An intelligent hemodynamic response analysis method to achieve prognosis and diagnosis of Huntington’s disease

Abstract

The development of a reliable and cost-effective Huntington’s disease (HD) detection is a challenging task due to non-specific clinical first symptoms. To address the challenge, this is the first study to comprehensively focus on proposing an automated HD detection system based on functional near-infrared spectroscopy (fNIRS) analysis through a standard decomposition technique and dynamic mapping neural networks. fNIRS is a highly cost-effective and more refined neuroimaging modality that noninvasively measures hemodynamic responses and neurovascular coupling mechanisms. Considering the non-stationary nature of the hemoglobin concentration changes, the proposed system has developed a new fNIRS-based biomarker of HD, namely time-varying singular value, to characterize the spatiotemporal characteristics of the oxyhemoglobin and deoxyhemoglobin signals. The classification has been performed using a support vector machine, recurrent neural network, and cascade forward neural network to discriminate healthy controls (HC) from presymptomatic (Pre-HD) or symptomatic HD (SHD) subjects. Moreover, in a comparative study, the effects of trajectory matrix size, clinical categories of HD, type of chromophores, and brain regions have been tested on the detection performance, separately.

To evaluate the proposed system, the fNIRS dataset of 12 Pre-HD, 15SHD, 29 HC for Pre-HD, and 33 HC for the SHD has been used. The method has achieved remarkable accuracy rates of 95.61% for Pre-HD vs. HC and 95.63% for SHD vs. HC. The comparative analysis leads to the outstanding performance of this system and its high robustness against affecting factors, providing a better trade-off between computational costs and detection performance.