The main scientific objective of JoinMi is to identify and to standardize AI approaches, techniques, and tools to improve hemophilic patient management, with a particular focus on joint health.
The research program has thus been structured into various scientific work-packages to reflect its key areas of innovation, from standardization to collection of patients’ self-acquired images and implementation of digital diagnostic tools. Currently WPs 1, 2 and 3 have been funded and have ongoing activities, meanwhile WPs 4 and 5 have been planned but need to be finalised.
WP1 - Standardization of US detection for hemophilia joint damage
This work package will establish clear, evidence-based protocols to make musculoskeletal ultrasound (US) a reliable tool for assessing joint health in hemophilia. Through a systematic literature review and expert consensus, standardized definitions and scoring of joint bleeding, synovitis, and osteochondral damage will be developed. The outcome will reduce operator variability, enhance diagnostic accuracy, and enable consistent comparison of results across clinical and research settings.
WP1 Leader: Roberta Gualtierotti
WP2 - Self-acquisition of US images by hemophilic patients
This work extends the GAJA system from knee to elbow and ankle ultrasound imaging for hemophilic patients. It tackles the greater positioning and alignment challenges of these joints using human-computer interaction and augmented reality (AR) for real-time guidance.
New deep learning models will detect key anatomical landmarks to support accurate imaging. A user-centered design process involving patients and clinicians will guide development, focusing on usability, performance, and responsiveness through software and hardware optimization.
A final longitudinal study will assess GAJA’s accuracy, usability, and independence across multiple joints.
WP2 Leader: Sergio Mascetti
WP3 - Tools for Computer Aided Diagnosis
The CADET prototype uses deep learning to support musculoskeletal ultrasound diagnosis, currently for the knee, performing tasks like scan identification, image sequencing, and joint distention detection.
This work package focuses on extending CADET to the elbow and ankle using approaches such as new data annotation, model adaptation, domain adaptation, and automated labelling with expert review.
Future goals include direct blood effusion detection, exploring static vs. dynamic imaging, and integrating explainable AI. Longer-term, the system will combine multiple images, videos, and patient data to improve diagnostic accuracy and clinical relevance.
WP3 Leader: Sergio Mascetti
WP4 – Biomarkers
Discovery
COMING SOON
WP5 – Health Technology Impact Assessment
COMING SOON