Targeting minimal residual disease in AML by using single-cell morphological and biophysical analysis with deep learning

Summary Conventional chemotherapy regimens induce complete remission in the majority of patients with acute myeloid leukemia (AML) and results in >95% reduction in tumor burden. However, the persistence of minimal or measurable residual disease (MRD) is considered to be a key determinant of relapse and poor patient prognosis in AML. The current clinical strategy at complete remission with MRD positivity is, depending on the level, to either provide additional dose(s) of the initial treatment or switch to a new therapy at the time of relapse. Our long-term goal is to instead target MRD by selecting a more optimal therapy before relapse as there are many advantages to treating patients with MRD-only disease rather than waiting for clinical relapse. This would be a transformative clinical advance for patients with persistent MRD if selective targeting of MRD either forestalled or completely obviated clinical relapse. We hypothesize that an optimal assay performed at the time of complete remission would not only predict therapeutic response but also identify relevant heterogeneity that is present within the patient’s MRD. We are proposing to achieve this assay by integrating Deepcell’s high-content label-free brightfield imaging and AI powered sorting with MIT’s high precision biophysical measurements and DFCI’s clinical resources and expertise. We expect to deliver a novel platform to enable real-time functional assessment of AML MRD therapeutic vulnerabilities with meaningful clinical impact. Furthermore, technical advances and experience gained here will facilitate future application of a similar device to the MRD state in other cancers.