Breaking the Silence on Alzheimer’s Dementia
Diagnosing Alzheimer's dementia usually requires broad clinical testing, but this ML algorithm that can run on a phone simplifies the job.
Alzheimer's dementia is a progressive neurodegenerative disorder that primarily affects older adults, although it can also occur in younger individuals. It is the most common cause of dementia, accounting for approximately 60-80% of cases. The cause of this condition is believed to be an accumulation of two types of abnormal protein structures in the brain: beta-amyloid plaques and tau tangles. These deposits lead to the impairment of neuronal communication and eventual brain cell death.
The impact of Alzheimer's dementia on patients is profound and extends to their families and the healthcare system as a whole. As the disease progresses, individuals with Alzheimer's often experience difficulties with memory, thinking, and problem-solving. They may struggle with simple daily tasks and lose their ability to recognize familiar people and places. Behavioral changes such as agitation, aggression, and mood swings are also common. The loss of independence and cognitive decline require ongoing care and support from caregivers, resulting in increased emotional, physical, and financial burdens.
Traditionally, diagnosing Alzheimer's dementia involves a comprehensive evaluation of an individual's medical history, cognitive functioning, and physical and neurological examinations. Typically, a thorough assessment is conducted to exclude other possible causes of cognitive impairment. Neuropsychological tests are employed to evaluate memory, attention, language, and executive functions. Brain imaging techniques, such as magnetic resonance imaging and positron emission tomography, may be used to assess brain structure and detect abnormal protein accumulations. In some cases, cerebrospinal fluid analysis can provide additional information about beta-amyloid and tau proteins.
Without a doubt, diagnosing this condition is time-consuming, expensive, and requires the input of highly trained medical professionals. Researchers at the University of Alberta and the Institute for Language and Speech Processing took note of the fact that Alzheimer's dementia patients generally present with a number of abnormal speech patterns, namely, they speak abnormally slow, frequently pause or hesitate, and they speak with reduced intelligibility. Leveraging this knowledge, they designed machine learning algorithms that can diagnose the condition from a voice sample with a reasonably good level of accuracy. And they can do so on minimal hardware, like a smartphone.
To ensure that their model was well generalized and truly capable of detecting pathological patterns of speech, they trained it with recordings of English speakers describing the contents of a picture, but validated it with recordings of Greek speakers describing the contents of a different picture. A dataset of 237 age- and gender-balanced audio files was acquired for this purpose.
To capture information about the most common types of speech problems that present, features were extracted that encode speech rate, pauses, and intelligibility. Multiple dimensionality reduction techniques (e.g. Principal Component Analysis, Latent Semantic Analysis) were tested to optimize the analysis pipeline. Several classification and regression models were also evaluated to find the most accurate algorithm for the job. The best performing classifier was a logistic regression model with L2-regularizer using the top 10 PCA components. This classifier yielded a nearly 75% average accuracy rate. A support vector machine, with a radial basis function kernel and regularization parameter set to one, proved to be the best regression model. The root mean square error of this algorithm was 6.487 ± 0.696.
The algorithms will need some improvement to be trusted in making clinical diagnoses, but this work shows that much of the expense and other barriers that prevent many early diagnoses from happening might be done away with in the future. One day, a person’s smartphone could alert them that they may be at risk for Alzheimer's dementia based on their telephone conversations, or a simple test performed with a diagnostic app.