College
College of Arts and Sciences
Mentor Information
Anas Kartoumah
Description
Dermoscopy refers to an imaging technique that magnifies skin structures not visible to the naked eye, aiding in lesion evaluation for diseases such as skin cancer. Deep learning, a type of artificial intelligence that uses neural networks to learn patterns from data, is used for its high performance in analyzing dermoscopic images, often achieving accuracy levels comparable to dermatologists. A systematic literature review was conducted using the PRISMA protocol where 388 total peer reviewed articles were extracted from PubMed and reviewed focusing on key words relevant to dermoscopic algorithms, deep learning, and dermatological imaging. After screening, we selected 33 articles based on an article’s integration of performance metrics and large population size to explore the diverse benefits of integrating deep learning with dermatological imaging approaches for melanoma diagnosis. We then analyzed the various model types, study designs, dataset types such as specificity, accuracy, and area under the curve (AUC) and reported outcomes. Across the 33 melanoma-focused studies, deep learning models proved high diagnostic performance, with reported accuracies in the high-80% to mid-90% range and sensitivity for melanoma detection often exceeding 90%. In some cases, fusion-based frameworks outperformed single-modality approaches by more than 6%, achieving accuracies above 90% and AUC values above 0.95. These findings consistently suggest benefits from deep learning enhanced dermoscopy used in diagnosing melanoma, supporting development and integration of artificial intelligence not only in dermatological specialties, but also by physicians in primary care facilities for its potential to increase accessibility in evaluating skin lesions accurately and quickly.
How Can Deep Learning Methods Enhance Dermoscopic Algorithms? Exploring the Synergy Between AI and Dermatological Imaging
Dermoscopy refers to an imaging technique that magnifies skin structures not visible to the naked eye, aiding in lesion evaluation for diseases such as skin cancer. Deep learning, a type of artificial intelligence that uses neural networks to learn patterns from data, is used for its high performance in analyzing dermoscopic images, often achieving accuracy levels comparable to dermatologists. A systematic literature review was conducted using the PRISMA protocol where 388 total peer reviewed articles were extracted from PubMed and reviewed focusing on key words relevant to dermoscopic algorithms, deep learning, and dermatological imaging. After screening, we selected 33 articles based on an article’s integration of performance metrics and large population size to explore the diverse benefits of integrating deep learning with dermatological imaging approaches for melanoma diagnosis. We then analyzed the various model types, study designs, dataset types such as specificity, accuracy, and area under the curve (AUC) and reported outcomes. Across the 33 melanoma-focused studies, deep learning models proved high diagnostic performance, with reported accuracies in the high-80% to mid-90% range and sensitivity for melanoma detection often exceeding 90%. In some cases, fusion-based frameworks outperformed single-modality approaches by more than 6%, achieving accuracies above 90% and AUC values above 0.95. These findings consistently suggest benefits from deep learning enhanced dermoscopy used in diagnosing melanoma, supporting development and integration of artificial intelligence not only in dermatological specialties, but also by physicians in primary care facilities for its potential to increase accessibility in evaluating skin lesions accurately and quickly.
