The use of co-occurrence features in medical imaging: An empirical study

Co-occurrence matrices are one of the most common second order statistical features used in the textural analysis of medical images. Here we review the application of co-occurrence matrices in medical imaging in general, paying particular attention to thoracic Computed Tomography (CT) scans. We note the wide variety of methods and user-defined parameter values that have been employed and test some of the common assumptions, presenting results on the classification accuracy in differentiating between seven common patterns found on CT scans of the thorax: normal lung, fibrosis, emphysema, ground glass, pleural effusion, muscle and fat.