FLIP-SIDE

Imaging studies such as CT scans are used in patient’s care to answer a specific question for clinicians to help in patient’s diagnosis and treatment. These images allow additional opportunistic measures of organs and body composition such as organ and muscle measures and this additional information provide deep insights for patients’ health as imaging biomarkers. Furthermore, these imaging biomarkers can potentially personalise complex treatments. Yet these are not routinely measured as they are not the primary clinical reason for undertaking the scans and are also laborious and time consuming. Artificial intelligence imaging models can therefore be utilised to automatically, opportunistically measure body compositions and draw additional insights without extra work for busy clinicians. Insights from these imaging biomarkers can then be used in disease identification, predicting health outcomes or in guiding therapy choices potentially transforming patient’s health care.

Undertaking training of such imaging AI models are traditionally forbidden due to data residing within health infrastructures. Federated learning approach allows models to be trained without the data leaving secure trust environments.

FLIP-SIDE project leverages on routinely collected images in partnering sites to train imaging AI models in a privacy conscious, secure, federated manner using FLIP platform. FLIP-SIDE will allow extraction of imaging biomarkers for a variety of clinical specialties to allow deeper understanding of patient’s disease, their outcomes and in guiding their therapies. Imaging models trained on diverse specialties across geographically distinct patient populations allow better performance through improved generalisability and validity. Our secure, privacy conscious, federated learning platform (FLIP) allow leveraging of our partnering sites fully by allowing model training across diverse populations.

Finally, imaging biomarkers extracted from FLIP-SIDE project will be evaluated in a variety of clinical scenarios from acute liver failure, critical illness, sepsis and oncology against health outcomes such as mortality, toxicity related to treatments. In these clinical scenarios, we will explore the association between imaging biomarkers and traditional biomarkers both in their ability to independently predict health outcomes and in their interaction together. Additionally, we will explore potential ability to personalise complex treatments using these imaging biomarkers.