A novel approach to smart-assisted schizophrenia screening based on Raman spectroscopy and deep learning.

Area of Science:

• Biomedical Spectroscopy
• Neuroscience
• Machine Learning in Healthcare

Background:

• Schizophrenia diagnosis remains challenging, necessitating novel biomarkers.
• Serum Raman spectroscopy offers a non-invasive approach for disease detection.
• Current spectral analysis methods may not fully capture complex spectral patterns.

Purpose of the Study:

• To develop an assisted screening method for schizophrenia using serum Raman spectra.
• To explore the utility of Markov transition fields (MTF) in analyzing Raman spectral data.
• To compare the diagnostic performance of 1D versus 2D spectral analysis for schizophrenia.

Main Methods:

• Collected serum Raman spectra from individuals with schizophrenia and healthy controls.
• Employed four convolutional neural networks for classification of spectral data.
• Introduced Markov transition fields (MTF) to convert 1D Raman spectra into 2D spectrograms.
• Developed and compared machine learning models based on 1D spectral sequences and 2D MTF spectrograms.

Main Results:

• The developed method demonstrated assisted screening capabilities for schizophrenia.
• Converting 1D Raman spectra to 2D spectrograms using MTF enriched spectral analysis.
• Models trained on 2D MTF spectrograms exhibited superior performance compared to those trained on 1D spectral sequences.

Conclusions:

• Serum Raman spectroscopy, enhanced by MTF-based 2D spectral analysis, shows promise as a screening tool for schizophrenia.
• The integration of MTF provides a novel approach to Raman spectral data processing.
• This method offers a potential advancement in the early detection and management of schizophrenia.