Comprehensive Summary
Chong et. al. trained an artificial intelligence (AI) model with thousands of sample slides from various dermatopathology institutions, the largest dataset of its kind to date. Over 34,376 slides were gathered, including images of normal skin, epidermal cysts, seborrheic keratosis, Bowen's disease (SCC in situ), basal cell carcinoma, melanocytic nevus, and malignant melanoma. Each of the samples were annotated, denoting the specific lesion area as well as pertinent clinical information. The dataset achieved over 99% syntactic accuracy, and is considered statistically diverse. Chong et. al created a large and diverse dataset to be used to expedite dermatological diagnosis with AI.
Outcomes and Implications
The dataset gathered through this study can serve as substantial machine learning training material to advance the AI-based dermatopathology scene. At present, rates of skin cancer are increasing, and as a result, the demand for pathological analyses of such lesions are on an upward trend. Being the largest dataset of its kind, spanning multiple types of skin lesions with thousands of image samples, the findings in this study will promote new discoveries in deep learning models meant to assist in and expedite diagnosing dermatological issues.