Abstract-Irregular streaks are important clues for Melanoma (a potentially fatal form of skin cancer) diagnosis using dermoscopy images. This paper extends our previous algorithm to identify the absence or presence of streaks in a skin lesions, by further analyzing the appearance of detected streak lines, and performing a three-way classi cation for streaks, Absent, Regular, and Irregular, in a pigmented skin lesion. In addition, the directional pattern of detected lines is analyzed to extract their orientation features in order to detect the underlying pattern. The method uses a graphical representation to model the geometric pattern of valid streaks and the distribution and coverage of the structure. Using these proposed features of the valid streaks along with the color and texture features of the entire lesion, an accuracy of 76.1% and weighted average area under ROC curve (AUC) of 85% is achieved for classifying dermoscopy images into streaks Absent, Regular, or Irregular on 945 images compiled from atlases and the internet without any exclusion criteria. This challenging dataset is the largest validation dataset for streaks detection and classi cation published to date. The data set has also been applied to the two-class sub-problems of Absent/Present classi cation (accuracy of 78.3% with AUC of 83.2%) and to Regular/Irregular classi cation (accuracy 83.6% with AUC of 88.9%). When the method was tested on a cleaned subset of 300 images randomly selected from the 945 images, the AUC increased to 91.8%, 93.2% and 90.9% for the Absent/Regular/Irregular, Absent/Present, and Regular/Irregular problems, respectively.

a b s t r a c tWe describe a novel approach to detect and visualize pigment network structures in dermoscopic images, based on the fact that the edges of pigment network structures form cyclic graphs which can be automatically detected and analyzed. First we perform a pre-processing step of image enhancement and edge detection. The resulting binary edge image is converted to a graph and the defined feature patterns are extracted by finding cyclic subgraphs corresponding to skin texture structures. We filtered these cyclic subgraphs to remove other round structures such as globules, dots, and oil bubbles, based on their size and color. Another high-level graph is created from each correctly extracted subgraph, with a node corresponding to a hole in the pigment network. Nodes are connected by edges according to their distances. Finally the image is classified according to the density ratio of the graph. Our results over a set of 500 images from a well known atlas of dermoscopy show an accuracy of 94.3% on classification of the images as pigment network Present or Absent.Crown

Abstract. Skin lesions are often comprised of various colours. The presence of multiple colours with an irregular distribution can signal malignancy. Among common colours under dermoscopy, blue-grey (blue-white veil) is a strong indicator of malignant melanoma. Since it is not always easy to visually identify and recognize this feature, a computerised automatic colour analysis method can provide the clinician with an objective second opinion. In this paper, we put forward an innovative method, through colour analysis and computer vision techniques, to automatically detect and segment blue-white veil areas in dermoscopy images. The proposed method is an attempt to mimic the human perception of lesion colours, and improves and outperforms the state-of-the-art as shown in our experiments.

Abstract. We present a method for automatically segmenting skin lesions by initializing the random walker algorithm with seed points whose properties, such as colour and texture, have been learnt via a training set. We leverage the speed and robustness of the random walker algorithm and augment it into a fully automatic method by using supervised statistical pattern recognition techniques. We validate our results by comparing the resulting segmentations to the manual segmentations of an expert over 120 cases, including 100 cases which are categorized as difficult (i.e.: low contrast, heavily occluded, etc.). We achieve an Fmeasure of 0.95 when segmenting easy cases, and an F-measure of 0.85 when segmenting difficult cases.