Automatic discrimination of melanocytic skin tumors

	Back to Success Stories Mainpage
	Back to What's New

Automatic discrimination of melanocytic skin tumors

M. Wiltgen¹, A. Gerger² and J. Smolle²
¹ Institute of Medical Informatics, Statistics and Documentation,
² Department of Dermatology, Division of Analytical-Morphological Dermatology,
Medical University of Graz, Austria

There is an increasing number of melanocytic tumors. The frequency of melanoma doubles every 20 years. At present there is a risk of 1:100 to fall sick with a melanoma. Reasons for the increasing number of melanocytic tumors are for example extreme sun exposure during sun-bathing. In the fight against skin cancer, researchers have high hopes in improved provisional screening methods, such as optimised computer aided diagnostic methods. Not every change of dermal tissue is dangerous. There are harmless cases too. Automatic analysis means, that the harmless (nevi) and malignant cases are discriminated by computer. This will optimise preventive medical checkups and early recognition of skin tumors. The detection of malignant changes of skin tissue in the early beginning will arise the success of the therapy. Most examinations are based on microscopic views of the tissue. To this purpose a biopsy of the tissue is prepared and stained using a fully automated device.

Automated image analysis of histological tissue is limited by the difficulty of recognizing special structures by computer. In contrast to isolated structures like blood cells, which can generally be well defined, the cells in histological tissues are arranged in various patterns showing variability in shape and appearance. The segmentation of different structures in histological tissues is therefore case dependent and cannot be done in a general approach. To avoid these problems, in tissue counter analysis (TCA) the images were dissected in square elements and the features were calculated for each element. In this way a priori definition and segmentation of the structures, which is the crucial point in automatic classification, was avoided.

graz1.jpg (47518 Byte)
Figure. 1. The tissue counter analysis consists of 3 parts: the feature analysis and extraction, the classification and the relocation. In feature analysis and extraction the images are dissected in square elements and the features, based on grey level histogram and co-occurrence matrix, are calculated inside each element. The classification is done by CART analysis, where the set of square elements are split into disjunctive nodes, representing different kinds of tissue. The relocation, that means the indication of the classified square elements superimposed to the image offers the possibility to evaluate the performance of the procedure.

The TCA consists of 3 steps, which are performed by different parts of the image analysis system (Fig. 1): The feature extraction, the classification and the relocation. In feature extraction the images are dissected in square elements and the features, describing the tissue, are calculated inside each element. The classification is done by CART (Classification and Regression Trees) analysis, where the set of square elements are split into disjunctive nodes, representing different kinds of tissue. The relocation, that means the indication of the classified square elements superimposed to the image offers the possibility to evaluate the performance of the procedure. The software for image analysis was developed with IDL.



Figure 2. The figure shows the histogram and co-occurrence matrix (as a plot) of the blue channel from a square element in an image of benign common nevi. The histogram has a skewness with a negative value, a low variance and the mean value lies in the range of high grey levels. The distribution of the elements in the co-occurrence matrix is concentrated in the range of high values with a low variance.

The aim of this study was to evaluate the possibilities of describing and discriminating common nevi and malignant melanoma tissue by the use of features extracted from histogram and co-occurrence matrix. 80 cases from microscopic views of benign common nevi and malignant melanoma were sampled. From this set 40 cases were randomly selected as learning set and the remaining 40 cases were used as test set. Each image was dissected in 256 square elements and 51 different features, describing histogram and co-occurrence matrix, were used. The tissue of benign common nevi appears faded rose red and homogeneous, the nuclei are small and spread out widely. The tissue of malignant melanoma appears dark with high contrast. The properties, enabling the discrimination between the different tissues, are described by the relevant features. The histogram skewness of the images of benign common nevi shows negative values, the variance is low and the mean value lies in the range of high grey levels (Fig. 2). The distribution of the elements in the co-occurrence matrix is concentrated in the range of high values and the variance of the element distribution is low. The histogram skewness of the images of malignant melanoma shows positive values, the variance is higher than in the case of benign common nevi and the mean value lies in the range of low grey levels (Fig. 3). The distribution of the elements in the co-occurrence matrix is concentrated in the range of lower grey values, the variance of the element distribution is high and the correlation low.



Figure 3. The figure shows the histogram and co-occurrence matrix (as a plot) of the green channel from a square element in an image of malignant melanoma. The histogram skewness has a positive value, a higher variance as in the case of benign common nevi and the mean value lies in the range of low grey levels with. The distribution of the elements in the co-occurrence matrix is concentrated in the range of lower values with a high variance.

The results from classification show a clear-cut difference between common nevi and malignant melanoma. The classification correctly classified 94,7 % of nevi elements and 92,6% of melanoma elements in the learning set. Discriminant analysis based on the percentage of "malignant elements" facilitated a correct classification of all cases in the test set (sensitivity = 100 %, specificity = 100 %). When the percentage of elements suggestive for malignancy in each case was evaluated, it turned out that a threshold level of 42% provides a correct classification of nevi and melanoma cases. The classification results were indicated in the original image in order to evaluate the performance of the procedure.

In conclusion, tissue counter analysis is a potential diagnostic tool in automatic or semi automatic analysis of melanocytic skin tumors.