Speaker
Description
Background: A fundamental challenge in clinical hematology is the inability to monitor bone marrow dynamics in real time without invasive procedures.
Objective: We present an Extended Kalman Filter (EKF) framework that reconstructs unobservable bone marrow cell populations (stem, progenitor, differentiated) from noisy measurements of peripheral blood inflammation.
Methods: The hematopoietic system is modeled as a five state nonlinear dynamical system with inflammatory feedback. We establish observability through Lie derivative analysis, proving all internal states can be reconstructed from peripheral inflammation when feedback pathways are active (g₃≠0 or g₂≠0). The EKF uses analytically derived Jacobians and is tested under varying process noise (0.1 to 3%), measurement noise (5 to 30%), and initial uncertainty (5 to 50%) for healthy and pathological (MDS/AML) conditions.
Results: The EKF demonstrated robust convergence across all scenarios. Higher process noise introduced oscillatory errors but maintained tracking accuracy, while increased measurement noise produced smoother estimates with slower convergence. The filter showed remarkable insensitivity to initial errors, converging reliably even with 50% error. Observability analysis revealed active inflammatory feedback (g₃≠0) is essential for reconstructing marrow inflammation from peripheral measurements.
Conclusion: The EKF provides accurate estimation of hidden hematopoietic states from peripheral blood measurements.
