Table 1 Summary of results from reviewed articles and identified limitations

Nath et al., 2017 [36]

Smartphone IMU

1: upper arm

ER: 0–6%

2: waist

ER: 0–1%

No real-time feedback

Validation of built-in smartphone IMUs to measure workers’ postures and identify risks

Calculated postures close to observation-based methods; reliable method for identifying postural risks and trunk flexion

Comparability limitation: Only tested in the context of 16 screw driving scenarios in one worker.

More variable error rate in arbitrary position readings.

O’Sullivan et al., 2012 [17]

BodyGuard: strain gauge

From spinous process of L3 to S2 calibrated to individual based on %ROM

- correlation to digital fluoroscopy: sitting vs standing r² = 0.94 vs 0.88

Real-time biofeedback (auditory or visual)

Validation of BodyGuard for analysis of vertebral motion in the sagittal plane (n = 12)

Slight and consistent underestimate of lumbopelvic flexion; validated method for use in laboratory and clinical settings.

Outcome limitations: Further validation required for use in individuals with low back pain

Bhattacharya et al., 1999 [18]

Ergonomic dosimeter

Trunk and upper dominant arm (housed in coveralls)

- Output stratified into risk categories based on ROM magnitude

No real-time feedback

Validation of system to measure postural angles of torso and upper arm in sagittal plane (n = 2)

Reliable system for the continuous monitoring of postural data in carpenters on construction sites

Selection bias: Small cohort not representative of general population

No data on correlation to video data during scenarios

Plamondon et al., 2007 [15]

Hybrid system: two IMUs linked by potentiometer

IMUs:

1: S1 (pelvic analysis)

2: T1 spinous process (thoracic flexion/lateral flexion and torsion)

- Error in degrees: 2.7, 1.9 and 5.2 respectively

No real-time feedback

Validation of hybrid system for 3D measurement of trunk posture; analysis of utility of potentiometer to increase validity (n = 6)

Root mean square error less than 3 degrees for forward- and lateral-flexion; potentiometer required when magnetometer signals corrupted

Comparability and outcome limitation: Error increased in long-duration dynamic tests (30 min) vs. short-duration (30 s) particularly without magnometer

Faber et al., 2009 [37]

MTx IMU System

1: sacrum

2: T9

3: movable (between 1 and 2)

Peak error rate ¬5 degrees

No real-time feedback

Determination of the possibility and optimal location of a single sensor for trunk inclination measurement (n = 10)

Optimal inertial sensor location for trunk inclination measurement 25% of the distance from the midpoint between the PSISs to C7 and was hence different to each subject.

Comparability bias: tested with straight legs; flexion of knees may impact trunk inclination when lifting an object, hence the optimal location may change

Gleadhill et al., 2016 [38]

SABEL Sense IMU

1: C7

2: T12

3: S1

No real-time feedback

Validation of inertial sensors for measurement of resistance exercise movement patterns (deadlift). 11 subjects provided 227 time points to analyse.

Timing validation results demonstrated a Pearson’s correlation of 0.9997 and supportive validity measures; validated for use in resistance exercise

Comparability bias: Only tested in the context of a conventional deadlift with ROM not specified

Yan et al., 2017 [24]

YEI 3-Space IMU Sensor

1: back

2: safety helmet

Real-time auditory alarm

Validation of a personal protective equipment involving IMUs for insecure motion warning

Successful validation of the proposed technology for real-time insecure motion warning

No comparison to analyse accuracy and no formal published output data provided.

Fathi et al., 2017 [34]

Shimmer IMU

1: cervical spine

2: thoracic spine

3: lower lumbar spine

Reported accuracy rate of 100% across pre-defined stages of ankylosing spondylitis

Reported real-time feedback but mechanism of the same not detailed

Proposal of wearable system able to detect spinal displacement and provide real-time warnings

System classification performance validated in differentiating between two incorrect postures (hunch back, slouch back)

Selection bias: Only evaluated in four subjects, no information regarding their health or tasks performed was provided

Abyarjoo et al., 2015 [14]

PostureMonitor: YEI 3-Space IMU Sensor

Attached to upper back of the user’s garment

Real-time auditory alarm

Verification of the PostureMonitor for the detection of poor posture and development of good postural habits

PostureMonitor reported sensitive as to detect and warn of poor posture.

