The advantages of generalized co-occurrence matrix in biomedical image analysis by self-organizing maps

Cancer diagnosis is one of the challenging works in the past decade and it has attracted much attention. Because that the laboratory methods and the medical equipment are very expensive and sometime are not available, development of computational methods in machine vision to predicting disease can save the lives of men in these regions. Image characteristics as features can include a particular or general information about the image such as color, shape or texture. The analysis of texture parameters is a useful way to obtain valuable information from medical images. There are several algorithms to extracting texture related features. In this study, we focused on statistical methods and computed generalized co-occurrence matrices for color medical images. The extracted features can be used to classify various biological tissues images by machine learning techniques. The proposed method can classify different types of malignancies and includes following steps: A generalized co-occurrence matrices is calculated from the tissues image. At the next step, the matrix is vectorized and considered as input to train a supervised self-organizing map (SOM). Results showed that generalized co-occurrence matrices method beside the classification ability of supervised SOM is highly capable to discriminate different types of Malignancies. To evaluate the performance of the built models, 10-fold cross validation was used. Experimental results achieved 92.7% on an imaging benchmark for biological applications called IICBU-2008 which was promising compared to the previous results.