Science Translational Medicine: Parkinson's disease surveillance has made a major breakthrough! Chinese scientists have invented an AI model that can sense and track PD patients through the wall!

*For medical professionals only

Parkinson's disease (PD) is the fastest growing neurological disease worldwide[1].

To date, there are still no drugs that can reverse or stop the progression of PD [2].

A major challenge in PD drug development and treatment is the lack of effective diagnostic markers [3].

At present, the diagnosis of PD is mainly based on clinical symptoms, but clinical symptoms often appear several years after the onset of the disease, which leads to the degeneration or even death of 50-60% of dopaminergic neurons in the brain at the time of diagnosis [2].

Therefore, it is urgent to develop early and effective diagnostic markers to assist the clinical diagnosis and treatment
of PD.

Recently, the team of Professor Dina Katabi of the Massachusetts Institute of Technology published important research results in the journal Science and Translational Medicine [4], and doctoral student Yingcheng Liu is the corresponding author and co-first author
of the study.

Screenshot of the front page of the paper

They used wireless sensors to collect signals reflected from PD patients and build predictive models using deep learning to assess the severity of PD, disease progression, and the effectiveness of drug therapy (Figure 1).

Figure 1.
Wireless signals capture signals reflected by the human body

In recent years, wearable diagnostic and treatment devices have emerged one after
another.

However, such devices need to be picked up frequently, charged, and complicated to use, causing huge problems
for the elderly.

Faced with this pain point, Liu and others have developed non-wearable devices based on radar and machine learning technology, which can sense and track the human body through the
wall.

They use instruments that emit low-power wireless signals that pass through walls and bounce through the human body, and then build machine learning models to analyze the reflected signals to identify the attitude and cadence of the subject (Figure 2).

Figure 2.
The team combined radar and machine learning to recognize human body postures

To validate the performance of the technique, Liu et al.
recruited 50 volunteers for the study, including 34 PD patients (age 69.
4±7.
6) and 16 non-PD participants (age 66.
4±13).

These devices were used to monitor and record the pace of all participants at home, and to average the data over time to reduce noise and disturbance
.
The results of the study showed that the average 14-day gait speed of PD patients was 0.
70 m/s, compared with 0.
91 m/s
for non-PD participants.
Compared to the pace of control group participants, PD patients had approximately 23% lower gait (Figure 3).

Figure 3.
Average pace of PD patients (top) versus non-PD participants (bottom) over time

Although the gait rate of patients with PD is significantly lower than that of non-PD participants, slow gait is not a clinical manifestation specific to
PD.

In order to clarify whether pace can be used as an indicator to assess the severity of PD, Liu et al.
correlated it
with the indicators commonly used in clinical PD diagnosis.
The results showed that home pace was closely related
to the total MDS-UPDRS score (Figure 4A) and the MDS-UPDRS Part III subscore (Figure 4B).

Figure 4.
Pace at home is positively correlated with PD test results

In addition, home gait was strongly correlated with MDS-UPDRS and Hoehn and Yahr stage, two indicators for assessing PD severity, but weakly correlated with TUGT and TMWT, two indicators for assessing clinical gait in PD (Figure 4C).

The reason for this is that there are many interfering factors in clinical gait assessment, such as the Hawthorne effect, that is, the evaluator's tendency to change the behavior of the assessee, and the data measured at home are more reflective of the daily habits
of PD patients.
Therefore, home gait is more suitable as a marker for assessing the severity of PD than clinical gait
.

Because participants naturally decreased their pace at home each year as they aged; So did the PD patients and control group participants have the same degree of reduction in home pace? Can a decrease in pace predict the progression of PD patients? To answer the above question, Liu et al.
plotted the volunteers' gait speed change curve
over a 12-month period.

First, the researchers randomly selected one participant from the PD group and one participant from the non-PD group and analyzed the change in gait speed over a year, and found that the gait decline rate of the control group participants was -0.
014 m/s, while the gait decline rate of the PD group was -0.
03 m/s (Figure 5A).

Looking at the pace curves of two randomly selected participants alone, it is impossible to judge whether PD is the main factor causing the decrease in pace speed, so Liu et al.
used a linear mixed-effect model to analyze the pace curves
of all participants in the control group and the PD group.
The results showed that while participants' pace slowed with age, PD patients declined twice as fast as the control group (Figure 5B).

This result also reconfirms the previous finding that there was a significant difference in pace between individuals with and without PD (Figure 3).

