Precision/Trueness (P90-P10)/2

JuKo

New Member
#1
Hello,

for my thesis I am comparing 5 dental 3d intraoral scanner with a reference model. The study model is very similar to this study: https://ijcd.quintessenz.de/ijcd_2019_01_s0011.pdf

In this study they are using (P90-P10)/2 in the statistics wich is in my opinion the semiinterquartile range of P90-P10, if I am right.
I made 12 scans for each scanner. That makes 60 scans in total.
For each scan I have around 7000 values.
Now my Question is how do I calculate Precision and Trueness, like it is done in the paper.
The statistic program i am using is SPSS V.26.
Do I need to do additonal 3D measurement, where I compare only the scans of each scanner with each other, to calculate the precision or are the values of the deviation to the reference model enough?

Thank you in Advance.
 
Last edited:
#2
Do I need to do additonal 3D measurement, where I compare only the scans of each scanner with each other, to calculate the precision or are the values of the deviation to the reference model enough?
You can follow 2 approaches at least:
1. In this method, all you need is a real model, as the gold standard, with known 3D coordinates of reference points marked on it (the more reference points, the better; but of course you cannot have too many reference points). You should carefully measure the 3D coordinates of those reference points (such as cusp tips of different teeth, contact points, etc) within a 3D frame. Then you will need to compare each digital point (for example the coordinate of the cusp tip of left canine) with its counterpart real-model (gold standard) reference point in all three dimensions, and compute the error as a euclidean distance between the coordinates of the digital point and its counterpart real-model point. Then you can calculate the average of these errors, and present the average and standard deviation of errors for that particular digital scanner, as its mean(SD) scanning error.

2. The approach of the paper: You can first digitally scan the real cast with a famous device that is already accepted as accurate. Then throw away the real case, and regard the digital reference cast as the gold standard. Is this acceptable? I don't think so, because the reference digital scanner would of course introduce a lot of error. But like this paper, lets assume that it has zero or minimal error. In this case, you need to compare each of those 7000 points (or a select handful of them, again like cusp tips, etc, etc) in your new digital models, with the counterpart point within the digital gold standard model. You should not compare each digital cast with all other casts; all casts should be compared only with the reference cast (assuming that the reference cast accurately models the real cast and that it has no error). Then you can calculate the precision and accuracy and trueness etc based on the euclidean error in each of those 7000 points, and summarizing the error statistics of all 7000 points as the mean, SD, etc, the way described in the paper, or the way described here: https://www.cherrybiotech.com/scientific-note/accuracy-and-precision-in-measurements