BCN20000: Dermoscopic Lesions in the Wild

Abstract

Advancements in dermatological artificial intelligence research require high-quality and comprehensive datasets that mirror real-world clinical scenarios. We introduce a collection of 18,946 dermoscopic images spanning from 2010 to 2016, collated at the Hospital Clínic in Barcelona, Spain. The BCN20000 dataset aims to address the problem of unconstrained classification of dermoscopic images of skin cancer, including lesions in hard-to-diagnose locations such as those found in nails and mucosa, large lesions which do not fit in the aperture of the dermoscopy device, and hypo-pigmented lesions. Our dataset covers eight key diagnostic categories in dermoscopy, providing a diverse range of lesions for artificial intelligence model training. Furthermore, a ninth out-of-distribution (OOD) class is also present on the test set, comprised of lesions which could not be distinctively classified as any of the others. By providing a comprehensive collection of varied images, BCN20000 helps bridge the gap between the training data for machine learning models and the day-to-day practice of medical practitioners. Additionally, we present a set of baseline classifiers based on state-of-the-art neural networks, which can be extended by other researchers for further experimentation.

Fig. 1

Dermoscopic images showcasing diagnostic challenges in BCN20000: (a) Lesion on a nail, (d) Lesion on mucosal tissue. (b) and (e) Lesions too extensive for the dermoscopy device aperture, illustrating size-related diagnostic obstacles. (c) and (f) present hypopigmented lesions.

Fig. 2

Samples from the BCN20000 dataset corresponding to (a) nevus, (b) melanoma, (c) basal cell carcinoma, (d) solar lentigo/ seborrheic keratosis, (e) actinic keratosis, (f) squamous cell carcinoma, (g) dermatofibroma and (h) vascular lesion.

Fig. 3

BCN20000 Dataset Preparation Pipeline. The process begins with the collection of images and metadata. A neural network is then employed to classify and separate between the image types. Patient identifiers are extracted from ‘Sticker pictures’ using a YOLOv3 network. Dermatoscopy’s diagnosis are revised by multiple reviewers for quality assurance. The resultant BCN20000 dataset is composed exclusively of dermoscopic images and metadata, divided into training and testing sets.

Fig. 4

Example of the sticker detection algorithm. Left: original image (blurred for privacy), right: detection of the YOLOv3 Architecture.

Fig. 5

Comparison between original (a,b,c) and processed dermoscopic images (d,e,f) using the proposed image cropping algorithm.