The second Periodic Reporting, corresponding to the period M19-M36, was submitted to the EC on May 28th, 2024 (M38) and it was officially approved on August 20th, 2024.
Along this reporting period, 23 deliverables were submitted and two milestones have been achieved.
In total, six project meetings have been organized: three General Assembly meetings and three Project Management Board meetings. Additionally, several WP meetings and other project meetings have been organized by the Project Coordinator.
The bi-annual technical and financial reports of the consortium (M19-M24, M25-M30 and M31-M36) have been coordinated, monitored and reviewed by the Project Coordinator.
During the first 36 months, the project has made substantial progress in creating advanced tools for a comprehensive skin examination system. Initially, the overall architecture of the iToBoS system was defined, specifying system requirements and ensuring interoperability among the key modules. Then, the Data Management Plan (DMP) was established and the iToBoS cloud platform was completed, implementing the cloud infrastructure in accordance with the DMP.
A major advancement during the early months was the definition and validation of the high-resolution imaging module (HRIM), including the design of liquid lenses for image acquisition. Progress continued with the integration of this HRIM into a first prototype of the total body scanner. The development of this prototype was based on an arch concept carrying the HRIMs. Due to component shortages causing delays in the construction of the total body scanner, a contingency plan was implemented, which included using the Vectra 360 scanner for initial image acquisition. Following this strategy, tools for image ingestion, anonymization, and masking of both images and patient data were adapted for use with the data acquired by the Vectra scanner. As the project advanced, algorithms were developed for image processing, 3D reconstruction, lesion detection and classification, and data integration, as well as tools for applying machine learning to the data gathered using the Vectra. A diverse dataset with various types of lesions from 496 patients of different skin types and UV damage has been collected during the reported period, corresponding to the clinical data acquisition trial, providing a resolution 60-80 microns per pixel. Manual annotation processes have been implemented to feed the AI algorithms with the data coming from the clinical data acquisition trial. As new datasets became available, the development of a mole change detector made significant progress, leveraging machine learning frameworks for multi-temporal analysis to identify changes in lesions over time. This system produces a change detection score that aids in assessing disease progression. An explanation module describing and visualizing the AI system’s reasoning during diagnosis has also been developed, enhancing the transparency and interpretability of AI-driven decisions.
Concurrently, the prototype of the Bosch Total Body Scanner was developed, capable of scanning a patient with a resolution of 20 microns per pixel. During the second reporting period, we transitioned from the arch concept to a flexible design using collaborative robots that allowed for dynamic adjustment of camera positions, accommodating a wide array of patient profiles. This second approach also addressed component scarcity in the market while ensuring comprehensive and precise imaging of the pigmented skin lesions.