A 70–75% Accurate AI Alzheimer’s Detection Model Using Smartphones

Once again, machine learning has discovered a fantastic use in healthcare. The difficulty of detecting Alzheimer’s has been met head-on this time.

Researchers working at the University of Alberta were working to develop a machine learning model that can detect dementia at a very early stage and can then flag the patient for the same. This machine learning engine, in the spotlight, can be accessed by low-end computational devices such as smartphones and can distinguish between Alzheimer’s patients and healthy individuals with a mean accuracy of  70-75%. It functions by analyzing the speaker’s speech pattern rather than focusing on what they are speaking. The tool can then provide crucial indicators, which can be used to provide better diagnosis and treatment of the disease.

Dementia caused by Alzheimer’s is a challenging task to intercept, especially in its early stages, where it matters the most because symptoms are very lowkey and can be mistaken for age-related memory issues. Early detection allows patients and doctors to attend to it sooner, minimizing the worst-case scenario.

Conventional methods of detecting brain changes associated with Alzheimer’s, such as lab work and medical imaging, are time-consuming, expensive, and typically not performed at early stages. Now, using a mobile phone and processing the speech input in such cases, that also at an early stage enhances and facilitates a better patient-physician relationship. This usage would lead to earlier treatment initiation and enable a possible simple intervention at home, which would help slow down the disease’s progression.

It is noteworthy to mention that this model does not aim to replace healthcare professionals; it rather aims to act as a tool, a telehealth service that will strive to provide a convenient way to identify potential concerns for patients who are facing geographical or language barrier and doesn’t have better facilities in their area. By triangulating the likely patients, healthcare providers can identify and prioritize identify and conditions reported by the markers.

The research group is focused on language-agnostic acoustic and linguistic speech features rather than vocabulary or specific words to develop the model, as concentrating on words can be misleading. Previous works involved analyzing the language used by Alzheimer’s patients, which posed computational challenges, specifically cross-language issues. The current approach emphasizes studying voice characteristics that transcend the language barrier. Patients with Alzheimer’s dementia tend to speak more slowly, experience more pauses or disruption in their speech, use shorter words, and have reduced intelligibility. Researchers have studied and translated these characteristics into speech features that the model can further attribute to analyze the condition.

The model itself is complex, but the user experience of the final tool incorporating it would be simple. Users would speak into the device, and it would analyze their speech and give out the result, whether they have Alzheimer’s or not. This information can then be shared with healthcare professionals, who can determine the best course of action for the individual. Although the model has been tested on English and Greek speakers, researchers are optimistic that this technology can be used across different languages with different dialects and tones.

There have been some previous works done around healthcare and technology which follow a somewhat similar approach. The computational psychiatry research group at the University of Alberta, led by Russ Greiner and Eleni Stroulia, has previously developed similar statistical models and tools for detecting mental health disorders such as PTSD, schizophrenia, depression, and bipolar disorder.

Any technical advancement would be appreciated in the healthcare field as it leads to better-informed decisions that are also within the appropriate amount of time and has the potential to bring down the cost of the healthcare service provided.


Leave a Reply

Your email address will not be published. Required fields are marked *