Automated classification of common skin lesions using bioinspired features

Abstract

Abstract-Ultrasound (US) imaging of skin lesions provides information supplementary to dermoscopy and helps in improving diagnostic accuracy. The aim of the current work is to explore the feasibility of using ultrasound image features derived from radiological experience to distinguish between common skin lesions. 5-18 MHz B-mode ultrasound images were acquired of incoming patients. Images containing lesions 1-2 mm thick were selected (N=248), with histology used to diagnose suspicious lesions. 73 melanomas, 130 BCC, and 45 nevi were studied. Following semi-automatic segmentation, a number of relevant features expressing the geometry of the lesion boundary and boundary layer, as well as the image characteristics of the lesion, lesion boundary layer, and post-lesion region were considered. With the exception of lesion echogenicity, all features had an area under the curve (AUC) value of above 0.70. The AdaBoost and Support Vector Machine (SVM) classifiers were then trained and tested using cross-validation of 50 random equal populations of melanomas, BCC, and nevi; each population was then 2-fold (holdout) cross-validated 50 times. When detecting one group against two other groups, the detection of cancerous lesions fared best, with an AUC of at least 0.84 and a specificity of at least 19% at 100% sensitivity for both classifiers. The results demonstrate the potential of clinically useful ultrasound-based automatic differential diagnosis of skin lesions, which could perhaps be attained by better segmentation, having more training data, using several images of the same lesion when performing classification, as well as refinements in the definition of image features.

Automated Skin Lesion Classification on Ultrasound Images

“…Attenuation-based features, such as attenuation, contrast of attenuation and heterogeneity of attenuation, examine the lesion region and its shadowing in the dermis region right under the lesion mask. Contrast parameters, such as lesion- contrast-based heterogeneity and the mean lesion contrast, examine the contrast of each boundary line emanating radially from the inner edge of the lesion [ 48 ]. The above two subgroups, as well as the mean boundary belonging to the boundary subgroup, were adopted from Csabai et al [ 48 ].…”

“…Contrast parameters, such as lesion- contrast-based heterogeneity and the mean lesion contrast, examine the contrast of each boundary line emanating radially from the inner edge of the lesion [ 48 ]. The above two subgroups, as well as the mean boundary belonging to the boundary subgroup, were adopted from Csabai et al [ 48 ].…”

“…Shape features are derived from parameters describing the shape of the lesion boundary. Some shape parameters, such as standard deviation of curvature and circularity, were extracted by Csabai et al [ 48 ]. Further shape features, such as axis ratio, perimeter-area ratio and compactness (ratio of perimeter and the length of the major axis of the lesion mask), are also extracted.…”

“…A number of more recent studies have moved towards providing a skin cancer classification framework. Csabai et al [ 48 ] and Andrékuté et al [ 49 ] combined an semi-automated (SA) segmentation method with a fully-automatic feature extraction and classification method based on acoustical, textural and shape features. Csabai et al [ 48 ] examined three kinds of lesion types, namely MMs, BCCs and benign nevi.…”

“…Csabai et al [ 48 ] and Andrékuté et al [ 49 ] combined an semi-automated (SA) segmentation method with a fully-automatic feature extraction and classification method based on acoustical, textural and shape features. Csabai et al [ 48 ] examined three kinds of lesion types, namely MMs, BCCs and benign nevi. Using a manually selected seeding region, an active contour model (ACM) was used to segment the lesion.…”