Technological Innovations for Parkinson Disease: Is Artificial Intelligence the Key to Better Diagnosis and Management?

Welcome to Health Connect, the podcast for health professionals where we will share the latest news and information on science and technology in the medical field. Today, we will talk about technological innovations in Parkinson disease.

Did you know that artificial intelligence can contribute to both early diagnosis of Parkinson disease and patient monitoring?

More than 200 years ago, in 1817, Dr. James Parkinson published a scientific work entitled “The Essay on Shaking Palsy” and with it the foundation of the disease that bears his name.1

After two centuries, we still struggle to understand and treat neurodegenerative diseases, such as Parkinson disease, which are growing exponentially, especially in industrialized regions. Given that no one is immune to them, specialists are concerned about something they are calling the “Parkinson pandemic”.1

Parkinson disease is a common progressive neurodegenerative disorder that can cause significant disability and decreased quality of life.1

Nowadays the disease is more present than ever, being the second most common neurodegenerative disease after Alzheimer disease, and affecting 0.3 percent of the global population. Surprisingly, it is estimated that only in 2015 Parkinson disease caused the death of 117,000 people. The American Parkinson Disease Association estimates that there are already 1 million people living with this condition in the U.S. alone and over ten million worldwide; and this number is expected to double by the year 2030, which will have a significant socio-economic impact.1

Statistically, Parkinson disease occurs in people over 60 years old, affecting one percent of this age group and increasing to 4 percent for people over 80.1 

However, it can also occur in people under 50, known as Young Onset Parkinson disease. Studies show that 5 to 10% of patients diagnosed with the disease are between the ages of 20 and 50 years. There are even cases in the literature of patients younger than 20 diagnosed with this illness, and even some rare cases with patients under ten years of age, with the first symptoms appearing as early as two years of age.1

During the 15 to 20 years prior to the onset of motor symptoms, patients experience a phase of “prodromal Parkinson disease”, through which neurodegeneration starts and progresses. Olfactory impairment, constipation, depression, rapid eye movement and sleep behavior disorder can be present in the prodromal period.1

Unfortunately, at the time of the diagnosis, a significant number of neurons that produce dopamine are already dysfunctional.1 At present, the clinical diagnosis of Parkinson disease mainly relies on some subjective evaluations, such as clinical symptoms, family history, and dopamine therapy response, leading to a misdiagnosis rate of nearly 25%. Therefore, it is very important to find an objective and efficient way to improve the diagnosis rate of this condition.2

Disease clinical presentation can take the form of many symptoms, but motor impairments are the most noticeable.1

Motor impairment in Parkinson disease has long been the focus of researchers, with significant advances being made in diagnostic accuracy, implementation, improvement of more accurate assessment scales, and better management of therapeutic strategies.1

Among parkinsonian motor symptoms, postural instability is one of the most debilitating. It severely affects balance and causes frequent falls and injuries, leading to increased hospitalization rates, loss of independence and higher mortality.3

Accordingly, the early identification of postural instability is a primary concern in the clinical management of these patients in order to optimize therapeutic strategies, prevent injuries and support the individual’s autonomy.3

The current challenge resides in the low sensitivity of routine clinical examinations to early recognize postural instability. Indeed, clinical assessment usually detects postural instability only when the patient becomes symptomatic and has complaints of balance issues and recurrent falls.3

Recently, there have been attempts to diagnose Parkinson disease early using artificial intelligence, especially using machine learning and deep learning algorithms.4 

Machine learning is used practically everywhere, from cutting edge technology such as mobile phones, computers, and robotics, to health care. Many researchers and practitioners illustrate the promise of machine learning-based disease diagnosis, which is inexpensive and time-efficient.5

While the individual’s ability restricts traditional diagnosis techniques, machine learning-based systems have no such limitations, and machines do not tire as humans do. As a result, a reliable method to diagnose disease in health care settings may be developed.5

Machine learning algorithms are being used on a variety of data modalities, including acoustic voice recording and handwritten patterns for the diagnosis of Parkinson disease.6

Mobile technologies such as smartphones and widely used low-cost sensors produce large amounts of health data. This data may be used by artificial intelligence to provide previously unattainable insights on the prevalence of diseases and patient status in everyday environments, as well as in clinical datasets. Thus, AI can help with patient symptom monitoring.6

So, machine learning technologies automate the study of multipurpose datasets to reveal models that would be difficult to detect using standard observation or statistical methods. Recent advances in data analysis and wearable sensors for human movement monitoring can promote objective and accurate clinical evaluation of neurological symptoms and improve outcome measures in clinical trials.7

As we know, gait disorders are a hallmark of the condition and they closely correlate with disease progression and are associated with a loss of independence and an increased risk of falls, affecting the individual’s quality of life.1,7 Gait disturbances, even if hardly noticeable, are described from the earliest stages of the disease and include shuffling gait, shortened stride length, reduced overall velocity, and increased stance phase, along with reduced or absent arm swing, reduced trunk rotation, and decreased amplitude of motion in the hips, knees, and ankles. In advanced stages, gait disorders often become increasingly complex, including motor blocks, festination, and imbalance.1

