MIT’s new AI model can successfully detect Parkinson’s disease
MIT researchers have developed an early-research artificial intelligence model that has demonstrated success in detecting Parkinson’s disease from breathing patterns. The model relies on data collected by a device that detects breathing patterns in a contactless manner using radio waves.
Neurological disorders are some of the leading sources of disability globally and Parkinson’s disease is the fastest-growing neurological disease in the world. Parkinson’s is difficult to diagnose as diagnosis primarily relies on the appearance of symptoms like tremors and slowness but these symptoms usually appear several years after the onset of the disease.
The model also estimated the severity and progression of Parkinson’s, in accordance with the Movement Disorder Society Unified Parkinson’s Disease Rating Scale (MDS-UPDRS), which is the standard rating scale used clinically. The research findings have been published in the journal Nature Medicine.
The researchers trained the model by using nocturnal breathing data (data collected while subjects were asleep) from various hospitals in the US and some public datasets. After training the model, they tested it on a dataset that was not used in training, and discovered it diagnosed Parkinson’s disease with an accuracy of about 90 per cent when it analyses one night’s sleep worth of data from a patient. They found that the model’s accuracy improves to 95 per cent when it analyses sleep data from 12 nights.
The relationship between Parkinson’s and breathing has been known since 1817, as observed by James Parkinson in his research. There has also been previous research into how Parkinson’s patients develop sleep breathing disorders, weakness in the function of respiratory muscles, and degeneration in brainstem areas that control breathing.
While MIT’s model is promising, it is still in an early stage of development. “Although the datasets are all from the United States, we note that the datasets have different races and ethnicities. However, we believe it is desirable to validate the model further on more diverse datasets from other countries. We would love to collaborate with medical institutes in India and other countries to extend the research to those communities,” Dina Katabi, co-author of the paper, told indianexpress.com over email.
Currently, the AI model is tested using data from either a wearable breathing belt used during polysomnography (sleep study) or a specialised device developed by Katabi and other researchers called the “Emerald Radio Device.” You can see a live demonstration of how the device can capture breathing patterns below.
But there is a possibility of using data from other devices that capture breathing data accurately enough. “Any device that can accurately obtain breathing signals would be suitable for use with our AI model. Smartphones today do not obtain a sufficiently accurate breathing signal, but it is perceivable that they can do it. Either way, obtaining accurate breathing signals is easy with either our radio device or a breathing belt,” added Katabi.
There is currently no cure for Parkinson’s disease, but Katabi envisions that using this technology for diagnosis could significantly shorten clinical trials for potential treatments, which could accelerate their development.
Also, this model could potentially be used for assessment in underserved communities, especially for those living in areas without significant medical access, and for patients who have difficulty leaving their homes due to the progression of the disease. The researchers also believe that the work and technology could potentially be extended to detect other neurological diseases, like Alzheimer’s disease. “But more research and experimentation need to be done before reaching a firm conclusion on whether that is possible,” commented Katabi.