In this blog we present more details about the iToBoS dataset: skin region images extracted from 3D total body photographs for lesion detection.
This dataset consists of high-resolution images of skin patches captured from multiple scans of patients at two clinical sites: the Clinical Hospital of Barcelona (Spain) and the University of Queensland (Australia). The images were acquired using the Canfield VECTRA WB360 system. This system, equipped with 92 fixed cameras arranged in 46 stereo pairs, uses xenon flashes to capture comprehensive images of each patient’s entire exposed skin surface in a single session.
The captured 2D images are processed by the VECTRA software to create a precise 3D avatar. To ensure patient privacy and comply with GDPR guidelines, the patient's head is automatically removed from the 3D avatars, ensuring patient anonymity and excluding facial features. The anonymized 3D body avatars are then divided into smaller, overlapping tiles. These tiles have an average dimension of 1012x827 pixels with a 45-pixel overlap between adjacent tiles, although tiles at the edges and bottom of the avatar have smaller dimensions. This process resulted in a dataset of 17,000 tiles. The images encompass diverse anatomical locations, including the torso, arms, and legs, and metadata such as the patient's age, anatomical region, and sun damage score are also provided.
Below are a few examples from the training set.
Find out more about the iToBoS 2024 - Skin Lesion Detection with 3D-TBP.