Using one or several texture features from published or new proposed texture characterization methods; participants are invited to use any classifier to classify the different data of the Challenge.
In a first step, participants are invited to calibrate the classifiers to reach the best score with 116 subjects, half of which from osteoporotic patients and the other half from control subjects.
In a second step, using the previous calibration parameters and 58 new test images for blind classification, participants will have to identify OP patients and CT subjects.
Participants are invited to submit three results:
1) The classification results of 116 training and test images (half OP and half CT)
In this step, participants will be asked to provide the following classification rates:
-- TP (True Positive, which is the number of OP people correctly identified),
-- FP (False Positive, which is the number of CT people incorrectly identified),
-- TN (True Negative, which is the number of CT people correctly identified),
-- FN (False Negative, which is the number of OP people incorrectly identified),
-- Sn (True positive rate or sensitivity) as Sn = TP/(TP + FN),
-- Sp (Specificity or True Negative Rate) as Sp = TN/(FP + TN).
2) The classification results of 58 blind test images
In this step, challenge participants are asked to classify the blind test subjects as either osteoporotic patients or control subjects. The results should be submitted as a CSV file where:
- the first column contains the name of each image,
- the second column contains the classification result for each image (1 for osteoporotic, 0 for control), and
- the following columns with the texture feature values for each image.
An extended abstract (2 pages maximum) explaining the methodology used and the classifiers characteristics (used features for the classifier).