How Picture Description Tasks Can Help Track Cognitive Health Remotely

Speech as a digital biomarker in clinical research

Changes in how we speak—how quickly we talk, how often we pause, and the words we choose—can be early signs of cognitive decline. People with mild cognitive impairment (MCI) or dementia due to Alzheimer’s disease (AD) often show subtle shifts in their speech patterns, such as slower speech, longer pauses, and changes in pronoun or noun use.^1,2,3

Researchers have been looking for ways to measure these changes in natural, everyday-like situations. One promising approach is the picture description task—a short, low-burden activity where participants describe what they see in an image. The task can be done remotely, captures naturalistic speech, and is well-suited for repeated testing over time.

DEFEAT-AD collaboration

Researchers at Amsterdam UMC led the DEFEAT-AD study to collect high-frequency speech assessments remotely in a sample of 50 cognitively unimpaired individuals. Participants completed a 5-day “burst” of testing using a tablet and the Winterlight Assessment app including two picture description tasks per day for five consecutive days. In the repeating condition, the same picture was shown every day. In the alternating condition, a different picture was shown each day. After a 2–3 week break, participants repeated the same 5-day burst with identical stimuli.

How reliable are speech measures?

In order for speech measures to be able to be used in higher frequency testing, we need to determine their reliability and susceptibility to practice or learning effects when these tasks are completed repeatedly over several days.

We found that most of our selected speech measures had acceptable test-retest reliability over the five days of testing, and it didn’t matter much if the same or different picture was shown. Some measures, including total words produced and number of objects described, showed some evidence of practice/learning effects, with increasing scores over the first few days.

These results show that picture description tasks can be a reliable way to track speech changes over time in a high-frequency, remote testing setup. Alternating pictures or including a practice phase before data collection may help reduce learning effects. Because picture description tasks are ecologically valid, easy to administer, and low burden, they are promising tools for long-term monitoring of cognitive health—especially in remote or decentralized clinical trials.