A fascinating new study from a team of US researchers has used machine learning techniques to develop algorithms that can analyze naturalistic driving data and detect mild cognitive impairment and dementia in a driver. The work is still in the preliminary stages, however, the researchers claim it could be possible in the future to detect early signs of dementia using either a smartphone app or devices incorporated into car software systems.
The influence of dementia on driving behavior is a reasonably well-studied topic. It is certainly unsurprising to observe driving behaviors change as neurodegeneration leads to cognitive decline. However, this new research set out to explore whether machine learning techniques could be used to identify patterns in driving data that can then detect either mild cognitive impairment (MCI) or dementia.
The research utilized data from a novel long-term study called LongROAD (The Longitudinal Research on Aging Drivers), which tracked nearly 3,000 older drivers for up to four years, offering a large longitudinal dataset.
Over the course of the LongROAD study, 33 subjects were diagnosed with MCI and 31 with dementia. A series of machine learning models were trained on the LongROAD data, tasked with detecting MCI and dementia from driving behaviors.
“Based on variables derived from the naturalistic driving data and basic demographic characteristics, such as age, sex, race/ethnicity and education level, we could predict mild cognitive impairment and dementia with 88 percent accuracy,” says Sharon Di, lead author on the new study.
Although age was the number one factor for detecting MCI or dementia, a number of driving variables closely followed. These include, “the percentage of trips traveled within 15 miles (24 km) of home … the length of trips starting and ending at home, minutes per trip, and number of hard braking events with deceleration rates ≥ 0.35 g.” Using driving variables alone, the models could still predict those MCI or dementia drivers with 66 percent accuracy.
It’s still early days for the work, with the researchers saying more investigation is needed to specifically understand the differences in driving characteristics between MCI and dementia. Plus, the study is clear in noting the modest number of MCI and dementia cases in the LongROAD data means larger investigations will be necessary to find out how broadly generalizable the prediction models are in real-world settings.
Nevertheless, the study does point to intriguing future scenarios where a smartphone app, or software inside a car, can constantly monitor your driving patterns to offer clues for early detection of cognitive decline before clinical symptoms become apparent. Of course, this all assumes we will still be actively driving our cars in the future.
“Our study indicates that naturalistic driving behaviors can be used as comprehensive and reliable markers for mild cognitive impairment and dementia,” adds senior author Guohua Li. “If validated, the algorithms developed in this study could provide a novel, unobtrusive screening tool for early detection and management of mild cognitive impairment and dementia in older drivers.”
The new study was published in the journal Geriatrics.
Source: Columbia University Mailman School of Public Health
'Quick hard braking' - not sure which way that would pan. Quick sharp braking could indicate fast cognitive ability and reflexes to an external event - or it could indicate the inability to perceive threats ahead on the road and braking at the last moment.
Also depends on where you are driving, Paris, Rome (always slamming on the brakes is the driving style!) or a quite rural road in England in the summertime!
Excluding demented city drivers, unexpected braking does tend to be a trait of older drives who are experiencing some cognitive decline, although at the same time it could be due to less practice and nervousness (many learner drivers do the same).
I have a friend who had a stroke a couple of years ago, and was driving until quite recently. I observed, while riding as a passenger, that his perceptions of the traffic conditions had become limited. He was much more likely to accelerate suddenly, braked late and hard, and would occasionally drift into adjacent lanes. I was forced on occasion to criticize his driving when I felt that he was putting us into danger- as much as I dislike being a "backseat driver".
I do not believe that he should be driving anymore, and will tell him that when I get an opportunity.
To a certain extent this is already being used by insurance companies with their reduced rates for drivers with black boxes that monitor speed, cornering, braking etc. Would the AI analysis even assist the black box predictive power?