Project-ID: FTI17-014
Project start: 01. März 2019
Project end: will follow
Runtime: 36 months / finished
Funding amount: € 170.000,00

Brief summary:

Diseases and injuries of the musculoskeletal locomotor system, as well as neurological disorders, can lead to pathological impairments of human motor function. To better understand gait impairments resulting from musculoskeletal or neurological disorders, it is important for clinicians and therapists to describe and analyze the patient’s gait pattern accurately. For this purpose, gait analysis has become a crucial assessment tool. Currently, the clinical three-dimensional gait analysis (3DGA) represents the “gold standard” in clinical gait analysis. 3DGA is used to quantify the mechanical processes of a patient's locomotor system from a kinematic and kinetic point of view. In everyday clinical gait analysis practice, clinicians and therapists examine a large number of patients. Often, specific cases turn out to be very similar to previously examined and treated cases. Information about the course of therapy and the associated treatment outcomes of historical cases could support clinicians and therapists during the examination of new patients. Medical history, 3DGA data, and information about treatment outcomes are nowadays stored in databases for documentation and evaluation, which have been compiled over the past decades. 3DGA databases can, however, hardly be examined manually in their entirety and the search for similar reference data is usually not possible. These databases implicitly contain a vast amount of valuable clinical knowledge, which is currently not exploited.
Automatic analysis methods bear the potential to provide a novel, efficient and objective way of accessing and making use of medical databases. The primary aim of this project is to apply data mining and machine learning to measurement data derived from clinical 3DGA to support clinical practice in gait analysis and decision-making. Therefore, the two main objectives are: (1) designing automatic classification algorithms that can robustly differentiate between a large range of gait patterns, e.g., different pathological patterns and healthy gait; (2) developing retrieval methods for the detection of similar gait behavior and associated diagnoses from large-scale clinical 3DGA databases. To meet the proposed objectives, higher-level feature representations should be learned automatically from the data to obtain abstract signal representations. We expect such representations to be more robust and expressive than currently used gait representations based on distinct spatio-temporal parameters, and Principal Component Analysis (PCA).
In the proposed project, an interdisciplinary group of researchers from the areas of computer science, physiotherapy, and biomechanics will closely collaborate to establish advanced analysis methods for modeling, classification and similarity retrieval of gait patterns. The developed methods will enable novel data-driven ways to access and analyze 3D gait databases and should lay the foundations for future clinical decision-support systems.