2.1. Questionnaires and Scoring Systems via Mobile Applications

Various questionnaires and scoring systems are available to evaluate one’s risk of stroke according to the self-reported personal health condition and lifestyle habits and behaviors [17]. Since the popularity of the smartphone is high, conveying the questionnaires or scoring systems via mobile applications (mobile app) increases the accessibility of this population-wide strategy for stroke risk stratification [18]. The Stroke Riskometer is a mobile app endorsed by the World Stroke Organization to evaluates one’s risk of stroke over the next 5 to 10 years based on 20 questions. In addition, it provides suggestions to lower the risk of stroke [8,19]. Another stroke risk system is specifically developed for the Chinese population to predict 10-year and lifetime stroke risk (Table 1) [20].

Table 1. Risk factors in a scoring system specifically developed for the Chinese Population.

These questionnaires are developed based on Framingham risk score, which is the oldest scoring system developed for stroke risk prediction [21]. During evolution of the past 20 more years, around 10 other scoring systems have been developed based on the same Framingham risk score [22,23]. These scoring systems are mainly applied on patients with atrial fibrillation (AF) to stratify their risk of stroke. AF is the most common disorder of heart rhythm, which accounts for at least 20% of all types of stroke [24]. AF means rapid and irregular beating of the atrial chambers of the heart resulting in abnormal heart rhythm. Blood can be stagnated and thrombus can be formed within the left atrial appendage due to the dysrhythmia. This can lead to cardioembolic stroke. The risk of stroke of patients with non-valvular AF is five times higher than those without AF [25]. These scoring systems help to stratify patients with AF to benefit most from anticoagulation, which is reported to reduce more than two-thirds of the incidence of stroke [25].

The scoring systems available as mobile apps are easy to access and intuitively interpret results. However, there are still limitations for these scoring apps. First, people need to know their common physiological parameters, such as systolic blood pressure (SBP), diastolic blood pressure (DBP), high-density lipoprotein (HDL) cholesterol, and low-density lipoprotein (LDL) cholesterol when conducting the questionnaires. Second, the systems do not allow real-time updates of the physiological parameters. Third, a scoring system may be not validated for people with various ethics or at various regions. Fourth, these assessments cannot suggest the risk of stroke in the near future, such as coming weeks or months.

To improve current scoring systems, mobile apps can be connected to wearable devices which record the real-time physiological parameters. The processing algorithms secure better assessment when using the variation of the imported physiological parameters. In addition, customizing the apps for individuals can increase the overall outcome of scoring systems.

2.2. Sensor for Air Pollution Embedded in Smart Phone

The association of air pollutants, such as airborne particulate matters up to 2.5 µm in diameter (PM_2.5), and various toxic gases with stroke have been evaluated this decade [26,27]. It is proved that short-term exposure of PM_2.5 and the toxic gases are associated with an increase in hospitalization due to ischemic stroke in China [28]. In fact, up to 29% of the risk of stroke is attributable to air pollution [4]. Worryingly, unlike most other modifiable risk factors, air pollution is unavoidable. Domestic and worldwide policies to lessen the impact of air pollution on risk of stroke are urgent needs.

An air pollution detector can validate both the daily concentration and the duration of such pollution. For example, W-Air is a platform embedded in a wristband for air pollution monitoring [29]. It allows measuring toxic gases of the environment without the influences of the gases emitted by the users. However, W-Air is not able to monitor PM_2.5. A miniaturized mass sensing based on a micro-electro-mechanical system (MEMS) structure is developed to monitor PM_2.5 [30]. However, the main criterion of air pollution sensor for stroke risk prevention is to continuously monitor both PM_2.5 and toxic gases, and the measured values should not be influenced by other scents emitted by the user. Also, the wearable device is better integrated with a mobile phone to avoid wearing extra devices [31]. Besides improving the sensor, increasing the public awareness of the risk of stroke resulting from air pollution represents a key step to gain the effectiveness of these sensors.

2.3. Devices for ECG Monitoring

Patients with atrial fibrillation, determined by the irregularity of the pulse rate, have a five-times higher risk of stroke than those without AF. Among those with clinical AF, one third of them have subclinical AF, which can also result in thromboembolic occurrence and leads to ischemic stroke [32]. People with subclinical AF have no obvious symptoms and often cannot be detected using a conventional short-term Electrocardiogram (ECG) exam, resulting in less attention to their risk of stroke. With the emergence of 24-h portable ECG or cardiac implantable electronic devices (CIEDs), such as a pacemaker, subclinical AF and device-detected AF can be detected, increasing the accessibility of the risk factors of stroke [33,34,35]. Moreover, with the dramatic increase in the accessibility to ambulatory ECG via wearable devices, other cardiac risk factors related to stroke can be monitored continuously and analyzed in real-time [36].

Apple Watch (Apple Inc., Cupertino, CA, USA) series 4 extracts heart rate from recorded photoplethysmography (PPG) [37,38]. The incorporated algorithm, which is the first to obtain Food and Drug Administration (FDA) clearance to determine AF, derives the pulse rate from the peak to peak interval of PPG pulsations [39]. However, pulse rate irregularity is not the only character of AF, other irregular electrical activity of ECG recording can be used to identify AF. Therefore, wearable devices that enable ECG monitoring benefit more on stroke risk prevention. Thus, another ECG device called KardiaMobile 6L from AliveCor, a smartphone attachment enabling 6-lead ECG recoding, is presented [37]. It is also an FDA cleared device for detection of AF. The integrated KardiaAI platform can distinguish AF between normal sinus rhythm based on the recorded ECG signals.