Outcome limitation: further testing required for validation, long-term testing required to assess the impact on the development of good postural habits

Cajamarca et al., 2017 [5]

StraightenUp: LilyPad Accelerometer ADXL335

Sensors attached to a brace:

1: upper trunk

2: central trunk

3: lower trunk

Precision rate across different pre-defined positions ranged from 99 to 100% (n = 9000)

No real-time feedback

Verification of StraightenUp for measurement of spinal posture and assessment of user experience

(n = 30, 9000 encounters)

Preliminary verification of postural classification; reported to be comfortable but difficult to apply; user preference for vibrotactile or smartphone notification for poor posture alerts

Outcome limitation: Further testing required for validation; device requires adaptation to become more user friendly

Not tests in real life setting

Valdivia et al., 2017 [39]

IMU MPU-9250 sensor

Sensor strapped to elastic band worn at the waist

Real-time feedback via exergame

Comparison of IMU sensor with Microsoft Kinect V2 for the use in a proposed exergame aimed at improving spinal posture

IMU more accurately but less reliably measures range of motion of the spine in comparison with the Microsoft Kinect V2; IMU exergame less engaging

Selection bias: Comparison of IMU and Microsoft Kinect between different subjects in an already low sample size

Wang et al., 2016 [26]

Zishi: 9-axis Adafruit IMU sensor

Two sensors within a vest:

1: T1

2: T5

Root mean square error range 2–5 degrees

Real-time visual and auditory feedback via Android app

Development, validation and incorporation of the Zishi in postural analysis and correction

Fifth iteration for the Zishi vest provided highly mobile smart textile for postural analysis

Outcome limitation: Further validation studies recommended; expansion to measure aspects of spinal posture (e.g. lumbar region) useful for better analysis of posture

Tanaka et al., 1994 [12]

Electromagnetic inclinometer LP06F1F1AA Murata

1: chest

2: thigh

3: leg

No real-time feedback

Proposal of wearable system for long-term measurement of human posture

Device able to record postural changes with an angular resolution of 12 degrees. No accuracy or error data provided.

Outcome limitation: Angular resolution inadequate for precise measurement; limited to sagittal plane

Wong et al., 2008 [33]

IMU: one tri-axial accelerometer and three uni-axial gyroscopes

Sensors strapped with elastic:

1: T1/T2

2: T12

3: S1

Error rate in postural assessment: < 3 degrees in sagittal and coronal planes, ICC > 0.829

Real-time auditory alarm

Proposal of posture monitoring system able to estimate spinal curvature changes in sagittal and coronal planes and provide postural analysis (n = 9)

Preliminary verification indicated high correlation with motion analysis system; verified for remote monitoring of trunk posture during daily activities

Outcome limitation: Lack of magnetometer did not allow for estimation of trunk rotation in transverse plane

Xu et al., 2017 [40]

9-axis IMU: MPU-9150 InvenSense

Eight IMUs placed symmetrically on left and right sides of torso at L4/L5

Real-time vibrotactile feedback

Proof-of-concept of wearable system for real-time postural balance and gait retraining using vibrotactile feedback (n = 4 and 6 in 2 studies)

Device able to monitor trunk tilt and provide meaningful vibrotactile feedback

Outcome limitation: Further testing required for validation as the current study was a proof-of-concept; battery life of IMUs only 1.5 h

Raw error rates not provided.

Bazzarelli et al., 2003 [28]

Hybrid system: electromagnetic technology and Analog Devices ADXL202 biaxial accelerometer

1: left scapula

2: right scapula

RMS error 1%

Real-time vibrotactile feedback

Proposal of hybrid system to replace braces in the correction of adolescent idiopathic scoliosis (n = 6)

Preliminary verification of hybrid system for monitoring progress and correction via biofeedback in adolescent idiopathic scoliosis with good sensitivity.

Outcome limitation:

Further testing required for validation. No current data in real user.

Dunne et al., 2008 [41]

Plastic fibre-optic goniometer

Markers placed on C7, T4, T7, T10, T12, L2, L4 + spines of left and right scapulae

No real-time feedback

Validation of plastic optical fibre sensor for monitoring seated spinal posture, as compared to visual analysis (n = 9)

Significant accuracy error ranging across 14.5% of the magnitude of the average range of motion of subjects

Outcome limitation: Further testing required for validation in clinical contexts

No error data provided.

Motoi et al., 2006 [42]

IMU: accelerometer and gyroscope

1: chest, housed in shirt pocket

2: lower thigh

3: upper calf

No real-time feedback

Proposal of wearable system for monitoring gait speed and angle changes of trunk, thigh and calf in sagittal plane (n = 3)

Preliminary verification of use of the wearable system for dynamic posture monitoring in sagittal plane

Comparison and outcome limitations: Poor wearability with sensors linked by a wire

No error rate comparison

Gopalai et al., 2012 [43]

MicroStrain’s wireless IMU

1: Attached to trunk via waist band

2: wobble board

Real-time vibrotactile feedback

Evaluation of real-time vibrotactile feedback for the warning of poor postural control (n = 24)

Preliminary verification of detection of poor postural control; improved postural control with vibrotactile feedback

Comparability limitation: Less related to spinal posture monitoring and more focused on postural stability using feedback system

Wu et al., 2014 [30]

Accelerometer

Vest containing:

1: below neck

2: chest

3: centre of mass

4: left hip

5: right hip

Angle errors within 0.5 degrees

No real-time feedback

Proposal of using multiple single-axis accelerometers to obtain titling angles

(n = 20)

Wearable system and time-less algorithm proposed verified for real-life applications

Outcome Limitation: Further testing required for validation in the suggested context (Parkinson’s disease) and other clinical contexts

Sardini et al., 2015 [19]

Inductive sensor

Shirt with an inductive sensor sewn to the back and front

Correlation coefficients range from 0.95–0.98 to optical system.

