Participants
Gait pattern categorization of chronic stroke patients
A population of 153 chronic stroke patients, included in a previous study [26], was chosen to perform the gait pattern categorization. All these patients underwent orthopedic procedures to correct equinovarus foot deformity and performed either prior and postoperative gait evaluation. Participants included in that study [26] satisfied the following inclusion criteria: (1) left or right hemiparesis because of ischemic or hemorrhagic stroke (diagnosis confirmed by computed tomographic scan/magnetic resonance imaging or clinical documentation or both); (2) age > 18 years; (3) time since stroke of at least 12 months; (4) mild spasticity level for all lower limb muscles (Modified Ashworth Scale ≤ 2).
The results of the postoperative gait evaluations were chosen for the gait categorization, being well representative of the walking ability of chronic stroke patients in a stable condition. During these assessments, all patients were ambulant, without using any special orthosis; some of them were helped by walking aids such as sticks (n = 70), tripods (n = 8), quadripods (n = 11), whereas the remaining group of patients (n = 64) did not use any aid.
The gait classification was based on temporal and spatial parameters able to identify the overall locomotor performance and the movement symmetry. The mean velocity was included as a variable for the cluster analysis, being defined as a reliable marker of functional disability [9] and being reported as the strongest determinant of group placement in a cluster analysis of stroke patients [27]. Besides, temporal parameters able to discriminate gait pattern in term of symmetry were chosen [24]. In particular, we considered the ratio between the values obtained by the paretic and healthy leg for the following parameters: stance time in percentage of stride time, swing time in percentage of stride time, and the intra-limb ratio of swing time against stance time. The double support time ratio was not considered in the gait categorization because it was unable to identify asymmetric individuals and the mean value did not differ a lot from healthy subjects [24].
A k-means cluster analysis was used to subgroup homogeneous gait patterns. A Mahanalobis distance criterion was adopted to eliminate any outlier from the data sample. The clustering technique is very sensitive to variables which are highly correlated, so all the variables were assessed for correlation and those highly correlated to others were removed. The selected variables were standardized before entering the cluster analysis. The Squared Euclidean distance measure was used and the number of clusters was optimized performing an a posteriori measurement of the silhouette coefficient which evaluated both cohesion and separation of the obtained centroids [28].
Choice of stroke participants
After having performed the cluster analysis of the population of chronic stroke patients, we chose a number of participants equal to the number of identified clusters: each patient was considered as representative of one cluster at baseline. Therefore, participants recruited in this study satisfied the same inclusion criteria of the population chosen for the gait categorization. In addition, patients were characterized by a joint mobility ranges which did not preclude pedaling (knee extension up to 150° and hip flexion up to 80°). The only exclusion criteria was an insufficient cognitive capacity to participate in the program, including receptive aphasia.
The chosen patients were prevented to perform any other lower limb intervention during the BF training.
Healthy subjects participants
A group of 12 healthy subjects (age 22.6 ± 3.3 years, height 171.8 cm ± 9.7 cm, weight 63.3 kg ± 8.9 kg) participated in the study in order to compute the normality ranges for both the pedaling and the walking test used to evaluate the motor recovery induced by the training.
Experimental setup
The THERA-live™ (Medica Medizintechnik GmbH, Germany) motorized cycle-ergometer was chosen for the treatment. It was equipped with a shaft encoder for the acquisition of the crank angle and with strain gauges attached on the crank arms to measure the torque produced by each leg during pedaling [25]. During the treatment, patients sat on a chair or a wheelchair in front of the ergometer and their legs were stabilized by calf supports fixed to the pedals.
A master computer, called master PC, running Matlab/Simulink® under Linux, acquired all signals coming from the ergometer with a sampling frequency of 200 Hz and calculated, at the end of each revolution, the BF indices. Then, these indices were sent to a second PC, called slave PC, which provided the visual biofeedback to the patients, displaying the values of the BF indices through a graphical interface implemented in Matlab. The communication between the PCs was obtained through LAN connection according to the UDP/IP protocol. The experimental setup is shown in Figure 1.
Intervention
The BF treatment was performed 3 days a week for two weeks, obtaining a total of 6 sessions. Each session lasted 14 minutes:
• 1 minute of passive cycling;
• 2 minutes of voluntary cycling without visual biofeedback (VOL1);
• 8 minutes of voluntary cycling with visual biofeedback (BF phase);
• 1 minute of passive cycling;
• 2 minutes of voluntary cycling without visual biofeedback (VOL2).
Passive cycling was guaranteed by the ergometer's motor which maintained the speed at a constant value of 30 rpm.
The communication between the two PCs, shown in Figure 1, was active only during the BF phase. During the other phases the data were only acquired and saved by the master PC.
