After a decades-long career in IT, Warwick Adams switched to winemaking. But about 10 years into his new career, he noticed he was losing his sense of smell and struggling with his handwriting. When those issues began to interfere with his work, he sought medical advice. He visited a range of specialists over the next two years before a neurologist finally told him he had Parkinson’s disease. “It was a real shock,” he said. “Of all the things I thought may have been my problem, Parkinson’s didn’t even enter my brain.”
Once the initial surprise wore off, Adams started to wonder why it had taken so long for him to get a diagnosis for a relatively common disease. “I did a little bit of research and discovered that there’s no quantitative test,” he said. “It relies entirely on the skill of the clinician.”
Having spent so many years working with computers, Adams wondered if a typing test might offer a simple, standardized way to check for fine motor impairments that could indicate Parkinson’s disease. So, he enrolled as a doctoral student at Charles Sturt University to test his hypothesis.
In doing so, Adams joined a growing group of researchers who are trying to develop software and hardware tools that can turn technology that people regularly interact with into effective approaches for detecting early deterioration in fine motor control. Recently, engineers at the University of California, Los Angeles developed a pressure-sensitive keyboard that can gather even more information than the commercially available options people have in their homes. But how exactly these tools will fit into the clinical landscape remains an open question.
The challenge of diagnosing Parkinson’s disease
Adams’ long diagnostic journey is an all-too-common common experience, according to Isabelle Buard, a neurophysiologist at the University of Colorado. Parkinson’s disease can manifest in a variety of ways, including both nonmotor symptoms, like the loss of smell, constipation, and changes in sleep patterns, and motor symptoms, such as tremors, muscle stiffness, and problems initiating and maintaining movements. Because these symptoms vary from person to person and overlap with other disorders, a nonspecialist, such as a primary care provider, may not immediately suspect Parkinson’s disease. “You need to be lucky that your primary care provider knows really well where to guide you,” said Buard.
Buard said that the best-case scenario is that a general practitioner or even a friend who has experience with Parkinson’s disease refers a person with early signs of the disease to a movement disorders specialist. The specialist can then assess the person using the Unified Parkinson’s Disease Rating Scale (UPDRS), a questionnaire which currently provides an official diagnosis for the disease. Specialists and even nonspecialists are fairly accurate when using the UPDRS, but it ultimately relies on the judgement of the doctor administering it. Furthermore, evidence suggests that a person’s symptoms may vary day-to-day, potentially affecting the results of any one assessment.
Dopamine transporter imaging can provide objective information to support a diagnosis because the loss of dopamine-producing neurons drives the motor symptoms of Parkinson’s disease, but not all scientists are convinced that these scans can currently play a diagnostic role without further research and development. Thus, new diagnostic methods that more objectively detect Parkison’s disease earlier are urgently needed.
A new diagnostic typing test
Adams’ goal was to develop a simple test that could help someone recognize their early symptoms as Parkinson’s disease and prompt them to seek the attention of a specialist. After graduating, he launched KeySense, an online typing test that anyone can complete at home for free. After completing the test, several machine learning algorithms determine the likelihood that the typist has Parkinson’s disease, and a final dashboard reveals their scores on characteristics such as rigidity, tremor, and consistency.
The algorithms build off the research Adams did in graduate school. He showed that the amount of time it takes for a person to release one key and press down on the next can readily distinguish between people with Parkinson’s disease and people without motor control issues. This data on keystroke dynamics can indicate if a person has a tremor, slowness of movement, or both. Several other research groups have independently published similar results, with the same stated goal of creating a screening tool for early detection of symptoms.
While wearable devices and smartphone applications could provide similar data, Adams thinks that typing on a keyboard is such a universal and standardized experience that keystroke data will most likely remain the best tool for at-home screening of fine motor impairment. “Typing is convenient,” he said. “Typing will always be around.”
