We conducted a pilot, before-after single group study and measured the effects of 3-months of IntelliRehab™ use. The study was approved by Human Research Ethics Committee USM (HREC); USM/JEPeM/18030172. All subjects provided a written informed consent prior to commencing.
Subjects
A total of eight adults (n = 8) with hemiparesis ischemic stroke were recruited from Hospital Universiti Sains Malaysia (HUSM), a suburban tertiary referral center for neurological disorders in the East Coast of Malaysia. The sample size was determined based on the mean difference (0.25) and standard deviations (0.20) from a previous study [14] calculated using PS Software [15]. In the previous study, FMA score was used as a measure to assess the participants. With a two-tailed significance level of p < 0.05 and power (1-β) of 0.8, the sample size needed was 8 subjects. The selection of subjects was based on purposive, convenience sampling.
Subjects recruited were hemiparetic patients who had first ischemic stroke episode within the last 4 months and were receiving conventional rehabilitation from the Rehabilitation Unit of HUSM, Kubang Kerian Kelantan at the time of recruitment. These patients were evaluated for eligibility based on the inclusion and exclusion criteria for this study as explained in Fig. 1. We included subjects who were 21-years or older, had no other known neurologic, psychiatric or developmental disabilities, demonstrated hemiparesis following an ischemic stroke, had first supratentorial unilateral ischemia in the cerebral vessels with stroke onset less four (4) months, preserved cognition, and intact vision and hearing, were medically stable, had persistent weakness involving the upper limb(s) and experience difficulty in accessing local stroke rehabilitation. We excluded patients who were unable to undergo an MRI, those with severe aphasia, apraxia, severe depression, severe shoulder subluxation, pain or shoulder dislocation, hemispatial neglect, skull breach or new stroke lesions during intervention period.
Intervention
IntelliRehab is augmented by customized sensor hardware that gamified physical therapy with ‘exergames’ as a mean to remotely monitor patient progress and compliance. This telerehabilitation tool consisted of (1) an intelligent virtual assistant, (2) wireless interaction sensors to capture body motions and, (3) a tool for custom exercises, all built for ease of clinical feedback. A new cloud-based platform was designed specifically to track patient data and incorporate multi-inputs, providing remote monitoring and analytics services. The IntelliRehab sessions were performed at home, where IntelliRehab tool enabled the subjects to perform their upper limb motor training at home as a telerehabilitation set-up (Fig. 2).
A cloud-based platform was designed specifically to track patient’s data and incorporate multi-inputs, providing remote monitoring and analytics services (maintained by project partner: MIRA Rehab Ltd UK). Each subject commenced first with initial training sessions (at least twice) for upper limb exercise using IntelliRehab apparatus. Monitoring and compliance were recorded with the subjects using video communication for 5 sessions every week over a period of 3-months. Each self-directed IntelliRehab session lasted between 45–60 min.
Outcome assessments
The MRI—arterial spin labelling (ASL) profile scan and functional assessments were performed at three time points (baseline, 1-month, and 3-months). Upper limb motor function was assessed on the Fugl-Meyer Upper Extremity Assessment (FMA), and impact of stroke-specific motor changes was assessed on the Stroke Impact Scale (SIS). The FMA score is used to evaluate the levels of upper limbs motor impairment and measure the movement, coordination and reflexes of the shoulder, elbow, forearm, wrist, and hand. Thus, it represents the functional outcome of the stroke survivors during their rehabilitation course. It has previously been established and tested as a reliable, valid, and specific tests for the motor function following a stroke [16,17,18].
Also, SIS is a questionnaire consists of nine questions that are used to evaluate how the stroke has affected the patient’s health, quality of life and activities of daily livings (ADLs) from the patient point of view. Some of these questions emphasize on ADLs related to motor functions such as in SIS-Q1, -Q5, -Q6, -Q7, -Q8 and -Q9. SIS has previously been proved as highly reliable and valid tests to assess ADLs after stroke event [19,20,21].
The feasibility of the intervention including technical problems, technical support and following on self-training and training exercises was evaluated during the intervention and subjects’ responses and feedback recorded by the research team in dedicated logbooks. Questions relating to participation in the research or concerns regarding risk in relation to interventions used in the study were addressed prior to obtaining an informed consent. All ASL MRI and USM IntelliRehab Assessment Proforma data acquired was kept securely in accordance with the Data Protection Act.
