4.1. Functions
The cleaning and data pre-processing stage had to deal with routine and simple tasks such as removing incomplete exercises,etc.
At the same time, though, it had to tackle more complex tasks that required the use of expert knowledge. A series of functions including expert knowledge were created to perform these tasks:
Firstly, the strength curves are preprocessed in order to eliminate inertia peaks, that is, peaks produced by machine inertia rather than by the patient’s actual strength (see Fig. 1, right).
Then, I4 detects exercise extensions and flexions that are invalid because the patient employed much less effort than in others, or movements that can be considered atypical as their morphology
is unlike the others.
The analysis of the strength curves run using the expert functions involves assessing different characteristics of the curve morphology.
These characteristics are what the specialist is interested
in, and they constitute the basis for patient assessment. The evaluated
aspects are:
Uniformity: how similar the exercise repetitions are.
Regularity: whether the curve has a smooth contour or a lot of
peaks.
Acceleration: a qualitative assessment of the time the patient
takes to reach the maximum strength value.
Troughs: prolonged drops and rises in the exercise strength
value.
Shape of the curve: overall assessment of the shape of the curve
based on the effort applied at the central angles (around 45),
the flattening of the curve and the angle at which each maximum
strength value is reached.
The design and implementation of the expert functions can be
described as an iterative and interactive induction process. Given
a number of strength curves, the expert evaluated each one and assessed
its characteristics. Then, tentative functions were implemented.
These functions were applied to a new set of curves, and
the results were shown to the expert for evaluation. This evaluation
led to some changes in function implementation, and so on.
This process ended when the functions provided the correct values
in a high percentage of the cases (over 98%). As we will see later,
these functions turned out to be useful for pre-processing purposes
in phase 2 of the system.