The use of machine learning in gait analysis has already shown promising results. Several studies have applied machine learning classification for the detection, quantification, and classification of gait abnormalities in people with this disease using gait data from several gait analysis systems.7

The automatic classification of gait impairments using machine learning algorithms, when combined with gait analysis using inertial measurement units, may allow a prompt and clinically meaningful assessment of gait abnormalities in people with movement disorders.7

Compared with optoelectronic three-dimensional motion analysis systems and instrumented walkways, wearable inertial measurement units allow retrieval of a wide range of gait data while lowering gait analysis costs and facilitating the measurement of walking characteristics outside laboratories.7

For instance, gait data extracted from a single inertial measurement unit in the lower back, can easily define spatiotemporal gait characteristics and pelvic kinematics based on trunk acceleration patterns during walking. In addition, data from these units can be used to compute trunk acceleration-derived gait indexes, which characterize the dynamic unbalance of subjects with movement disorders.7

Support vector machine, decision trees, random forest, k-nearest neighbor, and neural networks are the most-used supervised machine learning algorithms for classification purposes on gait data derived from wearables. Classification models with accuracies greater than 90 percent have been described using these algorithms.7

A study published in May 2022 by Trabassi and colleagues, aimed to determine which supervised machine learning algorithm can most accurately classify people with Parkinson disease from speed-matched healthy subjects based on a selected minimum set of gait features derived from inertial measurement units.7

They did it by extrapolating 22 gait features from the trunk acceleration patterns of 81 subjects with Parkinson disease and 80 healthy subjects, including spatiotemporal, pelvic kinematics, and acceleration-derived gait stability indexes.7

Researchers concluded that support vector machine, decision trees, and Random Forest showed the best classification performances, with prediction accuracy higher than 80 percent on the test set.7

These results further reinforce the applicability of supervised machine learning algorithms, especially support vector machine, in gait prediction, but also tree-biased methods such as Random Forest and decision trees.7

CAPSULE
Artificial intelligence is becoming increasingly common in healthcare, with applications ranging from screening and triage to clinical risk prediction and diagnosis. Under its current use, AI operates largely as a clinical tool used by physicians. However, certain uses may present novel legal challenges.8

Artificial intelligence’s accuracy is contingent on the reference data set that the algorithm uses to learn. This data set can vary in its representation of certain population characteristics, and this may impact the accuracy of its assessment. As the complexity of AI algorithms become more difficult to comprehend, they are referred to as “black boxes” to reflect their potential lack of transparency.8

If a physician relies on AI’s findings and they are incorrect or inaccurate, there may be implications for the patient in the form of either being subject to unnecessary treatment, unnecessary follow-up procedures or an undiagnosed condition.8

As the use of AI increases in healthcare, new legal questions may arise, including the governance of the AI models, the liability for harms suffered, and the changes in standards of care in medical practice.8
END OF THE CAPSULE

Welcome back. Even though Parkinson disease is a non-curable condition, the progression of the disease can be controlled by an early diagnosis and proper management. Artificial intelligence is a promising tool to detect early-stage Parkinson disease.4

High-resolution structural magnetic resonance imaging has been frequently used to detect the subtle changes of the human brain. In patients with Parkinson disease, numerous studies have found widespread brain atrophy in the cerebellar, subcortical, and cortical regions by using various structural magnetic resonance imaging features, such as gray matter volume, white matter volume, and cortical thickness. In addition, some cortical morphological features, including local gyrification index, local fractal dimension, and sulcal depth, are also increasingly used to detect the structural changes occurring in this disease.2

In recent years, machine learning technology based on structural MRI features has developed rapidly, showing huge advantages in assisting individualized diagnosis in various neurological and psychiatric diseases.2

Compared with traditional group-level analysis, machine learning technology takes inter-regional correlations into account, thereby providing increased sensitivity to subtle changes and spatially distributed difference. In patients with Parkinson disease, some structural features have been used to distinguish patients with the disease from normal controls. For example, by using the gray matter volume, white matter volume, and cerebrospinal fluid volume of substantia nigra, thalamus, hippocampus, frontal lobe, and midbrain, Rana and colleagues found that the machine learning model based on the substantia nigra had the highest accuracy, sensitivity, and specificity in differentiating patients from controls.2 

In a study published in 2022, Yang Ya and colleagues developed machine learning models for the diagnosis of Parkinson disease using multiple features from structural magnetic MRI. The researchers constructed a cerebellar model, a subcortical model, and a cortical model based on corresponding features, separately, and a combined model that integrated all selected features. They found that all their models had high diagnostic efficiency and clinical net benefits for Parkinson disease in both internal and external validation datasets; of all of them, the combined model performed best, followed by the cortical model.2

The most diagnostic discriminating brain regions identified by machine learning could then function as potential neuroanatomical markers of this condition, further deepening our understanding of its pathogenesis.2 

Their combined machine learning model based on multiple indicators may be of great value in assisting the clinical diagnosis of the disease and may become an effective and clinically applicable new method.2

