Throughout the project, substantial work was carried out across hardware development, data acquisition, algorithm design, and system integration to realise the technical and clinical objectives of iToBoS.
Early efforts were directed towards the development of the high-resolution imaging module (HRIM), a critical component of the total body scanner. This included the design and validation of a custom liquid lens system, integrated with automated image stacking algorithms to enable non-contact dermoscopic imaging. The HRIM was successfully incorporated into a robotic scanning platform that enables whole-body imaging with precise lesion targeting. The full scanner system was completed, tested, and used in clinical trials.
In parallel, the project established a clinical data acquisition study, enrolling 496 patients and collecting a rich dataset comprising whole-body images, dermoscopic image stacks, clinical variables, and polygenic risk information. Image resolution ranged from 60–80 microns per pixel, allowing detailed visual analysis of pigmented skin lesions. A multi-layer annotation pipeline was implemented, combining algorithmic pre-annotations with expert dermatological validation.
Building on this dataset, a range of AI modules were developed to support skin lesion detection, classification, change tracking and risk estimation. These models were trained and validated using multimodal inputs, including image tiles, 3D surface reconstructions, and structured patient data. The resulting outputs were consolidated into a Cognitive Assistant, capable of producing risk scores at both lesion and patient levels. A graphical user interface was designed and refined through an iterative process involving direct feedback from clinical end-users.
Additional components developed during the project include a change detection algorithm for multi-temporal lesion analysis, and an explanation module that enhances the interpretability of AI outputs through saliency maps and structured justifications. These features were embedded into the Cognitive Assistant to support clinical usability and decision-making.
In terms of project management and coordination, all planned deliverables and milestones were completed. The full system, comprising hardware, software, and data infrastructure, was deployed and validated in clinical settings.
From an exploitation standpoint, the project produced a set of Key Exploitable Results (KERs), including hardware components, AI models, data management tools, and user interfaces. Three patents were filed related to the liquid lens and image acquisition technologies. Dissemination activities were conducted at scale, with over 170 publications, active participation in major scientific events, and the release of two public datasets. Engagement with clinicians, researchers, and industry stakeholders was maintained through workshops, digital media, and targeted communication strategies. The outcomes of this phase have established a strong foundation for further deployment and post-project clinical studies.
