Identification negative BOLD responses using windkessel models
Alongside positive BOLD responses (PBR), a variety of negative BOLD responses (NBR) with distinct underlying mechanisms also occur. We identify five mechanisms of NBR: i) local/lateral/contralateral inhibition (LCI), ii) neuronal disruption of network activity (NDA), iii) altered balance of neuro-metabolic/vascular couplings (ANC), iv) arterial blood stealing (ABS), and v) venous blood backpressure (VBB). Detecting and classifying these mechanisms from BOLD signals is pivotal in understanding normal/pathological brain functions. This requires models and parameters with anatomical/functional interpretation that furnish the understanding of how these mechanisms are fingerprinted by their BOLD responses. Here, we used a windkessel model with viscoelastic compliance as well as dynamics of both neuronal and tissue/blood O2 to investigate the generation, detection, classification and interpretation of the BOLD hemodynamic response functions (HRF) of above mechanisms. Firstly, we evaluated the use of the general linear model to detect simulated NBRs. Secondly, we tested the ability of a machine learning classifier, built from a simulated ensemble of HRFs, to predict the mechanism underlying a new HRF. Crossvalidation indicates NDA and ANC can accurately be classified solely from fMRI BOLD signals; while LCI, ABS and VBB might require additional imaging modalities. Thirdly, we demonstrated that estimators of the model parameters determinant in the NBRs formation are accurate, and precise to certain resolutions. Finally, we successfully applied our detection/classification/estimation methodology to EEG-fMRI data in a clinical situation where several of these mechanisms could coexist. We believe that the proper identification and interpretation of NBR mechanisms have important clinical and cognitive implications in fMRI studies.