Accuracy refers to bullseye closeness, while precision refers to how tightly the darts are clustered. In biomechanics, both are really tricky to measure, so just be aware of what your system can measure and consider that when making conclusions.
One of our internal principles is to validate and test everything using data. Whether we are talking about how our product compares to other products, to previous versions of our own product, or to different technologies altogether, our gut typically tells us to just run the experiments, repeat them a few times, and then we see what we see. Data-driven decision making.
This leads to a couple of interesting concepts that we repeat, both internally and externally, all the time; precision and accuracy. I believe that every single engineer and biomechanist has heard/probably understands these concepts. But because these concepts are so common, I wanted to write a little bit about them and how they apply to this field.
Firstly, let’s digress a little bit to the standard analogy of a dartboard to get some of these definitions going. If we assume that the bullseye in darts is the highest score on the board, accuracy refers to the distance from each dart to the bullseye. As the collection of darts gets closer to the bullseye, they are more accurate. Precision is a bit trickier, and refers to how consistently you are throwing. So if, at the end of the round, the darts are clustered tightly, that is more precise than a larger cluster of darts, even if the larger cluster is closer to the bullseye. In other words, precision doesn’t care about the score you hit on the board, but just how close each dart is to each other.
Now, practically, how does this translate into biomechanics? Well, it’s actually pretty close to the dartboard analogy, but is slightly nuanced. Assuming we are talking about human pose estimation, accuracy refers to how close the measured pose is to the true pose. The sources of error between these two could be intrinsic system accuracy, sensor placement/movement (if you are using sensors), or how well the measurement technique actually translates to pose, to name a few. The precision of the system is just as important, and in the context of biomechanics is also nuanced. One way to imagine it is the following: a person comes to be measured on adjacent days, and literally nothing changes with their movement, the closeness of pose from these two measurements will refer to the precision of the system (now this doesn’t tell the whole story because there is inter and intra-rater reliability, but it’s a good start).
Given these definitions, it’s actually easy to see the importance of these two factors in the context of human movement measurements. So, which systems are accurate and which are reliable? Sadly for accuracy, we actually don’t know (shocking, right?). Despite marker based motion capture systems being grandfathered as the gold standard in every way, it’s not safe to assume that their measurements are perfectly accurate. Marker placement and marker movement on soft tissue are two issues alone that make this assumption questionable. There are biplanar x-ray systems that measure the position of the underlying bone from two x-ray views (while the subject is moving) and those can achieve mm accuracy, but their challenge is that the volume size is so small. So, if we are measuring what is happening at the knee, we can’t be measuring what is happening at the ankle. Looks like for accuracy, we are stuck between a rock and a hard place!
On a positive note, we can kind of measure the precision of a system by simply doing repeated measurements. Unfortunately in biomechanics, there is one critical factor to consider, that humans themselves aren’t entirely precise in their repetition of movements! So, even for a remarkably repeatable task like walking, there is intrinsic variability between strides; we don’t walk every stride identically!
When I encounter situations like this, I like to take a step back and think practically. How about a simple experiment that reflects an actual use case in biomechanics. A person comes in on subsequent days, and we measure them twice. Nothing has changed with the person, so we should measure the same thing as long as the underlying motion is fairly repeatable. If we do, then our system is precise! This is the important test that we should be considering when assessing the suitability of a system when measuring a particular movement. If we can’t do this, then there is no point in continuing, either the system isn’t good enough, or the action we are measuring is not repeatable. In some sense, if the objective is to measure a longitudinal change, then reliability may actually be more important than precision, because the goal is to measure change, not absolute values.
Basically, what can I say, this stuff is tricky and conflated by the fact that we are imperfect as humans! My only real advice here is to make sure you understand these concepts and just be aware of them as you collect data and make recommendations/conclusions from these measurements.
Comments