The pilot study tested the feasibility of supplementing MI training to physiotherapy. Specifically, MI training was embedded into physiotherapy and added after physiotherapy to learn a complex motor task: 'Going down, laying on the floor, and getting up again'. Furthermore, both MI integration approaches were compared to a control group that listened to tapes with information on stroke. All further factors regarding the study interventions remained the same for all groups. All groups received the same amount of attention and kind of physiotherapy content. They showed significant changes in the primary outcome measure time needed to perform the motor task from pre to post-intervention. The significant improvement could be maintained during follow-up period, which is an important aspect of therapy intervention studies . No group differences in time needed to perform the motor task was detected from pre to post-intervention.
Patients in all groups showed a high compliance and were highly motivated. Frequently named reasons for study participation were to help other patients after stroke with the research findings. Furthermore, patients were interested to learn the MI technique. All were able to learn the task, completed all 13 stages, and were able to improve the motor task performance regarding time and help needed considering the long time period and functional level since stroke onset and study participation.
Different MI integration approaches
Both MI interventions were designed based on currently accepted MI intervention paradigms. Embedded MI based on the work from Liu et al. and the PETTLEP framework from sports psychology [7, 18], whereas added MI was derived from the results of Page's publications [4, 5]. In a recently published systematic literature review on motor imagery elements the authors described 17 MI training session elements . Embedded MI (EG1) and added MI (EG2) differed in seven MI training session elements: integration, temporal order, supervision, location, position of the individual, instruction medium and instruction mode (for more details please refer to Table 3). Nevertheless, the current investigation suggests that the design differences have no influence on the effect of MI to learn the complex motor task. The same review analysed 129 MI interventions with positive changes in the pre to post-intervention assessments regarding their temporal parameters, suggesting an average MI training session duration of 17 minutes. Furthermore, we hypothesise that a MI intervention duration longer than two weeks including more MI training session is more important than the duration of one single MI training session. This hypothesis is supported by the results of the review mentioned above .
As suggested by Driskell et al. (1994), it is important to maintain patients' motivation for a positive overall effect of MI . In our study, some patients in EG2 mentioned that listening to the same tape became less interesting after the fourth time. On the other hand, patients in EG1, in particular patients ≥ 80 years of age, mentioned the difficulty to capture all details and motor task order to imagine during the first two sessions. Both occurrences showed that duration and content play an important role to learn and further use MI independently. Therefore, we suggest implementation of a modified content to be imagined, especially if the motor task to be imagined includes whole body movements more than focusing on one limb only, e.g. make a step with one leg to stand in stride standing.
The motor task
To the authors' knowledge, the motor task 'Going down, laying on the floor, and getting up again' was investigated in stroke patients for the first time. The motor task was modified after the work from Adams and Tyson (2000) . At T0 all patients were able to perform the complete motor task using a chair with no armrests and a thin mat. Pillows were only needed to pad 1) the head while side and supine laying, 2) knees due to temporal pain caused by degenerative joint diseases, and 3) arches of the feet and toes due to a temporally muscle tension increase or stretching of the muscles. All named reasons can be associated to the patients' age and the time period between stroke onset and study entry. As carried out in the current investigation, the motor task did not cause any harm to the patients. On the contrary, in combination with the applied physiotherapy the practiced motor task contributed to a decrease of fear of falling assessed by the Activities-Specific Balance Confidence Scale. The motor task seems to be feasible and practicable to be learned and performed by stroke patients. Therefore, for further motor task practice, we recommend using only seven of the 13 stages listed in Table 2. For both motor task related assessments as well as time and help needed, all raters showed a high inter-rater reliability. Furthermore, scoring the help needed to perform the motor task using the independence levels of the CMSA activity subscale was reasonable. The lower the assistance a patient required (higher CMSA level) the closer was her/his performance to healthy individuals . As expected, patients' level of help needed changed over time and was adapted to the actual situation according to the CMSA guidelines. Primarily, help was needed if the patients did not know how to proceed to the next stage of the motor task or if the therapists had safety concerns. We did not expect that the help provided reduced the time needed to perform the motor task compared to an independent motor task performance.
