Cerebral grey matter volume identifies healthy older drivers with a critical decline in driving safety performance using actual vehicles on a closed-circuit course

Handityo Aulia Putra¹Kaechang Park^2,3*Fumio Yamashita⁴

• ¹Nagaoka University of Technology, Nagaoka, Japan
• ²Kochi University of Technology, Kami, Kochi, Japan
• ³Kōchi University, Kochi, Kōchi, Japan
• ⁴Iwate Medical University, Morioka, Iwate, Japan

The study utilized the Random Forest machine learning method to identify healthy older drivers with a critical decline in driving safety performance (DSP) using cerebral grey matter (GM) volume. Ninety-four older drivers (45 males, 49 females; mean age, 77.66 ± 3.67 years) without dementia participated in the study, with their DSPs evaluated using actual vehicles on a closed-circuit course. The DSP was scored in six categories: DSP1, visual search behavior; DSP2, speeding; DSP3, signaling of the indicator; DSP4, vehicle stability; DSP5, positioning; and DSP6, steering. The total scores were calculated by a driving instructor; larger scores indicated safer driving performances. A score below 15% of the total was defined as a critical decline in DSP. Regional GM volumes were measured with voxel-based morphometry by magnetic resonance imaging (MRI). From 114 GM regions, eleven regions were selected under the optimization for the machine learning method. These GM regions were the left angular gyrus, frontal operculum, occipital fusiform gyrus, parietal operculum, postcentral gyrus, frontal cortex, planum polare, superior temporal gyrus, and the right hippocampus, orbital part of the inferior frontal gyrus, posterior cingulate gyrus, and posterior orbital gyrus. Some of them are involved in attention and spatial cognition, higher processing of visual information, and somatosensory processing, which are necessary for DSP execution. Although the accuracy and specificity of predictive performance are satisfyingly high, the other performances such as precision and recall/sensitivity are relatively low for implementation in the real world. In addition to the structural factor of regional GM volume, other factors such as neuronal connectivity evaluated by functional MRI (fMRI) should be further investigated. Since there is currently room for improvement with MRI in predictive performance, MRI's measurement may help deepen the understanding of the neuronal basis linking the identification of dangerous drivers and subsequently contribute to preventing traffic crashes by eliminating them.