How can I calculate the accurcy?

I have 17 images of patients with cerebral saccular aneurysms (each image has only one aneurysm), after applying detection and segmentation algorithm to segment aneurysms from the images:

    13 aneurysms in 13 images were detected\segmented.

    8 aneurysms (4 true positive aneurysms + 4 false positive ones) in 4 images were detected\segmented.

My question is:

1) How to calculate the accuracy?

Accuracy=items classified correctly\all items classified*

Accuracy=13/17 =76.74% ??

Accuracy=17/(17+4) =80.95% ??

2) What are other statistical measures could be used to describe the results?

3) What are your suggestions to improve the results?

• I have not tested the algorithm using images of healthy patients.

How can I calculate the accuracy?

Answer 1

[Image Analyst]

See the ROC curve:

https://en.wikipedia.org/wiki/Receiver_operating_characteristic