To compute the BF indices during the BF phase, the active torque profiles for each leg as function of the crank angle were obtained by subtracting the mean torque computed during passive cycling from the torque profile calculated during each revolution of voluntary pedaling. In this way, the inertial and gravitational contribution of the limbs were eliminated. Then, the BF indices for each revolution consisted of the mechanical work produced by the paretic (WPL) and healthy leg (WHL) and were computed as follows:
(1)
(2)
where T
PL
and T
HL
are the active torque profiles produced by the paretic and healthy leg, respectively, while θ represents the crank angle.
The slave PC displayed in real-time, at the end of each revolution, the values of work produced by the two legs, through a graphical interface consisting of two bars with a height proportional to the work values and a yellow band indicating the target (see Figure 1). Patients were asked to voluntary compensate a potential unbalance producing with each leg a value of work within the target band (yellow bands on the two bars). When the two work values were both within the yellow bands, the bars became green; otherwise they were red. To make the exercise more challenging, the target band increased the value of required work when the subjects were able to fulfill the goal for at least 7 over 10 consecutive revolutions. If the patients failed to maintain the increased target for 1 minute, the target decreased again not to discourage the subjects. The target value was subject-dependent and was fixed before the beginning of each session by means of a preliminary test. This test consisted of a 30-second period of passive cycling and a 30-second period of voluntary cycling during which patients were asked to pedal with maximal effort. At the end of the test, the values of WPL and WHL for each revolution were computed and the maximal value achieved by the paretic leg (WPLmax) was used to set the target interval used during the BF phase: the target could range between 80% WPLmax and 120% WPLmax and the target band was fixed at ± 10% WPLmax.
The proposed protocol was approved by the Ethical Committee of the rehabilitation center and each participant signed an informed consent.
Assessment
Participants were tested before, after the intervention and in a follow-up assessment one week after the end of the treatment by means of the following assessment tests:
1. a pedaling test, which comprised a 1-minute period of passive cycling and a 2-minute period of voluntary cycling. The same ergometer used for the BF treatment was employed for this test. Thus, the crank angle and the torque produced independently by the paretic and healthy leg were measured and sampled at 200 Hz.
2. a walking test on a 10-meter walkway. Patients were asked to walk without the shoes at a self selected speed. No constraints were imposed to the subjects and neither assistive devices were used during the test. Three-dimensional kinematics of the subject's lower limbs were recorded with the Elite clinic™ (BTS, Milano, Italy) motion analysis system (8 cameras, sample rate 100 Hz) using the SAFLo protocol [29]. Ground reaction forces were measured with two dynamometric force platforms (Kistler, Winterthur, Switzerland).
Data analysis
Intervention
The performance achieved daily during the BF phase was evaluated by means of the ratio between the number of symmetrical revolutions and the total number of revolutions (BFperf).
During VOL1 and VOL2, the values of WPL and WHL were computed for each revolution as in equations (1, 2). Then the pedaling unbalance (U) was defined as:
(3)
U could range from 0 (two identical works) to 100% (WPL negative or equal to zero).
Assessment
The pedaling test was evaluated in terms of WHL, WPL, and U computed at each revolution. During each assessment test, considering that patients were pedaling at 30 rpm for 2 minutes, the number of revolution was about 60.
Regarding the walking test, all raw data were filtered with a fifth order causal Butterworth filter (cutoff frequency of 5 Hz) and elaborated to compute kinematics, kinetics and standard temporal and spatial gait parameters [26, 29].
To evaluate gait symmetry two indices were computed:
- ST ratio, i.e., the ratio between the stance time in percentage of the stride time obtained by the paretic leg and the one obtained by the healthy leg. The ST ratio could be related to balance control issues leading the patients to shorten the paretic stance time [24].
- SV ratio, i.e., the ratio between the swing velocity obtained by the paretic leg and the one obtained by the healthy leg. The SV ratio could be related to an insufficient power generated to swing the paretic limb quickly and to an increased time for paretic foot placement [24].
All values of the temporal and spatial gait parameters reported are the mean values of 4 to 5 repeated gait trials along the walkway at the preferred speed.
Statistics
After having evaluated that all patients' parameters were normally distributed, an intra-subject one way Analysis of Variance (ANOVA, p < 0.05) was performed to compare pre-, post-training and follow-up outcome measurements. Moreover, a Mann-Whitney U test (p < 0.05) was used to compare patients' performance before training, after training, and at follow-up visits, with the group of healthy volunteers. A non-parametric test was preferred to identify any statistically significant difference between patients and healthy subjects, being the group of able-bodied participants not normally distributed.