Comprehensive Summary
Xiao et al. conducted a study investigating the effectiveness of serum Raman spectroscopy in schizophrenia screening. Serum Raman spectra are valuable in their accurate and rapid diagnosis using bodily fluid samples, here using serum centrifuged from blood samples of 48 schizophrenia patients and 66 healthy patients. Samples were pre-processed to mitigate interfering factors, rapidly screen samples, and ensure the model’s detection accuracy and stability. Pre-processed Raman spectra from schizophrenia and healthy patients had similar peak distributions, however, the intensities across the graph were lower in schizophrenia patients compared to healthy patients. Xiao et al. innovatively applied Markov transition fields (MTF) in Raman spectra to produce 2D spectrograms which contain more features than 1D spectral sequences and align better with convolutional neural networks (CNN), here used to classify pre-processed Raman spectroscopy. CNNs allow for more rapid screening of schizophrenia patients and in this study, Xiao et al. used four CNNs to analyze Raman spectra, comparing the effectiveness of MTF to 1D sequences. The authors found that inputting MTF images produced better sensitivity across all models compared to a 1D sequence input, and further, that two of the models produced significantly improved model performance with MTF input.
Outcomes and Implications
Xiao et al.’s work provides valuable insight into a potential methodology for a non-invasive, rapid screening that has potential for facilitating earlier detection and intervention in schizophrenia across widespread settings. However, patient samples for this study were obtained from a single institution and clinical challenges in obtaining them limit the study’s generalizability. The authors performed a 5-fold cross-validation to attenuate the impact of this on the model, yet acknowledged the need for increased sample size to improve the model’s reliability. Further this model is only a precursory validation of the method’s applicability and cannot be used in clinical diagnosis. Nevertheless, the authors indicate that they will continue to expand the study to better schizophrenia screening methods.