AI-assisted quantitative photon-counting-detector CT imaging for cytogenetic risk prediction and treatment response in multiple myeloma - PROJECT SUMMARY/ABSTRACT Focal bone marrow biopsies of the posterior iliac crest are the gold standard for diagnosing multiple myeloma, yet they do not reflect the spatial and genomic heterogeneity of the disease. Almost 25% of such patients do not have adequate samples for cytogenetic assessment need for risk stratification. This inconsistency in bone marrow sampling could result in falsely reassuring prognoses for patients with aggressive disease. Whole-body imaging presents an unrealized opportunity to derive image features in the bone marrow that reflect the cytogenetic diversity of clonal plasma cells. Several advanced imaging modalities are used for diagnosis and disease management. An alternative approach is to optimize one imaging modality for these tasks. Recently FDA approved photon- counting detector (PCD)-CT has technological advancements that permit simultaneous bone marrow composition quantification and lytic lesion detection throughout the whole skeleton. In addition, the quantitative accuracy of the spectral data can be used to determine changes in bone marrow and intralesional fat fraction, an indicator of treatment response. This proposal’s specific objective is to develop imaging based cytogenetic risk stratification models and a quantitative PCD-CT mechanism to accurately measure bone marrow composition. In Aim 1, we will develop and validate a deep- learning (DL) imaging-model for multiple myeloma cytogenetic risk stratification. In Aim 2, we will develop and validate physics-informed DL-assisted material decomposition to derive imaging biomarkers of multiple myeloma. In Aim 3, we will determine the clinical feasibility of the DL cytogenetic risk prediction model and imaging biomarkers for multiple myeloma prognostication and assessing response to therapy. This PCD-CT protocol will be a “one-stop shop” multiple myeloma tool for disease detection, prognostication, and treatment response assessment. The innovation of this proposal includes our development of an imaging protocol that leverages the many technical improvements of PCD-CT, AI image processing, and AI modeling to develop biomarkers from bone marrow composition. Secondly, a PCD-CT-based AI risk prediction model will be positioned to supplant focal invasive approaches that may not offer a comprehensive assessment of the entire heterogeneity of disease burden.
