Automatic Identification of Diagnostic Significant Regions in Confocal Laser Scanning Microscopy of Melanocytic Skin Tumors

Abstract

Summary Objectives: Confocal laser scanning microscopy (CLSM) is used for quick medical checkups. The aim of this study is to check the discrimination power of texture features for the automatic identification of diagnostic significant regions in CLSM views of skin lesions. Methods: In tissue counter analysis (TCA) the images are dissected in equal square elements, where different classes of features are calculated out. Features defined in the spatial domain are based on histogram (grey level distribution) and co-occurrence matrix (grey level combinations). The features defined in the frequency domain 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). Hundred cases of benign common nevi and malignant melanoma were used as the study set. Classification was done with CART (Classification and Regression Trees) analysis which splits the set of square elements into homogenous terminal nodes and generates a set of splitting rules. Results: Features based on the wavelet transform provide the best results with 96.0% of correctly classified elements from benign common nevi and 97.0% from malignant melanoma. The classification results are relocated to the images by use of the splitting rules as diagnostic aid. The discriminated square elements are highlighted in the images, showing tissue with features in good accordance with typical diagnostic CLSM features. Conclusion: Square elements with more than 80% of discrimination power enable the identification of diagnostic highly significant parts in confocal microscopic views of malignant melanoma.