Search

BACKGROUND Primary care providers, dermatology specialists, and health care access are key components of primary prevention, early diagnosis, and treatment of skin cancer. Artificial intelligence (AI) offers the promise of diagnostic support for nonspecialists, but real-world clinical validation of AI in primary care is lacking. OBJECTIVE We aimed to (1) assess the reliability of an AI-based clinical triage algorithm in classifying benign and malignant skin lesions and (2) evaluate the quality of images obtained in primary care using the study camera (3Gen DermLite Cam v4 or similar). METHODS This was a single-center, prospective, double-blinded observational study with a predetermined study design. We recruited participants with suspected skin cancer in 20 primary care practices who were referred for assessment via teledermatology. A second set of photographs taken using a standardized camera was processed by the AI algorithm. We evaluated the image quality and compared two teledermatologists’ diagnoses by consensus (the “gold standard”) with AI and histology where applicable. RESULTS Our primary outcome assessment stratified 391 skin lesions by management as benign, uncertain, or malignant. Uncertain lesions were not included in the sensitivity and specificity analyses. Uncertain lesions included lesions that had either diagnostic or management uncertainties. For the remaining 242 lesions, the sensitivity was 97.26% (95% CI 93.13%-99.25%) and the specificity was 97.92% (95% CI 92.68%-99.75%). The AI algorithm was compared with the histological diagnoses for 123 lesions. The sensitivity was 100% (95% CI 95.85%-100%) and the specificity was 72.22% (95% CI 54.81%-85.80%). CONCLUSIONS The AI algorithm demonstrates encouraging results, with high sensitivity and specificity, concordant with previous AI studies. It shows potential as a triage tool in conjunction with teledermatology to augment health care and improve access to dermatology. Further real-life studies need to be conducted on a larger scale to assess the reliability, usability, and cost-effectiveness of the algorithm in primary care.

Background Primary care providers, dermatology specialists, and health care access are key components of primary prevention, early diagnosis, and treatment of skin cancer. Artificial intelligence (AI) offers the promise of diagnostic support for nonspecialists, but real-world clinical validation of AI in primary care is lacking. Objective We aimed to (1) assess the reliability of an AI-based clinical triage algorithm in classifying benign and malignant skin lesions and (2) evaluate the quality of images obtained in primary care using the study camera (3Gen DermLite Cam v4 or similar). Methods This was a single-center, prospective, double-blinded observational study with a predetermined study design. We recruited participants with suspected skin cancer in 20 primary care practices who were referred for assessment via teledermatology. A second set of photographs taken using a standardized camera was processed by the AI algorithm. We evaluated the image quality and compared two teledermatologists’ diagnoses by consensus (the “gold standard”) with AI and histology where applicable. Results Our primary outcome assessment stratified 391 skin lesions by management as benign, uncertain, or malignant. Uncertain lesions were not included in the sensitivity and specificity analyses. Uncertain lesions included lesions that had either diagnostic or management uncertainties. For the remaining 242 lesions, the sensitivity was 97.26% (95% CI 93.13%-99.25%) and the specificity was 97.92% (95% CI 92.68%-99.75%). The AI algorithm was compared with the histological diagnoses for 123 lesions. The sensitivity was 100% (95% CI 95.85%-100%) and the specificity was 72.22% (95% CI 54.81%-85.80%). Conclusions The AI algorithm demonstrates encouraging results, with high sensitivity and specificity, concordant with previous AI studies. It shows potential as a triage tool in conjunction with teledermatology to augment health care and improve access to dermatology. Further real-life studies need to be conducted on a larger scale to assess the reliability, usability, and cost-effectiveness of the algorithm in primary care. Acknowledgments MoleMap NZ, who developed the AI algorithm, provided some funding for this study. HT's salary was partially sponsored by MoleMap NZ, who developed the AI algorithm. AB is a shareholder and consultant to Molemap Ltd provider of the AI algorithm. Conflicts of Interest None declared.

Background MoleMap NZ is a novel New Zealand–based store-and-forward telemedicine service to detect melanoma. It uses expert review of total body photography and close-up and dermoscopic images of skin lesions that are suspicious for malignancy. Objective The purpose of this study was to assess the effectiveness of MoleMap NZ as a melanoma early detection program. Methods We conducted a review of 2108 melanocytic lesions recommended for biopsy/excision by MoleMap NZ dermoscopists between January 2015 and December 2016. Results Pathologic diagnoses were available for 1571 lesions. Of these, 1303 (83%) lesions were benign and 260 (17%) lesions were diagnosed as melanoma, for a melanoma-specific benign:malignant ratio of 5.0:1. The number needed to biopsy to obtain 1 melanoma was 6. Among melanomas with available tumor thickness data (n = 137), 92% were Limitations Only lesions recommended for excision were analyzed. Pathology results were available for 75% of these cases. Tumor thickness data were available for 53% of melanomas diagnosed. Conclusions This real-world study of MoleMap NZ, a community-based teledermoscopy program, suggests that it has the potential to increase patients' access to specialist expertise via telemedicine. Additional studies are needed to more accurately define its efficacy.

Similarity in appearance between various skin diseases, often makes it challenging for clinicians to identify the type of skin condition, and the accuracy is highly reliant on the level of expertise. There is also a great degree of subjectivity and inter/intra observer variability found in the clinical practices. In this paper, we propose a method for automatic skin diseases recognition that combines two different types of deep convolutional neural network features. We hold the hypothesis that it is equally important to capture global features such as color and lesion shape, as well as local features such as local patterns within the lesion area. The proposed method leverages deep residual network to represent global information, and bilinear pooling technique which allows to extract local features to differentiate between skin conditions with subtle visual differences in local regions. We have evaluated our proposed method on MoleMap dataset with 32,195 and ISBI-2016 challenge dataset with 1,279 skin images. Without any lesion localisation or segmentation, our proposed method has achieved state-of-the-art results on the large-scale MoleMap datasets with 15 various disease categories and multiple imaging modalities, and compares favorably with the best method on ISBI-2016 Melanoma challenge dataset.