AI prediction of PET positivity in CT scans (PET-KM-CTs)

Patients with CT imaging sometimes remain in partial remission for years after the end of treatment without suffering a relapse of their disease or the residual scar tissue showing any metabolic activity or dynamics. This results in a high degree of uncertainty for follow-up care and for assessing the response. In the clinical picture of diffuse large B-cell lymphoma (DLBCL), machine learning will use a retrospective training dataset of synchronous or near-coincident CT scans and FDG-PETs to examine CT imaging for AI-supported pattern clues in the relevant lesion(s), which could be predictive of metabolic FDG behavior. This would allow for new or expanded analysis of routine or older cohort CT data in the pre-PET-CT era or for centers with unavailability of this technique. Furthermore, it is not unlikely that AI-assisted CT reanalysis will lead to information that contributes new insights independent of PET in the context of clinical or molecular patient data.