Diagnosis and severity assessment of Parkinson’s Disease through machine learning
4 January 2022
Parkinson's disease (PD) is a debilitating neurological illness that significantly reduces the mobility of its victims. It is caused by the degeneration of the central nervous system, and its patients exhibit a wide range of symptoms, including tremors, rigidity, postural instability, slowness, and in worse cases, physical incapacitation.
The Parkinson's Foundation estimates there are 10 million people worldwide living with the disease. Parkinson's disease is the second most age-related neurodegenerative illness after Alzheimer's disease. The lion's share of patients develop symptoms and get diagnosed with Parkinson's disease around the age of 60 or older.
In Malaysia, there are an estimated 20,000 Parkinson's disease patients. However, there is expected to be a sizeable leap in the number of patients in the coming years. "The prevalence of Parkinson's disease in Malaysia is expected to increase by five-fold over the next 20 years," says Dr Alpha Agape Gopalai from the School of Engineering, Monash University Malaysia.
The number of people diagnosed below the age of 50 may only account for four per cent of worldwide Parkinson's disease patients. However, it is still a significant amount that warrants scrutiny. Young-Onset Parkinson's disease (YOPD) affects patients between the ages of 21 and 50.
The varying stages of severity in patients can be represented by the Hoehn and Yahr (H&Y) score. Patients in stage one of the disease experience tremors and other mild forms of movement-related impediments that do not impact their daily activities.
The symptoms of stage two mirror that of stage one patients, although issues such as difficulty walking and awkward posture become apparent among the former. This physical deterioration then continues where patients in stage three start to experience slowness of movement, which can impair daily activities such as getting dressed and eating.
The physical decline continues where patients in stage four of the disease experience extreme symptoms. They are often unable to get on with their daily activities without seeking help from others and require additional aid such as a walker. Patients in stage five experience stiffness in their legs and often cannot walk anymore, becoming bed-ridden or wheelchair-bound.
An easily visible symptom of Parkinson's disease is the deterioration of walking patterns, muscle tremors and speech impairment. However, the disease begins much earlier before such symptoms become apparent to the observer. Therefore, early detection is imperative in arresting the decline of motor functions and ensuring patients have a better quality of life through proper medication or therapy. At present, the process of diagnosing the disease and determining its severity involves a series of visits to the hospital for a series of tests.
Srivardhini Veeraragavan set out to implement a system that would facilitate a preliminary self-test procedure based on wearable technology during her final year Bachelor of Mechatronics, Robotics, and Automation Engineering program at Monash University Malaysia.
Srivardhini's work led to the publication of a research paper on how a machine learning technique can be used in the diagnosis and severity assessment of Parkinson's disease. Srivardhini says this project was driven by a desire to reduce the number of trips that patients have to make to clinics for doctors to diagnose the disease and determine its severity in patients.
"Patients normally take multiple assessments where doctors have the opportunity to observe their movements. They are assigned tasks such as walking, speaking, or writing. From these visual observations, doctors then make the diagnosis and determine the severity of the disease," says Srivardhini, who is now pursuing her PhD at Monash University Malaysia.
Dr Alpha, who was the supervisor of Srivardhini's project, says that making multiple trips to see a doctor in the process of diagnosing and ascertaining the severity of PD can be physically, emotionally, and financially taxing for patients and their caregivers.
"Our aim was to investigate the potential of designing a routine which could probably help in, firstly, detecting the onset of PD, and, secondly, monitoring the progression of the disease. If this is done successfully, we could help in reducing the frequency of clinic visits for PD patients and provide clinicians with more data to help them understand the progression of the disease," he said.
The machine learning method developed by Srivardhini uses open-source data gleaned from a previous study (Goldberger et al., 2000) examining the gait of PD patients. The (Goldberger et al., 2000) study obtained the vertical ground reaction force (VGRF) of 93 Parkinson's disease patients by placing eight sensors apiece on the insole of the subjects' feet. These force-resistive sensors help determine the VGRF of a patient. VGRF refers to the amount of force that the ground exerts against a body it is in contact with. VGRF is widely used in sports performance analysis and rehabilitation. These sensors can also be easily incorporated into an insole or a shoe.
Using the VGRF data from (Goldberger et al., 2000), Srivardhini extracted key events while walking to be used as a training feature for the Artificial Neural Network (ANN) model that would help diagnose PD. ANN is a type of machine learning technique that loosely resembles the neurons in a biological brain.
Once a subject is ascertained to be positive for Parkinson's disease, Srivardhini then develops another ANN model to predict the severity of the patient's PD on the H&Y scale. "Our machine learning algorithm tries to map a patient's walking gait data to this clinical scale, and through this quantification, we can tell how severe their PD is, based on how they walk," she shared.
The severity assessment is critical to PD patients, who, because of their slow decline, may not realise their weakening state. "The more severe the stage of the disease, the more likely they are to fall and thus need more support. Often, this may not be something the patient notices because the decline is gradual," says Dr Alpha. "Falls among the elderly are among the leading causes of morbidity in our population."
Srivardhini's non-invasive method of diagnosing Parkinson's disease and determining its severity by solely using VGRF data resulted in an impressive 97.4 per cent diagnosis accuracy and 87.4 per cent severity accuracy based on the H&Y score.
Both Srivardhini and Dr Alpha say the positive outcome of this research can be used as a platform to expand the study further with a wider range of subjects database of PD patients with varying levels of severity.
"We have shown that the concept works and have also published our findings. The next phase in expanding this research will be increasing the database variation by working on the data collection so that the algorithm will continue to work over a larger population and demography," said Dr Alpha.
"This will help improve the learning capability of the algorithm and thus give better quality output estimates. I envision for this project a portable device that facilitates the assessment of PD severity and initial diagnosis testing outside a clinic or laboratory setting," he added.
Monash University has placed 57th in the world in the Times Higher Education World University Rankings 2022. Its research is strengthened through global collaboration with industry and government, and an improved industry income score demonstrates the success of these relationships.
Besides that, Monash University Malaysia has given out scholarships worth RM200mil to successful recipients for the past ten years.
If you are passionate about tackling core challenges facing the world, Monash University Malaysia is the perfect place for you. Learn more about our postgraduate coursework and research programs at www.monash.edu.my.