Sekundarni povzetek: |
In recent years, driving simulators have emerged as a promising alternative to contemporary driving ability assessment methods of neurological patients, as they offer an opportunity for a fast, standardized, and ecologically valid evaluation. However, despite growing research on the driving characteristics of neurological patients, no study has so far compared patients deemed fit and unfit to drive. To inform the development of future simulator tools for assessment and rehabilitation, the current study compared driving characteristics of patients, who were, based on a standard procedure in a competent rehabilitation facility, found to be fit, conditionally fit, or unfit to drive.
The study included 95 patients with various neurological diseases, participating in a comprehensive driver rehabilitation program, which combines clinical, functional, neuropsychological, and on-road assessment. The subjects drove through three high-risk scenarios in a driving simulator, simulating rural, highway, and urban environments. For each scenario, various descriptive variables were calculated from the driving data, describing reaction times, vehicle control, traffic rule compliance, and eye-tracking characteristics.
Group comparison using analysis of variance revealed significant differences in reaction times between the fit and the unfit group, regardless of the scenario. On the highway, the groups significantly differed in the variability of steering wheel angle, steering wheel reversal rate, turn signal neglect rate, and the use of the right side-view mirror. In the city, they differed in lane position variability, speeding rate, and the number of accidents. In some of the listed scenario-variable combinations, differences were also observed between the unfit and conditionally fit group, while the fit and the conditionally fit group only differed in the use of the rear-view mirror on the highway. The driving parameters were then used to train support vector machine classifiers. The best-performing models correctly classified 59% of drivers in the multiclass and 82% in the binary task, where only the fit and unfit drivers were classified.
The results show that driving simulators can indeed capture the differences in driving characteristics of neurological patients with different driving abilities. Except for reaction times, no variable exhibited significant differences in more than one scenario, which points to the importance of carefully designing the environments to best suit the desired measures of driving performance. The moderately successful performance of classification models indicates that the selected scenarios are not optimal for driver evaluation, but in terms of future development of such tools, the results are nonetheless promising. The study finishes by discussing ways to improve simulator-based methods and provides guidelines for their further development. |