Optimisation on the least squares identification of dynamical systems with application to hemodynamic modelling

Area of Science:

• Neuroscience
• Computational Biology
• Systems Biology

Background:

• Traditional least squares (LS) methods for dynamic modeling often produce biased and unstable predictions when using noisy input/output data.
• The cost function in LS methods relies on one-step-ahead prediction errors, which are sensitive to data noise.

Purpose of the Study:

• To develop a more robust dynamic modeling approach for noisy data.
• To improve the accuracy of model predictions in biological systems, specifically in relating cerebral blood flow and volume to neural activity.

Main Methods:

• Utilized model-predicted-output errors for parameter estimation, addressing the limitations of one-step-ahead prediction errors.
• Employed particle swarm optimization (PSO) to efficiently search for optimal model parameters due to the nonlinear cost function.
• Applied the novel algorithm to model the dynamic relationship between cerebral blood flow, cerebral blood volume, and neural activity.

Main Results:

• The model-predicted-output error approach demonstrated superior robustness in handling noisy input/output data compared to traditional LS methods.
• The developed algorithm produced more accurate predictions for the dynamic relationship between neural activity and hemodynamic responses.
• PSO effectively optimized the nonlinear cost function for parameter estimation.

Conclusions:

• The model-predicted-output error method offers a more reliable alternative to traditional least squares for dynamic modeling with noisy data.
• This approach enhances the prediction accuracy of neurovascular coupling models.
• The study highlights the potential of PSO in optimizing complex dynamic models in neuroscience research.