In the 12-month study, Liu et al.
also made an important finding: changes in MDS-UPDRS did not capture PD progression (P>0.
05), while the average home pace of PD patients decreased significantly and was significantly higher than that of the control group (P<0.
034).
<b10> These analyses suggest that home pace may be a more sensitive marker for assessing PD progression compared with MDS-UPDRS
scores.

Figure 5.
Patients with PD have a significantly slower pace at home within 1 year

In addition to being used to monitor the progression of PD, the system can also be used to assist in the treatment
of PD.

Studies have shown that within 3 to 5 years of levodopamine treatment, about 50% of PD patients experience motor fluctuations [5], and the effective time of each medication decreases with movement fluctuations, which is also known as the "switching phenomenon" [6].

So can monitoring your pace at home capture the effect of drugs on movement fluctuations?

To answer this question, Liu et al.
plotted the daytime gait curves of four PD patients and found that it fluctuated with medication taken (Figure 6).

That is to say, when the patient's gait speed will increase after taking the drug, but the effect of the drug will weaken 2 hours after taking the drug, and the gait speed will decrease
again.
Comparing each participant's daytime pace with their on/off state throughout the day, the researchers confirmed that stay-at-home pace could reflect the drug's effect
on movement fluctuations.

Figure 6.
Home cadence can be used as a marker of PD movement fluctuations and drug response

It is not difficult to see that the above results can guide the adjustment
of the patient's medication plan.

The researchers noticed that when taking two tablets at a time and three times a day, the movement of PD patients fluctuated greatly, and the peaks and troughs were obvious; After adjusting the medication regimen to one tablet at a time, six times a day, according to the gait change curve, the movement fluctuation of PD patients was significantly reduced, and the gait was more stable (Figure 7).

Figure 7.
The home gait curve can guide patients with PD to adjust their medication schedule

In addition to assessing PD progression, the system has great potential
for home health monitoring and other disease prevention.
Liu et al.
also found that monitoring data could reflect the health effects
of home isolation and hospitalization for atrial fibrillation on participants.

Overall, this study is of great significance
for the diagnosis and treatment of PD.

First of all, home pace can effectively reflect the progression of PD disease early and effectively, which is expected to become a new diagnostic
marker.
Second, the patient's response to drug therapy can be assessed at pace and provided a basis for
adjustment of the medication plan.
In addition, the system has great clinical potential for remote assessment of patients
living in remote areas, with limited mobility or cognitive impairment.

It is worth mentioning that this research team also published another PD-related research in Nature Medicine in August 2022 [7].

They used a similar system to monitor patients' nocturnal breathing signals and developed an AI-based algorithm to diagnose PD early, assess its severity and track the progression of the
condition.

With the advent of global aging, the need for new home health monitoring systems will become more urgent
.
The use of the above system for real-time fine monitoring of human posture is expected to provide data support
for the early diagnosis and treatment of a variety of geriatric diseases.

References

[1] Dorsey E R, Sherer T, Okun M S, et al.
The Emerging Evidence of the Parkinson Pandemic[J].
Journal of Parkinson’s Disease, 2018, 8(s1): S3–S8.

[2] Armstrong M J, Okun M S.
Diagnosis and Treatment of Parkinson Disease: A Review[J].
JAMA, 2020, 323(6): 548–560.

[3] Delenclos M, Jones D R, McLean P J, et al.
Biomarkers in Parkinson’s disease: Advances and strategies[J].
Parkinsonism & Related Disorders, 2016, 22 Suppl 1: S106-110.

[4] Liu Y, Zhang G, Tarolli C G, et al.
Monitoring gait at home with radio waves in Parkinson’s disease: A marker of severity, progression, and medication response[J].
Science Translational Medicine, 2022, 14(663): eadc9669.

[5] Stocchi F, Jenner P, Obeso J A.
When do levodopa motor fluctuations first appear in Parkinson’s disease? [J].
European Neurology, 2010, 63(5): 257–266.

[6] Lees A J.
The on-off phenomenon[J].
Journal of Neurology, Neurosurgery, and Psychiatry, 1989, Suppl: 29–37.

[7] Yang Y, Yuan Y, Zhang G, et al.
Artificial intelligence-enabled detection and assessment of Parkinson’s disease using nocturnal breathing signals[J].
Nature Medicine, 2022.

Responsible EditorBioTalker