With these projects, we want to investigate whether computerized speech analysis can be used to track the motor, emotional and cognitive fluctuations that a person with Parkinson’s disease (PD) can experience throughout the day. Moreover, it can also allow the early identification of motor, cognitive, or emotional side effects induced by medication or DBS, and thus preventing these side effects to worsen over time.
In the past, PD was seen mostly as a motor disease, in which slowness of movements, tremor or rigidity were the main symptoms. However, we now know that non-motor symptoms and in particular neuropsychiatric symptoms, such as anxiety, depression or apathy, actually have at least as much impact on the quality of life as the motor symptoms, if not higher. Therefore, it is extremely important to take into consideration not only the motor symptoms but also the neuropsychiatric symptoms of the disease while treating this disorder.
Dopaminergic replacing therapies are the mainstay of treatment and can give patients a very good quality of life and good control of symptoms during several years.
However, with the progression of the disease persons with PD (PwP) can start to experience fluctuations of their symptoms during the day, since the effects of medication start to last not as long as before. This means that the same person can alternate throughout the day between two completely different states. When medication is not working the person can be slow, with tremor, muscle stiffness, anxiety, low mood, lack of initiative and slowness of thinking, and when medication is working at its peak this can change to dyskinesia (involuntary movements ), euphoric mood, hypomanic behavior and difficulties to organize thoughts. This can happen several times during the day, and both ends of this spectrum can be very disabling.
When fluctuations starts to happen, the success of the treatment relies mostly on the optimum titration of the medication according to patient’s symptoms to try to avoid these two extremes. However so far we do not have an objective way to reliably track the symptoms of the disease, and in clinical practice the therapeutic decisions are based mostly on the person’s or/and caregiver’s recall of the symptoms in the previous days.
If fluctuations are not well controlled with oral medications, advanced treatments of PD are usually proposed. Within the advanced therapies, Deep Brain Stimulation (DBS) surgery is one of the most used treatments. In DBS, constant stimulation is delivered throughout the day and in this way, it is possible to smoothen symptom fluctuation. Although continuous delivering of stimulation is highly efficacious, it is now debated whether this is the best way to deliver stimulation. Two main reasons have been pointed out: 1) the symptoms of the disease are not always the same during the day, and 2) the person is not always doing the same type of activity. This means that sometimes we can be delivering too much stimulation that the person does not need, which also can increase the possibility of having side effects, and other times during more demanding activities, we may not be delivering enough stimulation.
Therefore, it is envisioned that in future smarter DBS devices should be capable to adjust in real time to the momentary needs of each specific person, while avoiding side effects. However, the best measurable indicators to be used, as feedback signals in these devices (also called adaptive-DBS) are not yet established.
Computerized speech analysis is a very promising way to measure fluctuations of PD since from speech we can extract not only motor information, but also emotional and cognitive content. Speech is easy to collect, its recording does not require laborious and time-consuming procedures. Therefore, we believe that computerized speech analysis might be a very good way to optimize medication of PD and to adjust DBS systems.
Additionally, machine-learning techniques are being increasingly applied in the healthcare sector. As its name implies, machine learning allows for a computer program to learn and extract meaningful representation from data in a semi-automatic manner. This would allow the integration of different types of information that can be used to individualize treatments of PD. In other words using machine learning, in the future we hope to decode for example speech data, brain signaling information and movement data from wearables, generating a response that supports the patient exactly at the time it is required, and taking into consideration not only the motor domain, but also the emotional and cognitive domains.
- Prof. Tobias Nef, ARTORG Center, University of Bern
- Prof. Jan Rusz, Czech Technical University of Prague