gms | German Medical Science

Automated image analysis of histological tissue is limited by the difficulty of recognizing special structures by computer. Microscopic views of histological tissues show structures mostly arranged in a variety of patterns. Therefore the automatic segmentation of different structures, like cells, nuclei, cytoplasm, vessels etc., is difficult, depending on the concrete tissue and cannot be done in a general approach [Ref. 1], [Ref. 2]. Tissue counter analysis (TCA) is based on the partition of the image into square elements of equal size where the features, describing the tissue, are calculated out [Ref. 3], [Ref. 4]. Therefore problems related to segmentation are avoided.

The TCA consists of 3 steps: The feature analysis and extraction, the classification and the relocation. We check the applicability of TCA in diagnostic discrimination of microscopic views of benign common nevi and malignant melanoma lesions 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. The classification was done by CART (Classification and Regression Trees) analysis. Each image was dissected in 256 square elements and 51 different features, describing histogram and co-occurrence matrix, were used. The square elements from the images of a learning set were classified. To evaluate the recognition rate the classification results were applied to individual cases of the test set. The classification results were indicated in the original image in order to evaluate the performance of the procedure (relocation).

Additionally a second class of features, defined in frequency domain, was tested for discrimination. These features are based on spectral properties of the wavelet Daubechie 4 transform (texture exploration at different scales) and the Fourier transform (global texture properties are localized in the spectrum).

The results from classification show a clear-cut difference between common nevi and malignant melanoma. With the features based on histogram and co-occurrence matrix the classification correctly classified 92,7% of nevi elements and 92,1% of melanoma elements in the learning set. In the test set, discriminant analysis based on the percentage of "malignant elements" showed a correct classification of all cases (sensitivity = 100 %, specificity = 100 %).

With the features derived from the wavelet spectrum the classification shows correct results for 88.778% of benign common nevi and 85.556% of malignant melanoma. The features from the Fourier spectrum provide 79.267% of correctly classified benign common nevi and 81.489% of malignant melanoma.

In conclusion, tissue counter analysis is a potential diagnostic tool in automatic or semi automatic analysis of melanocytic skin tumors. Grey level histogram and co-occurrence matrix features seem to be superior to wavelet and Fourier features in TCA of melanocytic skin tumors.