Early Detection of Dementia

A four-year project involving ten partners from academia and industry is currently underway to develop an integrated system for early diagnosis of dementia. Known as Dem@Care, the goal is to increase accuracy and affordability of detection through the analysis of voice patterns.

Dementia is defined as the loss of memory, cognitive skills and other brain functions beyond what is considered a normal part of the aging process. Currently, diagnosis is generally based on limited clinical observation during short visits to the doctor. This not only makes it difficult to thoroughly assess the patient, but also limits the ability of clinicians to develop effective and individualized preventive strategies.

The new system is based on past research which indicates that certain neurological and mental disorders, as well as diseases like laryngeal cancer, manifest themselves as changes in voice features such as quality, pitch, jitter, continuity, fluency, semantics and richness of vocabulary. Designed to be used in the home, it presents subjects with a series of questions adapted from neuropsychological tests, while a combination of sensors – including a microphone or telephone, accelerometers, video cameras and physiological sensors – records their responses for analysis on IBM’s Watson computer.

Experiments have shown the machine learning algorithms to be 85 percent accurate at categorizing patients into one of three groups – normal, mildly impaired, or positive for early dementia. Although it is not a formal diagnosis, the system can help clinicians better manage symptoms and gauge treatment based on behavior patterns. For example, if a patient reveals signs of apathetic behavior, medications and/or activity levels may be adjusted accordingly without requiring an office visit.