Please cite this article in press as: Zortea M, et al. Performance of a dermoscopy-based computer vision system for the diagnosis of pigmented skin lesions compared with visual evaluation by experienced dermatologists. Artif Intell Med (2013), http://dx.doi.org/10.1016/j.artmed.2013.11.006 ARTICLE IN PRESS G Model ARTMED-1317; No. of Pages 14 Artificial Intelligence in Medicine xxx (2013) xxx–xxx Contents lists available at ScienceDirect Artificial Intelligence in Medicine jou rn al hom e page: www.elsevier.com/locate/aiim Full length Article Performance of a dermoscopy-based computer vision system for the diagnosis of pigmented skin lesions compared with visual evaluation by experienced dermatologists Maciel Zortea a,∗ , Thomas R. Schopf b , Kevin Thon b , Marc Geilhufe a , Kristian Hindberg a , Herbert Kirchesch c , Kajsa Møllersen b , Jörn Schulz a , Stein Olav Skrøvseth b , Fred Godtliebsen a a Department of Mathematics and Statistics, University of Tromsø, 9037 Tromsø, Norway b Norwegian Centre for Integrated Care and Telemedicine, University Hospital of North Norway, 9038 Tromsø, Norway c Dermatology Office, Venloer Straße 107, 50259 Pulheim, Germany a r t i c l e i n f o Article history: Received 1 July 2013 Received in revised form 28 November 2013 Accepted 29 November 2013 Keywords: Computer-aided diagnosis Supervised classification Skin cancer detection Melanoma Pigmented skin lesions Dermoscopy a b s t r a c t Background: It is often difficult to differentiate early melanomas from benign melanocytic nevi even by expert dermatologists, and the task is even more challenging for primary care physicians untrained in dermatology and dermoscopy. A computer system can provide an objective and quantitative evaluation of skin lesions, reducing subjectivity in the diagnosis. Objective: Our objective is to make a low-cost computer aided diagnostic tool applicable in primary care based on a consumer grade camera with attached dermatoscope, and compare its performance to that of experienced dermatologists. Methods and materials: We propose several new image-derived features computed from automatically segmented dermoscopic pictures. These are related to the asymmetry, color, border, geometry, and tex- ture of skin lesions. The diagnostic accuracy of the system is compared with that of three dermatologists. Results: With a data set of 206 skin lesions, 169 benign and 37 melanomas, the classifier was able to provide competitive sensitivity (86%) and specificity (52%) scores compared with the sensitivity (85%) and specificity (48%) of the most accurate dermatologist using only dermoscopic images. Conclusion: We show that simple statistical classifiers can be trained to provide a recommendation on whether a pigmented skin lesion requires biopsy to exclude skin cancer with a performance that is comparable to and exceeds that of experienced dermatologists. © 2013 Elsevier B.V. All rights reserved. 1. Introduction The incidence and mortality rates of melanoma in Caucasian populations have been increasing for many decades [1]. Though only accounting for one tenth of the new cases of skin cancer, melanoma are associated with more than 90% of the skin can- cer deaths [2]. If detected at an early stage, the prognosis for the patient is excellent because the patient can be cured by simple exci- sion of the tumor. However, early diagnosis is very challenging as melanomas are easily confused with benign skin lesions. Dermoscopy is a method that allows doctors to examine struc- tures in the skin that are not visible to the naked eye. When practiced by experts, dermoscopy improves the diagnostic accu- racy of pigmented skin lesions (PSL) [3–5]. Several methods have ∗ Corresponding author. Tel.: +47 45683608. E-mail addresses: mzortea@gmail.com, maciel.zortea@uit.no (M. Zortea). been developed to help clinicians interpret the structures revealed through dermoscopy [6]. Well known algorithms include the ABCD rule of dermoscopy [7], the Menzies method [8], the Three-point checklist [9], the 7-point checklist [10], the CASH algorithm for dermoscopy [11], the Chaos and Clues algorithm [12], the BLINCK algorithm [13], and Pattern Analysis [14]. However, intensive and time consuming training is required to become an expert in der- moscopy. Furthermore, dermoscopy has its limitations, especially in the diagnosis of early melanoma [15]. In the early stages of the disease it may look like a common mole. Often there are no specific dermatoscopic features of melanoma, or the features appear subtle and are easily overlooked. There have been efforts to develop computer programs to diag- nose melanoma based on lesion images. Roughly, these studies follow intuitive steps in a standard pattern recognition processing chain: (a) image segmentation to separate the lesion area from the background skin, (b) extraction of image features for classifica- tion purposes, and (c) final classification using statistical methods. 0933-3657/$ – see front matter © 2013 Elsevier B.V. All rights reserved. http://dx.doi.org/10.1016/j.artmed.2013.11.006