Automatic regions of interest identification and classification in CT images: Application to kidney cysts

Recently, we proposed an original approach, in a statistical framework, for fully automatic detection of pathological kidneys in 2D CT images. In this paper, we propose some important improvements of our previous work and an attempt to classify the identified regions into pathological vs non pathological. To this end, we propose two indexing methods to construct the signatures coding the relevant information. The index is then used in a supervised classification technique to discriminate the kidney images. These approaches are tested on more than 500 clinically acquired images and promising results are obtained.