ASL MRI acquisition
MRI data acquisition was performed on an Achieva 3.0T (TX) MR system (Philips Medical System, Best, The Netherlands) using the sensitivity encoding (SENSE)-eight‐channel head coil (SENSE-Head-8) as follows. First, high-resolution sagittal T1- turbo field echo (TFE) images (that allow SENSE in slice encoding) were collected using 3D magnetization-prepared rapid acquisition with gradient echo (MPRAGE) with the following parameters: TFE shots = 88, TFE factor = 219, repetition time (TR) = 7.4 ms, echo time (TE) = 3.4 ms, Min. inversion time-delay (TI delay) = 850.8 ms, flip angle = 8°, Water-Fat Shift (WFS)/Bandwidth (BW) = 2.071/209.7 Hz/px, voxel size = 1.1 × 1.1 × 0.6 mm3, field-of-view (FOV) 250 × 241 × 150 mm2, acquisition matrix = 228 × 219, reconstruction matrix = 240, slice number = 250 slices. Secondly, 2D-pASL data were acquired using the following parameters: a single-shot 2D Fast Field Echo/Echo Planer Imaging (2D FFE/EPI) readout. SENSE was used to reduce the echo trains length by reducing the susceptibility related distortions and maximum water fat shift and short TE to keep a good signal to noise ratio (SNR). ASL Multi-Slice Single-Phase was acquired in ascending order, transverse orientation, and contained 20 slices of 6 mm-thickness with the following parameters: dynamic scans = 30, dynamic scan times = shortest, label thickness = 130 mm, label gap = 20 mm, inversion time-delay (TI delay) = 1800 ms, TR/TE = 4000 ms/20 ms, flip angle = 40°, FOV = 240 × 240 × 39 mm2, matrix size = 68 × 68, BW = 1260.4 Hz/px, WFS/BW = 14.943/29.1 px/Hz, Min. slice gap = 0 mm, Act. Slice gap = 0.6 mm, EPI factor = 33. A separate M0 image was not acquired; thus, the mean of the control images was used to measure M0 which was used in ASL analysis. During the resting state ASL scans, all subjects were instructed to keep their eyes closed, relaxed, and move as little as possible, without falling asleep.
ASL processing and analysis
ASL-MRICloud is based on cloud computing using the infrastructure of MRICloud.org [22, 23], as well as computational, and storage resources on a remote server. ASL-MRICloud (https://braingps.mricloud.org/home) [22] was used to process and analyze the acquired ASL MRI data. in the following steps:
Firstly, segmentation of the acquired 3D T1 was done using T1 Multi-Atlas Segmentation option. The 3D T1 data segmentation is a hierarchical brain segmentation [24]. This type of brain segmentation can divide the brain up to 289 regions based on the age of patient and the number of atlases used during the segmentation process. T1 output was skull-stripped and presented in Montreal Neurologic Institute (MNI) template space (2 × 2 × 2 mm).
Secondly, CBF Image Calculation and Processing data were analyzed using ASL v4 option. The quantification of CBF using single‐delay ASL was analyzed through the pASL image (control and label images) series that were corrected for motion by aligning to their respective first time point [25]. This was followed by the calculation of the difference between images (control − label). Estimation of equilibrium magnetization from the control images, after accounting for all RF pulses present in the pulse sequence and by assuming a tissue T1 was used to generate a global M0 which was used in CBF quantification.
Finally, CBF maps were calculated and co-registered with T1 Multi-Atlas results to apply brain parcellation to calculate the regional CBF based on the brain segmentation. The outcome CBF results were normalized and presented in MNI template space (2 × 2 × 2 mm) with regional values of up to 289 standard brain structures. In ASL v4, these parameters were selected for the analysis depending on our current MRI protocol including; pASL in labeling scheme; 2D in acquisition scheme; acquisition timing parameters (TI1[ms] = 800, TI[ms] = 1200, slice acquisition duration [ms] = 35); option “no” in background suppression; and M0 acquisition\estimation parameters (Tissue T1 [ms] = 1165), and assumptions for CBF quantification; (Blood T1 [ms] = 1650, brain/blood partition coefficient [ml/g] = 0.9, and labeling efficiency = 0.98). The complete data workflow is shown in Fig. 3.
Statistical analyses
Statistical Package for the Social Sciences (SPSS) version 22 (IBM Corp) was used to determine the scan time differences during three timepoints using repeated measures analysis of variance (ANOVA) test for CBF extracted values through ASL-MRICloud and for USM IntelliRehab Assessment Proforma (that represented by FMA and SIS) with a corrected threshold of p < 0.05. After that, the regions with significant CBF differences in the repeated measures ANOVA were extracted to be used as regions of interest (ROI). Subsequently, The Bonferroni post hoc analysis was done to determine the CBF changes and motor function changes represented by FMA and SIS during the 3-months duration (at three time points (baseline, 1-month, and 3-months) of IntelliRehab. Bonferroni correction compensates for increase of type 1 error that resulted from the likelihood of incorrectly rejecting a null hypothesis by testing each individual hypothesis at a significance level of 0.05/m where m is the number of hypotheses [26, 27]. The p value in the Bonferroni correction for ROI, FMA and SIS has been corrected using SPSS software. Finally, Pearson correlation test was used to determine the relationships between the significant afflicted brain regions of ASL Profile and the significant FMA and SIS tests. All p values were 2-tailed. Values of p < 0.05 were considered statistically significant.