Patch-type brain layer analyzer using neural network assigned high density near- infrared spectroscopy

Abstract

It has been known that near-infrared spectroscopy (NIRS) technology is not suitable for extracting brain layer information due to scattering nature of the photons. This fact prevents the application of the NIRS devices to the accurate diagnosis of brain diseases. This study proposes a brain layer analyzer that extracts the information of brain layer selectively by using AI-assisted NIRS technology. The proposed system collects high density reflectance of NIR lights and incorporates a neural network for the interpretation. Simulation results show that the proposed system demonstrates higher correlation ($R^2$=0.913) with the actual cortex oxygenation as compared to a conventional NIRS method ($R^2$ = 0.469). In addition, biomimetic phantom experiment demonstrates that the proposed system accurately measures the oxygenation of brain layer ($R^2$=0.986) irrespective of anatomical variations unlike conventional NIRS devices $R^2$=0.823). In the prospective observational study, the proposed system demonstrated the classification accuracy of 0.943 AUC between the healthy subject and the stroke patients.