Artificial Intelligence (AI) has blurred the lines between science fiction and reality with self-driving cars, humanoid robots, and virtual assistants. Recent progress in this technology has made Google Assistant's newest voice almost indistinguishable from a human's.
Researchers and companies are actively using AI to solve challenges in healthcare. Last month,
new work from
Dr. Babak Taati and his colleagues used deep learning, an area of AI, to help neurologists improve the treatment process for Parkinson's disease.
Individuals affected by Parkinson's disease can have reduced mobility and can experience loss of physical control. These symptoms worsen as the disease progresses. Although a cure does not currently exist, the severity of the symptoms can be reduced through medication or surgery.
A common medication for treatment is levodopa, a molecule that our body produces as part of normal function. Many patients who take this medication over long periods experience muscle spasms and involuntary movements.
Changing the dosage of the drug to reduce the symptoms of Parkinson's disease without causing spasms is difficult. Another challenge in evaluating these side effects is that the process is subjective and varies according to the particular experience of the specialist. Because of this, the specialist's assessment can differ considerably from what patients experience between visits.
To address this issue, Dr. Taati and his team captured series of short videos of patients after they received infusions of levodopa, and used the deep learning algorithm to track their body to measure the severity of the spasms and involuntary movements. The research team found that the AI algorithm performed as well or better than experienced neurologists.
"Our AI algorithm was able to objectively and accurately detect the onset and the remission of the spasms and involuntary movements and agreed with what the patients reported," explains Dr. Taati.
"Now that we know such an approach is feasible for Parkinson's disease, the next step is to validate it in more people and to improve the algorithm. The ultimate goal is to develop a clinical tool to help doctors design more effective treatments and minimize their side effects."
This work was supported by the Natural Sciences and Engineering Research Council of Canada, the Toronto Rehab Foundation, and the Toronto General & Western Hospital Foundation.
