Keywords
welding, iron, manganese, machine learning
Select the category the research project fits.
Life Sciences
Is this submission part of ICaP/PW (Introductory Composition at Purdue/Professional Writing)?
No
Abstract
Iron (Fe) is commonly found in elevated quantities in the human brain afflicted by neurodegenerative diseases. While it is unknown how Fe plays in the etiologies of these diseases, welders inhale large quantities of metal particulates in welding fume, including iron (Fe) and manganese (Mn). Mn is a neurotoxin that has been shown to accumulate in the brains of welding fume exposed workers by increasing the magnetic resonance imaging (MRI) R1 contrast. R1 and R2* are MRI parameters that are proportionate to Mn and Fe accumulation, respectively. Measurements of Mn in the brain could be confounded by accumulated Fe, mostly altering R2* contrast, but also R1 contrast to some degree. Therefore, monitoring the amount of Fe accumulating in welders is of consequential interest. While some groups, including our own, have reported increased R2* levels in region-of-interest (ROI) based analyses, such findings were inconsistent and targeted few brain regions. To enable an unbiased whole-brain analysis of Fe accumulation in the brain using MRI, 47 welders and 38 controls were recruited from a local manufacturer. Whole-brain R2* maps were co-registered with T1-weighted structural images using SPM12 and then segmented into 192 different brain regions using Freesurfer. R2* in these segmented ROIs within the brain (e.g. white matter tracts and basal ganglia nuclei) were separately averaged and compared between welders and controls. Student’s t-tests showed no statistically significant differences between controls and welders. Therefore, a more comprehensive analysis using machine learning was used to determine if any patterns using all 192 regions could discriminate between controls and welders. Principle component analysis (PCA) was performed on five different statistics of R2* distributions in each ROI: mean, median, skew, 90thpercentile, and maximum value. For example, PCA performed on R2* mean showed that only 32 principle components (PCs) were required to explain 90% of all variation in mean between all 192 ROIs. A support vector machine (SVM) with a linear kernel was employed using these 32 PCs but could not distinguish between welders or controls better than chance. Similar results were found for the other four statistics. These null results suggest that R2*, and thus brain Fe accumulation, cannot distinguish welders from controls. This provides some evidence that measures of Mn accumulation shown in previous work in the same cohort is only caused by elevated Mn brain levels and not confounded by elevated iron levels.
Recommended Citation
Davis, Jennifer, "Whole-Brain Approaches for Investigating Iron Accumulation by R2* show no Excess from Occupational Exposure to Welding Fumes" (2019). Purdue Undergraduate Research Conference. 2.
https://docs.lib.purdue.edu/purc/2019/Oral_Presentations/2
Whole-Brain Approaches for Investigating Iron Accumulation by R2* show no Excess from Occupational Exposure to Welding Fumes
Iron (Fe) is commonly found in elevated quantities in the human brain afflicted by neurodegenerative diseases. While it is unknown how Fe plays in the etiologies of these diseases, welders inhale large quantities of metal particulates in welding fume, including iron (Fe) and manganese (Mn). Mn is a neurotoxin that has been shown to accumulate in the brains of welding fume exposed workers by increasing the magnetic resonance imaging (MRI) R1 contrast. R1 and R2* are MRI parameters that are proportionate to Mn and Fe accumulation, respectively. Measurements of Mn in the brain could be confounded by accumulated Fe, mostly altering R2* contrast, but also R1 contrast to some degree. Therefore, monitoring the amount of Fe accumulating in welders is of consequential interest. While some groups, including our own, have reported increased R2* levels in region-of-interest (ROI) based analyses, such findings were inconsistent and targeted few brain regions. To enable an unbiased whole-brain analysis of Fe accumulation in the brain using MRI, 47 welders and 38 controls were recruited from a local manufacturer. Whole-brain R2* maps were co-registered with T1-weighted structural images using SPM12 and then segmented into 192 different brain regions using Freesurfer. R2* in these segmented ROIs within the brain (e.g. white matter tracts and basal ganglia nuclei) were separately averaged and compared between welders and controls. Student’s t-tests showed no statistically significant differences between controls and welders. Therefore, a more comprehensive analysis using machine learning was used to determine if any patterns using all 192 regions could discriminate between controls and welders. Principle component analysis (PCA) was performed on five different statistics of R2* distributions in each ROI: mean, median, skew, 90thpercentile, and maximum value. For example, PCA performed on R2* mean showed that only 32 principle components (PCs) were required to explain 90% of all variation in mean between all 192 ROIs. A support vector machine (SVM) with a linear kernel was employed using these 32 PCs but could not distinguish between welders or controls better than chance. Similar results were found for the other four statistics. These null results suggest that R2*, and thus brain Fe accumulation, cannot distinguish welders from controls. This provides some evidence that measures of Mn accumulation shown in previous work in the same cohort is only caused by elevated Mn brain levels and not confounded by elevated iron levels.