Computational vision systems for the detection of malignant melanoma

Abstract

Abstract. In recent years, computational vision-based diagnostic systems for dermatology have demonstrated significant progress. We review these systems by first presenting the installation, visual features utilized for skin lesion classification and the methods for defining them. We also describe how to extract these features through digital image processing methods, i.e. segmentation, registration, border detection, color and texture processing, and present how to use the extracted features for skin lesion classification by employing artificial intelligence methods, i.e. discriminant analysis, neural networks, and support vector machines. Finally, we compare these techniques in discriminating malignant melanoma tumors versus dysplastic naevi lesions.

Symmetry Extraction in High Sensitivity Melanoma Diagnosis

“…Reference Used Databases [13] Melanoma vs. Clark Nevus [16] Melanoma vs. Nevus [22] Melanoma vs. Dysplastic Nevus [25] Melanoma vs. Nevus This proposal Melanoma and Dysplastic Nevus vs. 5 kinds of Nevus…”

“…Mean and variance of the gradient are the features applied in order to discriminate between the two classes when evaluating the radius evolution. The second group is also composed of geometric features [22]. These features are obtained applying the formulas below:…”

“…Although this work belongs to the third group and it uses well known features, its novelty is based on increasing the importance of the symmetry when feature extraction step is applied [19][20][21][22], since asymmetry shows an uncontrolled growth of cells, leading to melanoma cancer. As an important example of how the second and third group can be used together, in [10] a system architecture was presented based on Support Vector Machine which deals with the ABCD rule.…”

Incorporating Colour Information for Computer-Aided Diagnosis of Melanoma from Dermoscopy Images: A Retrospective Survey and Critical Analysis

“…Most of these methods try to imitate the ABCD rule of dermoscopy where asymmetry is quantified with respect to a symmetry axis that bisects the lesion. In automatic image analysis approaches, the symmetry axis is determined in a variety of ways, such as principal axes (moment) of inertia of the lesion shape [52]. The asymmetry is then quantified by overlapping the two halves of the lesion along the symmetry axes and dividing the nonoverlapping area differences of the two halves by the total area of the lesion [52–56].…”

Search for Resolution Invariant Wavelet Features of Melanoma Learned by a Limited ANN Classifier

“…In the review [52] authors surveyed computer-based systems according to acquisition, feature definition, extraction and skin lesion classification. Authors concluded that some widely used lesion parameters like lesion size, shape, color, and texture do not correspond to known biological phenomena and the structural patterns that are considered essential for manual lesion categorization are absent in the analysis due to their complexity.…”