Melanoma is the deadliest form of skin cancer. While curable with earlydetection, only highly trained specialists are capable of accuratelyrecognizing the disease. As expertise is in limited supply, automated systemscapable of identifying disease could save lives, reduce unnecessary biopsies,and reduce costs. Toward this goal, we propose a system that combines recentdevelopments in deep learning with established machine learning approaches,creating ensembles of methods that are capable of segmenting skin lesions, aswell as analyzing the detected area and surrounding tissue for melanomadetection. The system is evaluated using the largest publicly availablebenchmark dataset of dermoscopic images, containing 900 training and 379testing images. New state-of-the-art performance levels are demonstrated,leading to an improvement in the area under receiver operating characteristiccurve of 7.5% (0.843 vs. 0.783), in average precision of 4% (0.649 vs. 0.624),and in specificity measured at the clinically relevant 95% sensitivityoperating point 2.9 times higher than the previous state-of-the-art (36.8%specificity compared to 12.5%). Compared to the average of 8 expertdermatologists on a subset of 100 test images, the proposed system produces ahigher accuracy (76% vs. 70.5%), and specificity (62% vs. 59%) evaluated at anequivalent sensitivity (82%).
展开▼
