Reliable assessment of infant motor patterns and cognitive development is especially important during early infancy, as it is during the first year that infants set the foundations for their upcoming life. Typically applied clinical methods are various assessment scales, tests and questionnaires, which can lack objectivity and precision. In the field of infant posture and movement analysis, clinical methods are often used in combination with sensor-supported measurement systems, such as video cameras, optoelectronic measurement systems, inertial measurement systems, force plates, and pressure mattresses. These approaches are usually more accurate, reliable and objective, but can be subject to shortcomings, such as view angle imitations, self-occlusion, wiring, and complex system setup procedures. CareToy project Consortium has recently proposed and developed a dedicated sensor-supported gym for stimulation and assessment of infant activity.
First part of dissertation focuses on presentation of technical and clinical aspects of Care-
Toy environment. The technical part provides a description of the included sensor modules,
comprising pressure mattresses, IMUs, video cameras, and sensorized toys. Individually
these sensors were already used successfully, but applications of the proposed sensor combination are rare. The clinical part of CareToy environment gives a description of the clinically supported infant rehabilitation protocol, focusing on training goals and training scenarios.
These were prepared for various body positions in accordance with general developmental
milestones of infants.
Second part of dissertation provides a description of the developed data processing and
sensor fusion algorithms. Data analysis flow is presented sequentially, ranging from raw, unprocessed data to final numerical parameter results. First are presented data pre-processing and sensor fusion algorithms of pressure mattress and IMU data, addressing the shortcomings of the newly proposed sensor combination, such as pressure mattress data bias, superposed noise, and possible trunk IMU displacements. Various methods are described, such as unscented Kalman filtering, pressure imprint data moment calculation, and weighted valuation, whereas two separate versions of sensor fusion algorithms are prepared for real-time processing and the post-processing approach. Following this, the development of sensorsupported computer model for head movement analysis is presented. The model includes line-of-sight algorithm, head-tracking algorithm, and two-dimensional intensity profile analysis.
Furthermore, methods for motor pattern parameter assessment are provided, comprising
analysis of rolling and toy grasping activity, as well as postural stability evaluation.
Finally, a description of algorithms for body parts recognition using colour space filters in
video recordings of infants is given. Results confirm accuracy and suitability of proposed
algorithms on the selected data set of video recordings, but reveal certain shortcomings, such as limited view-angle of cameras and self-occlusion.
The third part of dissertation covers validation of the proposed sensor system for infant
motor pattern parameter assessment. First the preliminary study, focusing on using the
measurement system, namely combination of pressure mattress and IMUs, for movement
analysis of healthy infants is presented. Results were acquired by means of real-time data
processing and sensor fusion algorithms. Evaluation was focused on determination of optimal combination of data processing algorithms and assessment of suitability of the developed computer model for infant head movement analysis. Data comparison to results of the reference video review confirmed reliability and accuracy of the developed computer model in combination with sensor data processing algorithms, successfully detecting all performed head movements. The sensory fusion algorithms contributed to the robustness and precision of the system, while the numerous statistical parameters provided a complete description of head imprint position and displacement data distribution. Following this, validation of
CareToy system and implemented algorithms was performed with a reference optoelectronic
measurement system Optotrak and a dedicated baby doll. Data post-processing and sensor
fusion algorithms were applied on pressure mattress and IMU data for calculation of
numerical parameters. Study comprised influence evaluation of measurement system simplifications by using only 1 IMU per each arm on the accuracy of motor pattern parameter estimation and additional accuracy validation of the developed computer model for head movement assessment. To support data of the validation study, a pilot study on a healthy infant was performed. Finally the randomised-controlled trial study (RCT) is presented. Motor pattern parameter values were calculated using the developed sensor data post-processing algorithms and were provided for trunk movement analysis, forearm posture and movement evaluation, toy grasping analysis, and COP movement assessment. Statistical analysis of correlation among numerical parameters and AIMS clinical assessment scores was performed, along with statistical comparison of data for various training goals and motor ability levels.
Study results verify that the proposed measurement system in combination with the developed sensory data processing and fusion algorithms, and the chosen numerical motor
pattern parameters is appropriate as a device for stimulation and simultaneous assessment
of infant activity. The applied techniques remove the effects of aforementioned drawbacks
well. The chosen numerical parameters are appropriate for description of infant activity and
discrimination of infant stimulation-based responses. Correlation of numerical parameters
and clinical scores implies adequacy for full, all-round description, as well as single motor
pattern subfield evaluation. Results confirm suitability of designed training scenarios for infant activity stimulation, including trunk rotation, arm posture, reach-to-grasp manoeuvres,and posture stability subfields. The combination of the proposed CareToy gym and developed data processing methods thus has great potential and represents an important step on the route towards developing an objective, accurate tool for unobtrusive sensor-supported assessment of infant motor pattern development.