Detection of Chylous Plasma Based on Machine Learning and Hyperspectral Techniques

Area of Science:

• Medical technology
• Biomedical engineering
• Spectroscopy

Background:

• Chylous blood is a primary reason for discarding donated blood.
• Accurate and rapid pre-donation screening for chylous blood is essential.

Purpose of the Study:

• To develop and validate a rapid diagnostic method for identifying chylous plasma using hyperspectral imaging and machine learning.
• To assess the effectiveness of different machine learning algorithms for chylous plasma classification.

Main Methods:

• Utilized the Gaia hyperspectral sorter to capture plasma images across 254 bands (900-1700 nm).
• Applied and compared four machine learning algorithms: decision tree, Gaussian Naive Bayes (GaussianNB), perceptron, and stochastic gradient descent.
• Performed feature dimension reduction on spectral data to optimize classification accuracy.

Main Results:

• Decision tree and GaussianNB models achieved over 90% accuracy with original data.
• Decision tree model demonstrated superior performance with 93.33% accuracy after feature dimension reduction.
• The decision tree model consistently showed over 80% accuracy for chylous plasma classification before and after feature reduction.

Conclusions:

• Hyperspectral imaging combined with machine learning provides a rapid and effective method for identifying chylous plasma.
• This approach has the potential to significantly reduce blood resource wastage and enhance medical staff efficiency.
• Characteristic spectral bands identified offer potential for robust plasma classification and identification.