I am having TP, TN, FP and FN for detected and classified ST segments for detection of ischemia in ECG signals
Dear Luca..
Thanks for advise. I have threshold value, how i decide score and label for plot in MATLAB. could you please suggest me any related command.
Thank you
A ROC curve is a useful way to study and illustrate the way a parameter affects a classifier. Say you want to study the effect of parameter A. Choose a value for A and perform the classification. Count correct decisions and incorrect decisions and obtain sensitivity and specificity. This gives you ONE point for the ROC curve. Now change A and repeat the procedure till you have enough points to plot the ROC curve. If your ROC curve has a sharp 'knee' towards the left-hand top corner it means your paramenter is good. The closer the ROC curve is to the main diagonal the worse the parameter is. One way of quantifying this is measuring the area under the ROC curve (AUROC): Area close to 0.5 meaning a poor parameter and area close to 1 meaning a good parameter.
Dear Fernando,
Thank you very much for nice lines and explanation about ROC. I have threshold values, but it required score and label also, how i decide this. Thank you
Hi, Amit Manocha,
I don't understand what you mean by "I have threshold values, but it required score and label also, how i decide this." To draw the ROC curve you need to know how many decisions using a certain value for your parameter were correct, how many false negatives you had (failed to detect something that occurred) and how many false positives you had (your algorithm decided it was an event, but it was not). That means you need to have an annotated database: A database for which you KNOW what (and when) your events to be detected are. This would have been created by a panel of experts. One such database is the MIT-BIH database, which is widely used to assess the development of automatic QRS detectors and ECG classification algorithms.
Hi Frenando Soares,
Thank you very much for the reply. I mean to say, I have threshold value by which TP, TN, FP, FN, sensitivity and specificity have been calculated for detected ischemic beats for ESC ST-T database. In MATLAB, for a plot of the ROC curve, I required score and labels. I have own developed classifier, for which how I can assign labels and scores. So, could you please suggest me any command in MATLAB or in SPSS for plotting ROC curve.
Thank you!
Amit
Save each point's coordinates (sensitivity and 1-specificity) and then plot them! What is the problem?
Sir,
I had tried, a simple plot (Se, 1-Sp), but it is not giving desired results for follwoing data
Se Sp
99.01 98.48
98.98 99.63
98.76 98.70
98.64 99.20
98.41 98.75
98.27 99.13
98.71 99.57
98.55 98.65
98.65 97.83
98.72 99.56
96.61 96.77
98.57 98.52
97.05 98.32
97.39 96.85
98.65 98.09
97.88 98.78
97.99 98.05
97.82 96.82
98.17 98.56
98.45 98.69
98.19 98.79
97.01 98.63
98.50 97.22
please suggest me.
Thank you
Your Sensitivity and Specificity are not behaving as they should. What should happen is that if you change your parameter one way (say decrease it) then Sensitivity will increase and Specificity will decrease (or vice versa), and if you change your parameter the other way, then the opposite should happen. Your results don't show that. Also it seems you are not changing the parameter enough, as your Se and Sp are always close to 100%. Finally, if you express them in percentages, then you should plot (Se, 100-Sp), of course, but that does not help with the data you sent... Sorry I cannot be more helpful without actually doing it... Best of luck.
I think you should try : http://in.mathworks.com/matlabcentral/fileexchange/19950-rocout=roc-varargin-
Hope this helps !
- Badges
- Science topic
- Similar topics
- Engineering
- Control Theory
- Control Systems
- Control Systems Engineering