NIH Research Festival
–
A two-stream convolutional neural network (TCNN) for breathing pattern classification has been devised for continuous monitoring of patients with infectious respiratory diseases. We have applied this method to classify different breath patterns using tissue hemodynamic responses collected from 14 participants. The hemodynamic responses were measured using a wearable near-infrared spectroscopy (NIRS) device, which was placed on the chest of a participant. Four breathing patterns, namely normal, slow, rapid, and breath holding, were included in the experiment. We hypothesize that tissue hemodynamic responses measured with our NIRS device are sensitive to respiratory changes, and these changes can be quantified by correlating to the changes in oxy- and deoxyhe-moglobin concentrations at the chest wall tissue. The TCNN consists of a convolutional neural network (CNN) based autoencoder and classifier. The encoder of the autoencoder generates deep compressed feature maps, which contain the most important information constituting the data. These generated maps are concatenated with the feature maps generated by the classifier to classify breathing patterns. The TCNN is designed by improving and modifying the pre-activation residual network (Pre-ResNet) developed for 2-dimensional image classification. The proposed TCNN overcomes the problem of decreasing learning performance as the layers deepen in the single-stream classification model. Compared with a single-stream classification model without an autoencoder, the proposed TCNN method demonstrated approximately 2.6% higher classification performance. Furthermore, the proposed method achieved the highest classification accuracy of 94.63% com-pared to state-of-the-art classification models.
Scientific Focus Area: Microbiology and Infectious Diseases
This page was last updated on Tuesday, August 6, 2024