Skin cancer, the most common human malignancy, is primarily diagnosed visually, beginning with an initial clinical screening and followed potentially by dermoscopic analysis, a biopsy and histopathological examination. Automated classification of skin lesions using images is a challenging task owing to the fine-grained variability in the appearance of skin lesions. Deep convolutional neural networks (CNNs) show potential for general and highly variable tasks across many fine-grained object categories. Here we demonstrate classification of skin lesions using a single CNN, trained end-to-end from images directly, using only pixels and disease labels as inputs. We train a CNN using a dataset of 129,450 clinical images-two orders of magnitude larger than previous datasets-consisting of 2,032 different diseases. We test its performance against 21 board-certified dermatologists on biopsy-proven clinical images with two critical binary classification use cases: keratinocyte carcinomas versus benign seborrheic keratoses; and malignant melanomas versus benign nevi. The first case represents the identification of the most common cancers, the second represents the identification of the deadliest skin cancer. The CNN achieves performance on par with all tested experts across both tasks, demonstrating an artificial intelligence capable of classifying skin cancer with a level of competence comparable to dermatologists. Outfitted with deep neural networks, mobile devices can potentially extend the reach of dermatologists outside of the clinic. It is projected that 6.3 billion smartphone subscriptions will exist by the year 2021 (ref. 13) and can therefore potentially provide low-cost universal access to vital diagnostic care.

We identified circulating CD8+ T-cell populations specific for the tumor-associated antigens (TAAs) MART-1 (27-35) or tyrosinase (368-376) in six of eleven patients with metastatic melanoma using peptide/HLA-A*0201 tetramers. These TAA-specific populations were of two phenotypically distinct types: one, typical for memory/effector T cells; the other, a previously undescribed phenotype expressing both naive and effector cell markers. This latter type represented more than 2% of the total CD8+ T cells in one patient, permitting detailed phenotypic and functional analysis. Although these cells have many of the hallmarks of effector T cells, they were functionally unresponsive, unable to directly lyse melanoma target cells or produce cytokines in response to mitogens. In contrast, CD8+ T cells from the same patient were able to lyse EBV-pulsed target cells and showed robust allogeneic responses. Thus, the clonally expanded TAA-specific population seems to have been selectively rendered anergic in vivo. Peptide stimulation of the TAA-specific T-cell populations in other patients failed to induce substantial upregulation of CD69 expression, indicating that these cells may also have functional defects, leading to blunted activation responses. These data demonstrate that systemic TAA-specific T-cell responses can develop de novo in cancer patients, but that antigen-specific unresponsiveness may explain why such cells are unable to control tumor growth.

The question whether tumorigenic cancer stem cells exist in human melanomas has arisen recently1. Here we show that in melanomas, tumor stem cells (MTSC) can be isolated prospectively as a highly enriched CD271+ MTSC population using a process that maximizes viable cell transplantation1,6. In this study the tumors sampled were taken from a broad spectrum of sites and stages. High viability FACS isolated cells resuspended in a matrigel vehicle were implanted into T, B, and NK deficient Rag2−/− γc−/− mice (RG) mice. The CD271+ subset of cells was the tumor initiating population in 9/10 melanomas tested. Transplantation of isolated melanoma cells into engrafted human skin or bone in RG mice resulted in melanoma from CD271+ but not CD271− cells. We also showed that tumors transplanted by CD271+ patient cells were capable of metastasis in-vivo. Importantly, CD271+ melanoma cells lacked expression of TYR, MART and MAGE in 86%, 69% and 68% of melanoma patients respectively suggesting why T cell therapies directed at these antigens usually result in only temporary tumor shrinkage.