科研之友
微信
新浪微博
Facebook
分享链接
摘要
There are three variables involved in modeling measurement errors - type, severity, and selectivity. Whereas clinicians typically utilize a graphical technique to grade color deficiencies based on D-15 panel tests, Vingrys and King-Smith developed a quantitative scoring technique for panel color tests, which models all three factors by utilizing an overall moment of inertia for color difference vectors (CDVs) calculated in 1976 CIELUV space. We propose a least squares analysis via linear regression of the errors (Delta U, Delta V) according to the following equation: Delta V = m Delta U, where m = slope of best-fit line determined by linear regression. Error type is determined by the angular proximity of the best-fit line to known confusion axes representing protan, deutan, tritan or unspecified color defects. The severity is the sum of the CDV lengths of all errors made, and the selectivity is determined by the adjusted variance of the least squares fit. We determined normative threshold values for type, severity and selectivity by inspecting 142 cap arrangements with tentative diagnoses. We then analyzed 49 standard D-15 cap arrangements of subjects with definitive diagnoses to determine the sensitivity and specificity of our method to defect types. The results were then compared with those of Vingrys and King-Smith. Our linear regression technique provides an improved assessment of error arrangements that represent subtle unspecified color defects. However, our model appears too sensitive to atypical repositioning errors made when the majority of errors lie along known confusions. Used in conjunction with previous quantitative methods, linear regression by least squares proves a useful tool in the classification of errors of D-15 color panel tests.
- 出版日期2011