Real-time vibrotactile feedback

Validation of wearable system for monitoring seated posture at home through comparison with optical measuring system (n = 4)

Validated for the use of monitoring seating posture in a variety of functional activities within the home

Outcome limitation: Only measures spinal posture in sagittal plane; further testing in a greater variety of contexts required for wider validation

Tsuchiya et al., 2015 [20]

Flex sensor + accelerometer

Accelerometers (2) placed at upper lumbar spine + sacrum, flex sensors (3) placed between

No real-time feedback

Proposal of wearable system to measure the shape of lumbar skin to identify lumbosacral alignment changes in 3 positions xray (n = 4)

Lumbosacral alignment and lumbar load accurately estimated using wearable system

Comparability limitation:

Less related to spinal posture monitoring and more focused on lumbosacral dimension estimation

Miyajima et al., 2015 [44]

Six-axis IMU: accelerometer and gyroscope across knee, hip and spine.

1: lumbar spine

2: thigh

3: calf

Mean angle error < 3.5 degrees across sensors

No real-time feedback

Verification of wearable system for monitoring lumbar torque through comparison with optical capture system (n = 1)

Estimation error of lumbar joint torque < 11 Nm based on inclination angle data; preliminarily verified.

Comparability limitation: Assumption that all angles were at 0 degrees when subjects were standing straight. More subjects needed.

Petropoulos et al., 2017 [6]

SPoMo: six-axis IMU (accelerometer and gyroscope)

1: upper back

2: lower back

Mean square error range: 0.001–0.05

Real-time vibrotactile feedback

Proposal of SPoMo for the real-time automatic monitoring of spinal posture in sitting

Average mean square error suggests SPoMo is a reliable tool for monitoring sitting spinal posture

Comparability limitation:

Accumulated error due to gyroscope drift, requires well refined calibration and filtering of data for long-term use

Lou et al., 2012 [29]

Smart garment: IMU (three-axis accelerometer and two-axis gyroscope)

1: upper back

2: lower back

Error in static measurements of 2 degrees

Real-time vibrotactile feedback

Verification of smart garment for posture monitoring during daily activities; analysis of efficacy of vibrotactile feedback compared to video (n = 4)

Measurement accuracy within 5 degrees over 90% of the time during daily activities

Outcome limitation:

No indication of whether long-term use with vibrotactile feedback can lead to long-term postural change

Data only on single plane kyphosis measured.

Bell et al., 2007 [21]

Fibre-optic goniometer

L5/S1

No data on error rate or accuracy.

No real-time feedback

Proposal of wearable system using fibre-optic goniometers to identify activities and associated lumbar postures (n = 5)

System reported as comfortable and unobtrusive; motion profiles accurately identified work-related activities and quantify lumbar postures

Outcome limitation:

Postural identification is not currently automated in the proposed system, preventing real-time feedback restricting usability.

No comparison data for accuracy.

Ribeiro et al., 2016 [45]

Spineangel: triaxial accelerometer

Attached to belt

Real-time auditory alarm

Investigation of the extent to which the Spineangel can reduce exposure to poor posture associated with low back pain

Within-day measurement error of 5 degrees and between-day measurement error of 8 degrees

Outcome limitation:

Study published was a protocol for the ELF cluster randomised controlled trial, results not yet published

Harms et al., 2009 [16]

SMASH accelerometers

Fixed on shirt:

1: C7

2: T10

3: L5

4: scapula

5: shoulder

Absolute sensor error less than 5 degrees in 84% of cases.

No real-time feedback

Validation of system involving accelerometers fixed to shirt to measure trunk inclination in children, as compared with vision-based system (n = 21 subjects across 6 positions)

Single scapula sensor most valuable in assessing Posture based on the least error derived

Comparability limitation: The shirt to which the sensors were affixed was loose fitting, thus allowing sensor movement and subsequent error particularly in setting of head movement and significant trunk flexion.

