An algorithm for processing and analysis of Gas Chromatography-Mass Spectrometry (GC-MS) signals for early detection of Parkinson's disease

Abstract

Parkinson's disease (PD) is one of the most prevalent neurodegenerative disorders, affecting about 1% of the people over 65. Early detection of PD is important as it enables slowing or stopping its progression. Analyzing Volatile Organic Compounds (VOCs) in blood, which may be involved in biochemical pathways specific to the disease, may reveal biomarkers that could potentially assist in early detection of PD stage and progression. VOCs can be detected by using the Solid-Phase Micro Extraction (SPME) technique, together with Gas-Chromatograph Mass Spectra (GC-MS). The output of the GC-MS is a signal with multiple peaks, where the locations of the peaks and their areas represents the various VOCs and their corresponding concentrations. In this study we have developed an algorithm for automatic analysis and processing of GC-MS signals based on VOCs in blood samples. The algorithm detects all peaks (local maxima) in this signal, with the corresponding amplitude and retention time, and computes the areas of the peaks. The algorithm aligns and groups all peaks from different samples of a similar retention time, and normalizes the areas according to internal standard area. Based on the amplitude and area, only peaks whose peak-to-noise ratio is high enough and which are substantially different from blank peaks are considered for further analysis. Two groups of rats were considered for the experimental evaluation of the algorithm: a control group of sham rats (n=9), and a second group of pre-Parkinsonian rats (n=10). A supervised learning algorithm was employed on a partial set of features to evaluate the possibility of using the areas of the peaks in order to predict whether a sample was from an intact or an impaired rat. A 6-d feature vector was constructed using the areas of peaks with large difference between groups. Using a leave-one-out cross-validation, a correct identification rate of 84.2% was achieved. Although these results are preliminary, they may indicate the potential of using the GCMS signal obtained from blood VOCs for discriminating between intact and PD induced samples.