main Results:

• Identified statistically significant differences in plasma Raman spectra between ROP groups.
• Achieved high accuracy (AUC > 0.90) in differentiating control vs. moderate/advanced ROP, and mild vs. moderate/advanced ROP.
• Demonstrated successful classification between mild and moderate ROP (AUC > 0.90).

Conclusions:

• Plasma Raman spectroscopy combined with machine learning offers a viable, non-invasive approach for ROP diagnosis.
• This method can effectively distinguish between different ROP severity stages, including moderate from mild.
• The technique is suitable for widespread adoption in hospitals lacking specialized ophthalmologists.