Moving on to other AI studies on Parkinson disease, a study in the US by Saiful Islam and team developed an artificial intelligence system to remotely assess the motor performance of individuals with Parkinson disease. Their machine learning model was developed to predict the severity of the symptoms in Parkinson disease based on extracted features from a finger-tapping video.9

The finger-tapping task is commonly used in neurological exams to evaluate bradykinesia in upper extremities, which is a key symptom of Parkinson disease.9

Researchers obtained data from 250 global participants, 172 with Parkinson disease, and 78 controls, who completed a finger-tapping task with both hands. Participants used a web-based tool to record themselves with a webcam primarily from their homes.9

The study quantified 42 features measuring several aspects of the finger-tapping task, including speed, amplitude, hesitations, slowing, and rhythm. They found that 22 features were significantly correlated with the severity scores, which reflects their promise for use as digital biomarkers of symptom progression. These features are clinically meaningful as they capture several aspects of speed, amplitude, and rhythm of the finger-tapping task.9

The authors remark that this tool is not intended to replace clinical visits for individuals who have access to them. Instead, the tool can be used frequently between clinical visits to track the progression of the disease, augment the neurologists’ capability to analyze the recorded videos with digital biomarkers, and fine-tune the medications.9

Even for those with access to care, arranging clinical visits can be challenging, especially for older individuals living in rural areas with cognitive and driving impairments.9

Periodic clinical evaluations and medication adjustments can help control symptoms and improve quality of life. This developed tool could take a more active role by automatically assessing the symptoms frequently and referring the patient to a neurologist if necessary.9

Thanks for joining us on this episode of Health Connect. Don't miss out on our next episode.

Discover more medical news on Viatris Connect.

 

References:

  1. 1. Ileșan RR, Cordoș CG, Mihăilă LI, Fleșar R, Popescu AS, Perju-Dumbravă L, et al. Proof of Concept in Artificial-Intelligence-Based Wearable Gait Monitoring for Parkinson's Disease Management Optimization. Biosensors (Basel). 2022;12(4):189. Available at: https://www.mdpi.com/2079-6374/12/4/189
  2. 2. Ya Y, Ji L, Jia Y, Zou N, Jiang Z, Yin H, et al. Machine Learning Models for Diagnosis of Parkinson's Disease Using Multiple Structural Magnetic Resonance Imaging Features. Front Aging Neurosci. 2022;14:808520. Available at: https://www.frontiersin.org/articles/10.3389/fnagi.2022.808520/full
  3. 3. Castelli Gattinara Di Zubiena F, Menna G, Mileti I, Zampogna A, Asci F, Paoloni M, et al. Machine Learning and Wearable Sensors for the Early Detection of Balance Disorders in Parkinson's Disease. Sensors (Basel). 2022;22(24):9903. Available at: https://www.mdpi.com/1424-8220/22/24/9903
  4. 4. Paul S, Maindarkar M, Saxena S, Saba L, Turk M, Kalra M, et al. Bias Investigation in Artificial Intelligence Systems for Early Detection of Parkinson's Disease: A Narrative Review. Diagnostics (Basel). 2022;12(1):166. Available at: https://www.mdpi.com/2075-4418/12/1/166
  5. 5. Ahsan MM, Luna SA, Siddique Z. Machine-Learning-Based Disease Diagnosis: A Comprehensive Review. Healthcare (Basel). 2022;10(3):541. Available at: https://www.mdpi.com/2227-9032/10/3/541
  6. 6. Rana A, Dumka A, Singh R, Panda MK, Priyadarshi N, Twala B. Imperative Role of Machine Learning Algorithm for Detection of Parkinson's Disease: Review, Challenges and Recommendations. Diagnostics (Basel). 2022;12(8):2003. Available at: https://www.mdpi.com/2075-4418/12/8/2003
  7. 7. Trabassi D, Serrao M, Varrecchia T, Ranavolo A, Coppola G, De Icco R, et al. Machine Learning Approach to Support the Detection of Parkinson's Disease in IMU-Based Gait Analysis. Sensors (Basel). 2022;22(10):3700. Available at: https://www.mdpi.com/1424-8220/22/10/3700
  8. 8. Jassar S, Adams SJ, Zarzeczny A, Burbridge BE. The future of artificial intelligence in medicine: Medical-legal considerations for health leaders. Healthc Manage Forum. 2022;35(3):185-189. Available at: https://journals.sagepub.com/doi/10.1177/08404704221082069
  9. 9. Islam S, Rahman W, Abdelkader A, Yang PT, Lee S, Adams JL, et al. Using AI to Measure Parkinson’s Disease Severity at Home. arXiv:2303.17573. 2023. Available at: https://arxiv.org/abs/2303.17573

Os links para todos os sites de terceiros são oferecidos como um serviço aos nossos visitantes e não implicam endosso, indicação ou recomendação do Health Connect. Os artigos vinculados são fornecidos apenas para fins informativos e não visam implicar uma atribuição pelo autor e/ou editor. O Health Connect se isenta de qualquer responsabilidade pelo conteúdo ou pelos serviços de outros sites. Recomendamos que você analise as políticas e condições de todos os sites que escolher acessar.

NON-2023-9673
NON-2023-2512