The goal of this competition is to develop state-of-the-art machine learning techniques for detecting multiple skin lesions in clinical images. These images resemble photographs taken with a standard camera or smartphone, often used for digital healthcare services. Your algorithm could be valuable in environments lacking specialized care and could serve as a critical initial step in facilitating early detection of abnormalities in suspicious lesions.
Diagnosing skin cancer traditionally relies heavily on the expertise of dermatologists and the use of dermatoscopy. This non-invasive approach uses a dermoscope to enhance the view of sub-macroscopic structures in pigmented skin lesions, which vary widely across dermatological conditions. While dermoscopy has improved diagnostic precision, its accuracy is closely tied to the clinician’s level of expertise. Further, the process of taking a dermoscopic image of every suspicious lesion is a very tedious process. This necessitates the need for computer-aided diagnosis (CAD) systems using conventional cameras, especially in environments where dermatological expertise is scarce. This would enable non-specialist practitioners, such as general physicians without dermatological training, to identify suspicious lesions with ease, thereby facilitating early detection of abnormalities and allowing for timely intervention and improved prognosis. Additionally, tracking changes in suspicious lesions over time becomes feasible, enabling researchers to study the progression of the disease and assess the effectiveness of various treatments.
Detecting lesions in different regions of the body is a critical initial step in gaining such valuable insights to make informed decisions regarding patient care and treatment. This competition, therefore, challenges you to develop state-of-the-art machine learning techniques for detecting multiple skin lesions in clinical images. These images are tiles extracted from anonymized 3D avatars generated by the Canfield VECTRA WB360 system that captures comprehensive images of each patient’s entire skin surface. As such, this competition leverages 3D total body photography (TBP) to present a novel dataset comprising multiple scans of hundreds of patients across two different continents.
Your work will contribute to advancing the timely diagnosis and treatment of skin cancer. Therefore, we urge all teams, regardless of their placement in the competition, to publish a manuscript on arXiv detailing their solution and to open-source their code.
Find all the information about the Competition, Calendar and Register at iToBoS 2024 - Skin Lesion Detection with 3D-TBP | Kaggle.
Download documents: