Integration of Raman Scattering and Machine Learning Methods for Improving the Sensitivity of Chronic Heart Failure Stage Diagnosis

Abstract

This study proposes the use of Raman scattering in combination with machine learning techniques to study chronic heart failure. The effectiveness of predictive machine learning models depends on the availability of large and diverse datasets, but spectroscopy still faces the challenge of insufficient data. To address this issue, we propose applying augmentation methods to artificially increase the volume and diversity of Raman spectra. The study included 62 blood serum Raman spectra from 22 patients with stage 1 chronic heart failure (CHF), and 89 blood serum spectra from 29 patients with stage 2 CHF. These initial 151 spectra were divided into training and testing sets in a 4:1 ratio. For the training set, various augmentation methods were applied. These included spectral-temporal techniques such as displacement along the wave number axis and stretching along the intensity axis, as well as the Wasserstein GAN (WGAN) method with augmentation coefficients of 5 and 10. In order to evaluate the effectiveness of artificially increasing the volume of spectral data, we built predictive binary models using the partial least squares discriminant analysis (PLS-DA) method. The results showed that there were significant improvements in the classification of stages 1 and 2 CHF cases with a 10-fold increase in spectral data using both spectral-temporal methods and the WGAN technique, achieving 0.81 and 0.83 ROC AUCs, respectively, compared to 0.71 for the initial spectral set. This work was supported by the Russian Science Foundation under Grant No. 25-75-10097, https://rscf.ru/project/21-75-00146/.

Speaker

Yulia Khristoforova
Samara National Research University
Russian Federation

Discussion

Ask question