The data summary and analysis were completed on November 10th, 2022. Figure 1 presents the flow diagram of study selection. Our initial search retrieved 1,027 articles from four databases and 13 additional articles from other sources (e.g., reference lists from original work and review articles). After the removal of duplicates and the screening of title, abstract, and full text for study design and outcomes, 26 original research articles were eligible and included in the systematic review. Reasons for exclusion in this phase included: not related to gait, not an fMRI study, prototype description only, or not an MRI-compatible device with mechanical structure enabling the control or adjustment of stimulation parameters.

The RISMA flowchart of publication screening. Abbreviation PRISMA: Preferred Reporting Items for Systematic Reviews and Meta-Analyses; fMRI: functional magnetic resonance imaging

A total of 13 different devices were reported in 26 total studies, including a foot-sole stimulation system [20], a dual-drive foot-sole stimulator [21], a Korvit boot system [22,23,24], a foot pedal manipulandum [25], a plantar flexion force measure apparatus [26], a brain discovery pneumatic orthosis (Bra.Di.P.O.) [27, 28], a bipedal device (i.e., one pedal for each of left and right foot) [29], a torque-measuring apparatus [30], a pedaling device [31,32,33], a pseudogait-magnetic resonance compatible device (pseudogait-MRCD) [34, 35], a cylindrical treadmill device [36], a magnetic resonance compatible stepper (MARCOS) [37,38,39,40,41], and a lower-extremity motion simulator (LOMS) [42,43,44,45] (Table 1).

The structure of included devices can be divided into three main parts: the power supply unit, the execution unit that is oftentimes placed in the MRI scan room, and the control unit placed outside the scan room (Fig. 2; Table 2).

The conceptual diagram of device structure

For power supply, four devices used an air compressor to drive a pneumatic actuator [20,21,22,23,24, 42,43,44,45], while such information is missing for the other nine.

The execution units consist of three structures: support to the lower limb, sensor, and actuator. Types of lower limb support included a plastic boot (number of devices (N) = 3) [20,21,22,23,24], syringes (N = 1) [26], a customized pedal (N = 5) [25, 27,28,29,30,31,32,33], a special treadmill (N = 2) [34,35,36], and a customized lower limb exoskeleton (N = 2) [37,38,39,40,41,42,43,44,45]. Sensor types included force/torque sensors (N = 2) [26, 30], displacement sensors (optical encoders [31,32,33,34,35] or a rotary potentiometer [42,43,44,45])(N = 3), and the combination of at least two types of sensors (e.g., the Bra.Di.P.O. used both a custom-built analogue optical encoder and a pressure transducer [27, 28])(N = 3) [27,28,29, 37,38,39,40]. Five devices did not report sensor information. For the types of actuators, pneumatic actuators were used in seven devices [20,21,22,23, 27, 28, 36,37,38,39,40,41,42,43,44,45], while such information was missing for the other six.

Control units consisted of valves, a controller, a data acquisition system, and a user interface. The details of valves were provided in five devices [20, 21, 27, 28, 37, 38, 42,43,44], and that of controller in four [20, 21, 37, 42, 44], data acquisition system in five [29,30,31,32, 34, 37], and user interface in six devices [20, 21, 30,31,32, 34, 37] (See Table 2 for details).

In order to prevent unwanted task-related movement of the body as well as the extraneous motion of head that may interfere with image quality, all devices implemented different accessories, such as Velcro straps or belts (N = 6) [26, 29,