Other Open Access
Noel C. F. Codella; M. Emre Celebi; Kristin Dana; David Gutman; Brian Helba; Harald Kittler; Philipp Tschandl; Allan Halpern; Veronica Rotemberg; Josep Malvehy; Marc Combalia
This is the challenge design document for the "International Skin Imaging Collaboration (ISIC) Challenge: using dermoscopic image context to diagnose melanoma", accepted for MICCAI 2020.
Skin cancer is one of the most frequent types of cancer and manifests mainly in areas of the skin most exposed to the sun. Despite being the less frequent of the common skin cancers, melanoma is responsible for 75% of deaths from skin tumors. It can appear at any age, and it is the first most diagnosed cancer among patients from 25-29 years old, the second among 20-24 years-old and the third solid tumor among 15-19 years-old. Melanoma is one of the cancers with more years of productive life lost and is the most expensive cancer when expressed in terms of cost per death in Europe. Since skin cancer occurs on the surface of the skin, its lesions can be evaluated by visual inspection. Dermoscopy is an imaging device composed of a magnifying glass coupled with polarized light, which permits visualizing more profound levels of the skin as its surface reflection is removed. Prior research has found that this technique permitsimproved visualization of the lesion structures, enhancing the accuracy of dermatologists. Typically, experts search for specific structural and color cues that help them determine if a lesion is of a particular type of skin cancer, rule sets such as the "ABCD rule" have been used to standardize the clinical procedure associated with the diagnosis of skin cancer.
Many medical institutions are not only using but also capturing images of the skin lesions using specializeddermoscopic adapters coupled with high-resolution cameras. The increase of imaging data of this modality has motivated the appearance of computer vision algorithms that aim to diagnose the lesions on the dermoscopic images automatically. Earlier systems relied on the extraction of handcrafted features from the skin lesions, similar to the rule sets the clinicians were using to perform diagnosis. Researchers tried to develop highly specialized algorithms which would extract color, border features, symmetry, and a bunch of other types of diagnostic criteria that was later on appended on a machine learning classification algorithm. However, the increased availability of dermoscopic images has motivated the appearance of more sophisticated algorithms based on deep learning, mainly on convolutional neural networks. These algorithms are no longer based on rule sets since the feature extractor is already embedded in their architecture. A significant player in the adoption of these algorithms in the community has been the The International Skin Imaging Collaboration (ISIC), which has been organizing yearly challenges since 2016, where participants are asked to develop computer vision algorithms to help with the classification of dermoscopic images skin lesions.