Both algorithms of the discussed devices are designed to determine the presence of atrial fibrillation. However, it has been shown that AF is not always necessary for the formation of thrombus and the occurrence of embolization [40]. Abnormal atrial structure and function can also result in thrombosis and then increase the risk of stroke, even in people without AF [41]. These abnormalities found in P-wave indices, Q-wave, QRS/QT duration, other waveform angles and slopes, and HRV are associated with risk of stroke [42,43,44].

The wearable devices enabling the monitoring of these factors are often in the form of a chest patch, chest strap, or garment. The usability and user comfort of these devices lead to a trade-off with the number of ECG channels. The limited recording lead reduces the applicability of these devices on stroke risk prediction compared to conventional 12-lead ECG. Besides the challenge of balancing the form factor and applicability of wearable ECG devices, there are other limitations. The first limitation is that the ECG measurement is recommended to be conducted when resting, since the motion artifact disturbs the recorded PPG or ECG signals. Second, the algorithms developed and released nowadays are restricted to irregularity of pulse rate due to the complexity of real-time analysis of other cardio abnormalities related to risk of stroke. Third, consultation of professionals for further inspection and diagnosis when abnormal ECG signals are detected is still needed. Fourth, the recorded results can be biased when lack of active and continuous monitoring.

To increase the impact of ECG devices on stroke risk prediction, developing user friendly devices to record meaningful signals for stroke risk analysis is the first step. Secondly there is a need to increase their accessibility in underdeveloped countries, where prevalence of stroke is often higher. The third step is to develop algorithms with high accuracy and specificity to reduce unnecessary anxiety or further testing. Fourth, combine EHRs with the recorded ECG to customize the algorithms resulting a dynamic prediction system. Fifth, increase the awareness of stroke risk prevention using wearable devices and encourage the users to actively take actions when receiving warming signals of abnormal ECG.

2.4. Devices for Vascular Related Risk Factors Monitoring

Since stroke is a neurovascular disease, the abnormalities of vascular related risk factors strongly associate to the risk of stroke. Here, devices enabling continuous monitoring of blood pressure, pulse pressure, arterial stiffness, and blood flow dynamics for stroke risk prediction are introduced.

2.5. Devices for Carotid Plaque Characterization and Cerebral Microembolization Monitoring

Non-invasive imaging techniques, such as computed tomography (CT) and magnetic resonance imaging (MRI) are the gold standards to characterize the structure of blood vessels (accumulation of plaques, development of atherosclerosis and reduction of lumen diameter) and detect blood flow speed. Here, two more compact and user-friendly imaging techniques compared with CT and MRI are introduced: ultrasound for carotid plaque monitoring and TCD ultrasonography for embolic signal detection.

2.6. Gait and Motion Monitoring

When stroke cases occur, the patients experience muscle-related difficulties, such as blindness or blurred vision, unclear voice, walking or maintaining balance problem, difficulty moving some muscles, and weakness in the limbs or muscles.

A gait monitoring system including an accelerometer and pressure sensors to record gait speed, foot pressure, and ground reaction force is proposed for stroke risk prediction [83,84]. Another promising system for stroke risk prediction is developed to perform gait analysis without specific alignment motions. The algorithm along with the system distinguishes the level of disability of stroke survivors by analyzing the asymmetry of gait parameters measured from lower limbs [85]. Another study statistically analyzes gait nonlinear patterns to distinguish healthy young subjects (23–29 years old), healthy elderly subjects (71–77 years old), and patients with Parkinson’s disease. The complexity measures, walking stride time series, can be potential parameters to predict the risk of stroke [86]. The machine learning technique is used to extract the features and perform classification. In addition, wearable sensors for motion and surface electromyography (EMG) monitoring are widely used to evaluate the rehabilitation of muscle function and motion ability on stroke patients [87]. The EMG signals indicating the transformation of impaired to normal condition of muscles can be applied as features to predict the occurrence of stroke when a part of muscle function is weakened. A wireless body area network composed of multiple sensor nodes and machine learning algorithms are applied to analyze the motion performed by human body. This is a promising system for stroke risk prediction [88].

Besides the loss of muscle function of the limbs, the ocular muscles often become uncontrolled in stroke patients. Therefore, a goggle which combines a pair of glasses and a motion detecting camera is designed by Neurobit for early stroke detection or prediction [89]. By recording the eye movements when a subject is performing the assigned tasks, the algorithm of the system stratifies the risk of stroke. Moreover, people with stroke can experience difficulties in speaking clearly and fluently. A mobile application program is developed to predict risk of stroke by analyzing the users’ voice input [90].

2.7. Devices for EEG Monitoring

The above presented devices, except TCD, monitor the physiological parameters indirectly related to cerebrovascular itself. Wearable devices enabling brain activities monitoring should increase the prediction outcome. Since the changes of cerebral blood flow impact the activities of neurons, EEG signals contain stroke risk predictors. The alpha waves decline when the CBF decreases. The theta and delta waves appear when the CBF further decreases [91,92,93]. Besides the sub bands analysis of EEG, the local brain symmetry index (BSI), relevant delta power, and relative local delta to alpha ratio (DAR) calculated from the recorded EEG signals are used as indices for detecting stroke.

2.8. fNIRS Devices for Hemodynamic Signals Monitoring

Neural electrical activities recorded using EEG are associated with the cerebral hemodynamic, which can be monitored using functional near-infrared spectroscopy (fNIRS). This neurovascular coupling means that when neuron activity arouses, CBF around the area increases to supply more oxygen for neuron activity. fNIRS measures the concentration change of oxygenated hemoglobin ([HbOxy]) and that of deoxygenated hemoglobin ([HbDeoxy]) [94,95]. Together with the derived parameters, HbT (the combination of the previous two parameters) and rSO₂ (regional cerebral tissue oxygenation), these values are used to evaluate the hemodynamic states of stroke patients.