Research on disease diagnosis based on teacher-student network and Raman spectroscopy.

Area of Science:

• Biomedical Spectroscopy
• Machine Learning in Healthcare
• Computational Biology

Background:

• Diabetic nephropathy and primary Sjögren's syndrome are significant health concerns.
• Traditional Raman spectral feature selection methods alter original data characteristics.
• Need for advanced diagnostic tools for complex diseases.

Purpose of the Study:

• To develop a Raman spectroscopy-based diagnostic method using a teacher-student network.
• To perform feature selection while preserving original data characteristics.
• To improve diagnostic accuracy for diabetic nephropathy and primary Sjögren's syndrome.

Main Methods:

• Utilized a teacher-student network architecture with Raman spectroscopy.
• Employed five flow learning models as the teacher network and a neural network as the student network.
• Multi-layer perceptron was used for classification, with feature selection based on accuracy, precision, recall, and F1-score.

Main Results:

• Achieved optimal accuracy rates of 98.3% for diabetic nephropathy diagnosis.
• Achieved optimal accuracy rates of 100% for primary Sjögren's syndrome diagnosis.
• Demonstrated significant accuracy improvements over existing research (14.02% and 10.48% higher, respectively).

Conclusions:

• The combination of Raman spectroscopy and teacher-student networks is feasible for disease diagnosis.
• This approach effectively performs feature selection while retaining original data.
• The method shows high potential for accurate diagnosis of diabetic nephropathy and primary Sjögren's syndrome.