Recognition of chronic renal failure based on Raman spectroscopy and convolutional neural network

Chronic renal failure (CRF) is a disease with a high morbidity rate that can develop into uraemia, resulting in a series of complications, such as dyspnoea, mental disorders, hypertension, and heart failure. CRF may be controlled clinically by drug intervention. Therefore, early diagnosis and control of the disease are of great significance for the treatment and prevention of chronic renal failure. Based on the complexity of CRF diagnosis, this study aims to explore a new rapid and noninvasive diagnostic method.

In this experiment, the serum Raman spectra of samples from 47 patients with CRF and 53 normal subjects were obtained. In this study, Serum Raman spectra of healthy and CRF patients were identified by a Convolutional Neural Network (CNN) and compared with the results of identified by an Improved AlexNet. In addition, different amplitude of noise were added to the spectral data of the samples to explore the influence of a small random noise on the experimental results.

A CNN and an Improved AlexNet was used to classify the spectra, and the accuracy was 79.44 % and 95.22 % respectively. And the addition of noise did not significantly interfere with the classification accuracy.

The accuracy of CNN of this study can be as high as 95.22 %, which greatly improves its accuracy and reliability, compared to 89.7 % in the previous study. The results of this study show that the combination of serum Raman spectrum and CNN can be used in the diagnosis of CRF, and small random noise will not cause serious interference to the data analysis results.