Using machine learning for early screening and detection of Alzheimer’s disease

Early diagnosis of dementia patients enables early intervention that could potentially prevent or delay disease onset. Alzheimer’s disease (AD) is the most common form of dementia. Despite the importance of early diagnosis, 3 in every 4 individuals with dementia have not been diagnosed yet. One possible way to overcome this problem is to devise a screening method that could be implemented in non-clinical settings.

Drawing tests are commonly used in clinical practice to screen and diagnose AD. The changes in an individual's drawing capability represent a sensitive indicator of AD and mild cognitive impairment (MCI), an early stage of dementia. An example of the drawing tools is the trail-making test, which measures the person’s processing speed and drawing ability in terms of task completion time. In clinical settings, multiple tests are utilized to enhance AD screening and diagnosis. The outcome of the drawing tests requires evaluation by clinical specialists, which may act as a barrier to early screening. Therefore, developing automated drawing tests in non-specialist settings could mend the gap.

Kobayashi et al. used several machine-learning algorithms to diagnose AD. The sample consisted of 144 participants, including patients with AD, MCI, and cognitively normal (CN) individuals. The subjects underwent seven cognitive assessment tests that evaluated global and specific cognitive domains. Then, the participants took five drawing tasks, which were later used to evaluate cognitive measures using machine learning. The objective was to determine whether the drawing features extracted from different tasks were associated with diagnostic status and cognitive measures. Also, researchers assessed the feasibility of utilizing a five-task combined model versus a single-task model in identifying and evaluating cognitive performance.

Results revealed that the model combining features from all the five drawing tasks outperformed the models based on characteristics from a single task. Regarding diagnostic categories, the five-task model achieved an accuracy of 75.2% in assessing AD, MCI, and CN. This was 7.8% higher than the best single-task performance. Also, the combined model could differentiate between AD cases and CN with an accuracy of 96.8%, higher than any single-task model. Regarding MCI and CN differentiation, the combined model achieved an accuracy of 82.2%, which was 7.6% higher than any single-task model.

The authors concluded that using a multiple-assessment approach could potentially capture different, complementary aspects of cognitive impairment. This could be used to better screen and detect AD and MCI early on, using scalable, automated models.

‍

Source: Kobayashi M, Yamada Y, Shinkawa K, Nemoto M, Nemoto K, Arai T. Automated Early Detection of Alzheimer’s Disease by Capturing Impairments in Multiple Cognitive Domains with Multiple Drawing Tasks. Journal of Alzheimer's Disease. 2022 Jun 13(Preprint):1-5.

‍