Leung et al., 2012 [23]

Limber: accelerometer, IMU, strain gauge

Accelerometers: shoulders + IMUs: spine and neck, contained in hoodie; (stretch sensors on wrist)

Game-like positive and negative feedback regarding posture on computer

Proposal of two prototypes to encourage maintenance of good posture whilst sitting over the duration of the workday (n = 4)

Enable a minimally disruptive and highly engaging method for monitoring and correcting poor posture in an office-style workplace

Outcome limitation: Concerns with comfort, aesthetics and incorporation with work protocol; further testing required for validation

No formal data provided.

Hermanis et al., 2015 [46]

9 axis IMU: accelerometer, gyroscope, magnetometer

Sensors contained within a 7 × 9 grid that is attached to the back of a vest

Real-time visual feedback via Android app

Proposal of Wearable Sensor Grid consisting of IMUs to monitor posture

No validation testing conducted

Outcome limitation:

With no validation published as of yet, this remains a prototype with unknown validity

Giansanti et al., 2009 [3]

IMU: 3 uniaxial accelerometers, 3 gyroscopes

Sensor mounted at L5 (close to centre lf mass)

Real-time auditory feedback; sound volume correlating with degree of flexion

Proposal of using wearables and auditory feedback to improve postural control (n = 9)

Reported improvement in balance and decrease in energy expenditure with use of this auditory biofeedback wearable system

Comparability limitation: Specific auditory feedback requires intact hearing in users, this may limit use of this device in the elderly and those with hearing deficits; less related to spinal posture and more to postural control

No data on sensor accuracy,

Millington, 2016 [22]

Lumo Lift: IMU sensor: tri-axial accelerometer, gyroscope, magnetometer

Lumo Back: accelerometer

Prana: sensor measuring posture and breathing

Lumo Lift: worn under clothes under the clavicle

Lumo Back: waist

Prana: waist

Lumo Lift: real-time vibrotactile feedback

Lumo Back: real-time monitoring through smartphone app

Prana: push alert reminders to sit/breath better and real-time monitoring through app

Qualitatively assess commercial wearables available for postural analysis

Haptic surveillance of posture enables shared responsibility of postural monitoring

Outcome limitation: Qualitative analysis of these devices, therefore no validation on the accuracy and validity of these devices in various clinical contexts

Felisberto et al., 2014 [13]

BodyMonitor: IMU: tri-axial accelerometer, gyroscope, magnetometer

1: upper torso

2: hip

3: leg

No formal error rate, however was capable of detecting 70% of “incorrect activity” definitions.

No real-time feedback

Proposal of monitoring posture in the elderly with aim of decreasing premature nursing home admissions (n = 5, across multiple movement and orientation states)

Verification of using the wearable system for the identification of various body postures

Outcome limitation: Further testing required for validation; only tested identification of poor/good posture whilst sitting

Lin et al., 2016 [47]

Microelectro-mechanial tri-axial accelerometer

1: lower cervical spine

2: middle of the chest

3: L3 (centre of mass)

4: right waist

5: left waist

Error rate in previously published work from group of 0.466 degrees.

Real-time visual feedback via smartphone app

Proposal and validation of wearable system incorporating five sensors affixed to a vest for real-time posture monitoring

Wearable system is comfortable, washable and easy to wear; all proposed functions of the system were validated

Selection bias: Tested in elderly subjects with the smartphone app driving technology anxiety. Total subjects not provided.

Voinea et al., 2016 [48]

IMU

Five sensors affixed to shirt in midline running from upper thoracic to lower lumbar spine

No real-time feedback

Proposal of model that converts orientation angles from the wearable system to calculate the curvature of the spine

Maximum error percentage < 5%, proposed mathematical model validated for reproduction of spine curvature; suitable for postural monitoring

Comparability limitation: Only uses one axis from the IMU; development to analyse all axes should further validate this system in kyphosis, lordosis and scoliosis. Total subjects not provided.

Kang et al., 2017 [35]

Smart garment: IMU sensors, metal composite embroidery yarn

IMU sensors: left and right shoulder, left and right waist.

Anterior/posterior direction tilt angle error of less than 4 degrees.

No real-time feedback

Proposal of garment to measure postures; compared with motion capture camera system

Reported reasonable estimate of pitch and roll motion; feasible for postural monitoring

Comparability and outcome limitation; Posture estimates require an algorithm to compensate for the coupling of body motion

Charry et al. 2011 [49]

DorsaVi’s ViMove: IMU sensors (one tri-axial accelerometer, one single axis gyroscope)

1: L1

2: S1

RMS error range 1.9–2.5 degrees across flexion and lateral flexion and 4.1–5.2 degrees for twisting motion.

No real-time feedback

Proposal and assessment of accuracy of ViMove in measuring 3D orientation of lumbar spine (n = 2)

Once the raw inertial signals were processed by the Positional Algorithm there was a “good agreement” with Optotrak System

Selection bias: Only tested on two subjects; further research with a larger sample size required to determine if suitable for clinical use