ROC curve how to determine the diagnostic threshold

There are diagnostic indicators, diagnostic results, SPSS can be directly made ROC curve, do not need to calculate their own 1-spe and sen. Different versions of the ROC curve location is not the same, look for analysis or graphing two menus can find the ROC curve of this command.

ROC curve closest to the upper left of the critical value of the point for the best critical value. Generally, statistical software will provide a table of all the critical values, each critical value corresponds to a different sensitivity and specificity, and then calculate the critical value corresponding to the point with the largest Jordon index, which is the final result.

Properties of ROC Curve

(1) The change of β value is independent of the change of d', and when examining the effect of the change of β value on P(y/SN) and P(y/N), it is found that: when β is close to 0, the probability of hitting is almost 0, i.e., the signal is all accepted as noise; when β is close to infinity, the probability of false alarm is almost 0, i.e.

The curve reaches an optimal standard βOPT? somewhere.

(2) The curvature of the ROC curve response sensitivity index d': diagonal line, representing P(y/SN)=P(y/N), that is, the subject's discriminatory power d' is 0, the further the ROC curve is from this line, indicating that the more discriminatory the subject is, the greater the value of d' is. The value of d' will of course be larger. From the above, it can be seen that the change in d' causes the ROC curve to form a cluster of curves, and the change in β reflects the change at different points on one of the curves in this cluster.

In addition, if the axes of the ROC curve are changed to Z-score coordinates, we will see that the ROC curve changes from a curvilinear form to a linear form. This coordinate transformation can be used to verify an important assumption of signal detection theory, namely the variance chi-square assumption.

(3) Complementary properties:

For a particular ROC curve, d' is constant, so it is also called an iso-receptive curve.

The diagonal line represents a line where discrimination is equal to 0, also called a line of pure opportunity.

The farther the ROC curve is from the line of pure opportunity, the stronger the subject's discrimination.

The ROC curve is different for subjects with different discriminatory power.

Baidu Encyclopedia - Receiver Operating Characteristic Curve