An intelligent keyboard
Trinny Tat, now a postdoctoral scholar in Jun Chen’s bioengineering lab at the University of California, Los Angeles, became aware of the research being done by Adams and other scientists when she was planning her own dissertation project. It was similar to work that her advisor had previously done, showing that keystroke data could reliably differentiate between individuals to verify a person’s identity.
At the time, Tat was looking for a medical application for the latest discovery from Chen’s research group: giant magnetoelasticity, a property of certain soft materials that produce magnetic fields of increasing intensity, depending on the amount of pressure applied to them. She realized that materials with this property could enhance the data collection capabilities of a conventional keyboard by allowing clinicians to collect both timing and pressure data from a typist.
Tat and her colleagues made a cover that could fit over a conventional keyboard. Each key in their cover consisted of a layer of soft, magnetoelastic material and a layer of conductive material. Because a magnetic field can induce an electrical current of proportional intensity, this setup allowed the researchers to use the electrical output they measured on external instruments to calculate the initial pressure applied to each key by the typist. Tat’s team then connected the cover to a Bluetooth device, which would ultimately allow a clinician to collect data wirelessly on a paired smartphone while a patient types on a computer.
Tat was working on her PhD at the height of the COVID-19 pandemic, so she had difficulty recruiting patients with Parkinson’s disease to study the effectiveness of her design. But she and her team managed to enroll three people with Parkinson’s disease and 13 healthy participants for a pilot study. The researchers used this initial data to show that their design was sensitive enough to pick up on the pressure and timing differences among people with and without fine motor impairments. They also used the data to train algorithms capable of categorizing future participants based on those characteristics.
Now, she has plans to conduct a larger study to assess the diagnostic accuracy of their intelligent keyboard design. Ultimately, she and her colleagues see the potential for their device to serve not just as a screening tool or a diagnostic complement, but also as an instrument that could monitor changes in a person’s fine motor skills over time. That could prove especially valuable in the context of Parkinson’s disease, because neurologists adjust their treatment recommendations based on how a patient’s symptoms are progressing. “But that’s very ambitious, and that’s hard to do,” said Tat. “For now, we’re only doing diagnostics.”
An uncertain future for typing tests
According to Buard, movement specialists who want to assess fine motor impairment in people with Parkinson’s disease use established clinical tools like the grooved pegboard test. “The issue with keyboard typing is that there’s no standardized tests that have been validated in research,” she said. “Things move slowly, so maybe in a few years.”
However, if a clinician were to use keystroke dynamics to go beyond early diagnosis and monitor disease progress over time, Buard worries that typing could ultimately prove too complex a task to provide precise diagnostic information. “There may be some issues in terms of comprehension, hearing, reading,” she said. In her own research, she strives to tease apart the distinct symptoms of Parkinson’s disease. For instance, she showed that fine motor impairment can become more severe regardless of whether or not a person has a tremor, confirming that these two phenomena are independent. Before she could be confident in using a clinical typing test, Buard would want to have a clear, holistic picture of all the factors that might influence a patient’s performance.
On the other hand, when considering the use of typing tests for early detection and screening, she noted that fine motor impairment and tremors usually follow nonmotor manifestations of the disease. “Loss of smell is one of the first symptoms,” she said. Consequently, if the goal is detecting signs of the disease as early as possible, then focusing on a sniff test, for instance, might be more effective.
Adams thinks that Parkinson’s disease is so complex and varied that multiple, widely available methods for identifying it will be necessary to cut down on the long diagnostic delays that currently exist. Right now, KeySense prompts users to answer questions about changes in their sense of smell and sleeping habits and factors those responses into its assessment. Adams said that he would welcome tools that could quantify those changes or scans that could more easily and reliably visualize disease indicators in a person’s brain. But for his part, he’s continuing to refine KeySense, implementing incremental improvements to its code, to make it as accurate as it can be. “We’re whittling down, finer and finer as time goes by,” he said.
