Knee Adduction Moment (KAM)

Estimation of Knee Adduction Moment (KAM) from wearable IMUs during walking.

Nodes Required: 7

• Sensing (7):

  • left foot (top of the foot, with switch pointing towards body)

  • right foot (top of the foot, with switch pointing towards body)

  • left shank (Midway between Femur Lateral Epicondyle and Fibula Apex of Lateral Malleolus)

  • right shank (Midway between Femur Lateral Epicondyle and Fibula Apex of Lateral Malleolus)

  • left thigh ( Midway between Femur Greater Trochanter and Femur Lateral Epicondyle)

  • right thigh ( Midway between Femur Greater Trochanter and Femur Lateral Epicondyle)

  • pelvis (Midway between Left and Right Anterior Superior Iliac Spine)

• Feedback (0)

Algorithm & Calibration

Calibration Process:

No initial static calibration is performed to compensate for misalignment with the segment, so the user should be standing upright when starting the trial.

ML Model Statistics

Our KAM app has two modes, real-time and offline. Due to limited computational resources, the sage system cannot run the full-size model in real-time. Thus, the real-time model has a reduced size and lower performance.

• Real-time model with size reduction: RMSE = 0.9 (%Body weight * Body height), r = 0.79

• Offline full-size model: RMSE= 0.6 (%Body weight * Body height), r = 0.87,

RMSE is the root mean square error, and r is the correlation coefficient.

Citation Information

Please cite this paper IMU and Smartphone Camera Fusion for Knee Adduction and Knee Flexion Moment Estimation During Walking

For convenience, here is the BibTeX Citation:

@ARTICLE{9826418,
  author={Tan, Tian and Wang, Dianxin and Shull, Peter B. and Halilaj, Eni},
  journal={IEEE Transactions on Industrial Informatics}, 
  title={IMU and Smartphone Camera Fusion for Knee Adduction and Knee Flexion Moment Estimation During Walking}, 
  year={2023},
  volume={19},
  number={2},
  pages={1445-1455},
  keywords={Knee;Cameras;Foot;Estimation;Feature extraction;Sensors;Legged locomotion;Deep learning;joint kinetics;osteoarthritis (OA);portable sensing},
  doi={10.1109/TII.2022.3189648}}

Description of Data in Downloaded File

Calculated Fields

• time (sec): time since trial start

• weight_kg: the users weight in kilograms as entered on the app configuration panel

• height_meter: the users height in meters as entered on the app configuration panel

• KAM: right knee adduction moment

• Stance Flag:

  • 0 for swing phase or not walking

  • 1 for stance phase

Sensor Raw Data Values

Please Note: Each of the columns listed below will be repeated for each sensor

• SensorIndex: index of raw sensor data

• AccelX/Y/Z (m/s^2): raw acceleration data

• GyroX/Y/Z (deg/s): raw gyroscope data

• MagX/Y/Z (μT): raw magnetometer data

• Quat1/2/3/4: quaternion data (Scaler first order)

• Sampletime: timestamp of the sensor value

• Package: package number of the sensor value