Differential value of iToBoS

iToBoS tackles the limitations of currently available systems, by designing a novel tool that will help practitioners during the diagnostic process. The diagnosis for a given exploration will be enhanced with all the data available for the patient, including, but not being limited to, genomics, clinical history, previous dermoscopy, etc. The result will be a cognitive assistant that integrates information from several sources to provide a personalized diagnostic for each patient.

This diagnosis system collating the data from different sources will be based on novel technologies arising from the field of explainable AI. In iToBoS we want to refrain from the common black box decisions taken by Deep Learning (DL) methods. Instead, the DL methods to be developed will provide a more transparent decision, in the form of human-understandable explanations of the results obtained. This will allow the dermatologist to take informed decisions based on the outputs of the DL system.

Moreover, in iToBoS we want to improve the process of systematically exploring the skin of a patient by developing a new scanner able to automate the Total Body Skin Examination (TBSE) process. Following the same philosophy, this new source of data will be collated to the other types of information of a given patient to enhance her/his diagnosis.

After the exploration, the system will be capable of automatically generating the 3D map of the moles of the patient by AI detection and tracking across images. With privacy in mind, the scanner will not collect data from the head/face of the patient, and the 3D reconstruction of the patient will be anonymized by creating a 3D avatar of the patient, allowing the precise location of moles across explorations, but decoupling the location of the mole from the body-structure of the patient.

The project envisions the construction of three scanners, which will be installed in three different hospitals to perform data collection first (M17-36), and a clinical feasibility study for the validation of the system (M38-47). This period will also serve to obtain prospective data with the new scanner, which will not only be used to develop the algorithms associated to this new type of data, but also to create an annotated dataset that will be released publicly in order to foster the development of cognitive assistant algorithms, such as the one envisioned in this project, within the scientific community. As an example, the dataset will be made publicly available as part of the test/training data of two challenges for skin lesion analysis to be organized within the project.

By providing a standardized way of scanning the skin of a patient with sufficient quality for diagnosis, and an AI-powered tool to assist diagnosis in a human-friendly form, we foresee that the solution presented in iToBoS will become the de-facto standard for skin examination.