ml-medimage

The ML-MedImage framework provides an environment to evaluating multi-label learners to the automatic annotation task of two-dimensional medical images. The label are assigned conform to the IRMA code. In this framework, ten subsets are built from a set with more than 12.000 ray-X medical images from chest region. The EHD, Gabor, LBP and SIFT techniques are used to feature the samples from formed subsets. From theses subsets, the performances of various multi-label learners are evaluated on image annotation task. Ten iterations are performed to this evaluating. For each iteration, a subset is used to train step and nine remaining subsets to test step. The learners used are BRkNN, ClassifierChain(RandomForest), LabelPowerset(kNN) and MLkNN. Beyond from this approach, an alternative approach is evaluating too. In this other approach, the classification is performed to axis from IRMA code instead of to assign the labels to all axes in one step like to first approach. The evaluating provides results to various measures for each iteration. These results are grouped by measure in individual files to that can get means and standard deviation from each iteration. The measures considered in experimental evaluating performed are Average Precision, Hamming Loss and Micro F.