We address the problem of knee pathology assessment by using screw theory to describe the knee motion and by using the screw representation of the motion as an input to a machine learning classifier. The flexions of knees with different pathologies are tracked using an optical tracking system. The screw parameters which describe the transformation of the tibia with respect to the femur in each two successive observation are represented as the instantaneous screw axis of the motion given in its

more »

... otion given in its Plücker line coordinate, along with its corresponding pitch. The set of screw parameters associated with a particular knee with a given pathology is then identified and clustered in R 6 to form a "signature" of the motion for the given pathology. Bone model and two cadaver knees with different pathologies were tracked, and the resulting screws were used to train a classifier system. The system was then tested successfully with new, never trained before data. The classifier demonstrated a very high success rate in identifying the knee pathology.