Motor imagery ability
Scoring for the visual and kinaesthetic subscales at PRE are comparable with published data of stroke patients by Malouin and colleagues in 2007 . All three groups started almost at the same visual MI ability level. As expected, both MI integration approaches helped to improve patients' visual MI ability from PRE to T1. In general, kinaesthetic values were lower than visual values but patients in CG scored lowest at PRE. At T1 both experimental groups decreased, whereas CG increased the scoring. At FU EG2 and CG decreased the kinaesthetic scoring almost to the same value but EG1 increased the MI ability to a higher level than at PRE. We hypothesise that those patients in EG1 and EG2 learned to clearly distinguish between visual and kinaesthetic MI during the investigation. Therefore, they were able to show the difference in the scoring at T1 and FU. Contrary, not all patients in CG were able to differentiate to the same amount as in EG1 and EG2. This indicates that patients might have to be asked at all measurement events if they can differentiate between visual and kinaesthetic MI. The application of the Imaprax software before administering the KVIQ clearly helped to determine the patients' preferred MI perspective. It serves as basis for the use of the first person perspective during the KVIQ. Overall, patients in EG1 were able to improve their kinaesthetic MI ability at FU, whereas patients in EG2 got worse.
The decision to extend the study sample up to 15 patients per group was based on two reasons: Firstly, based on previous therapy intervention studies in our clinic a high drop out rate was expected. Secondly, MI interventions based on previous motor imagery studies published by Page et al. (2001) and Liu et al. (2004) [4, 7] reported high effect sizes. Unfortunately, other researchers conducting MI intervention studies at the same time as the current pilot study reported no effect of their motor imagery interventions [11–13, 34]. To not to underestimate or overestimate the effect of MI the pilot study sample has been raised to obtain more detailed data providing sufficient information for a subsequent Phase III study.
Based on the classification by Thabane et al. the pilot study outcome can be classified as feasible with modifications . Results of the current investigation have to be interpreted with caution due to the following limitations: Firstly, the sample size in all three groups was too small consequently increasing the risk of a type II error. Secondly, notwithstanding the randomised group allocation, patients in the three study groups were not comparable in all baseline characteristics. Though randomly allocated, patients in CG experienced significantly more falls since stroke onset and needed more time to perform the motor task than both experimental groups. Furthermore, CG showed the lowest scoring in the Berg Balance Scale and the Activities-Specific Balance Confidence Scale. Therefore, CG had the highest potential to improve their outcomes, in particular, their motor task performance. Due to the small sample size for each group statistical analyses corrected for baseline imbalances would not have been appropriate. A motor impairment assessment, e.g. the CMSA, would have added a better description of the patients' functional status at study entry. This has been omitted due to the already long duration of up to three hours of the measurement events. Thirdly, the motor task including whole body movements might have been too complex for stroke patients to imagine. Published successful MI investigations had chosen single limb or bimanual movements, e.g. turning a page, grasping a cup, and hang out laundry [5, 7]. Klausler (1991, cited in Jarus, 2000) pointed out that older adults pay more attention to irrelevant task details or could have problems with the information organisation . Therefore, we propose to cut a complex motor task that involves the whole body into shorter pieces to be imagined and give the patient the opportunity to add piece after piece to a consolidated motor task part for forward and backward chaining.
Finally, the MI assessments Imaprax and KVIQ at BL and T0 were used as familiarisation sessions to learn how MI works and can be used. More effort should be undertaken to prepare the patient for a MI intervention, e.g. make sure that patients know the difference between visual and kinaesthetic imagery and can distinguish between internal and external MI perspective.
Recommendations for further MI investigations
An appropriate sample size of a comparison of embedded and added MI would be 33 per group if time needed to perform the motor task (continuous data level) would be chosen as primary outcome measure (see section 'Type II error and sample size calculation' above). If help needed to perform the motor task would be chosen as primary outcome measure (ordinal data level) a much larger sample size would be required suggesting a multicentre study design. We suggest replacing the Berg Balance Scale with the CMSA to perform a group allocation based on stratified randomisation to correct for imbalances in patients' motor function. Regardless their motor function level, patients were well adapted to maintain balance in different positions and situations assessed with the Berg Balance Scale. Patients with a low motor function level achieved a Berg Balance Scale scoring above 45 points, which is an indication that they are safe in independent walking despite their low motor function level . Furthermore, a detailed MI ability assessment and MI familiarisation sessions should be administered to enable the patient to know important MI training session elements, e.g. distinguishing between visual and kinaesthetic MI modes and an internal or external MI perspective. For both MI integration approaches it is proposed to include a progression of the content if a complex motor task will be investigated. A clear description of the implemented MI training session elements and temporal parameters would be helpful to interpret study results within available literature.