Deep brain stimulation (DBS) has the potential to ameliorate the motor symptoms associated with Parkinson’s disease, such as bradykinesia, rigidity, and tremor. However, the precise therapeutic mechanism underlying DBS in Parkinson’s disease remains inadequately comprehended, impeding the advancement of personalized DBS treatments. This study introduces a bio-inspired multi-scale brain model driven by blood oxygenation-level-dependent (BOLD) signals to explore the neural mechanisms underlying DBS effects on Parkinson’s disease. The model integrates microscopic neural dynamics with macroscopic brain function, utilizes functional Magnetic Resonance Imaging (fMRI) data to uncover the neural basis behind observed brain functional changes. The experiments include 27 Parkinson’s disease patients and 30 healthy controls. Each Parkinson’s disease patient has been conducted DBS surgery targeted on subthalamic nucleus (STN), and the fMRI data are recorded both during DBS-ON and DBS-OFF conditions. Based on our proposed model structure, we fit all the free coupling parameters within the microscopic cortico-basal ganglia-thalamic circuit (CBTC) to match the subject-specific functional connectivity matrix calculated from the fMRI data of each subject. After model validation, we further conduct a three-step deep exploration based on it. Firstly, it is found that an increase in GABAergic transmission into the thalamus has been associated with the exacerbation of rigidity symptom (p = 0.005**), whereas a reduction in GABAergic projections from interneurons within the cortex to pyramidal neurons has been correlated with an elevation in the severity of bradykinesia (p = 0.023*), indicating a “push” effect in the CBTC to impel the symptom-specific coupling intensity to an abnormal state. Secondly, the elevation of GABAergic signaling from the external globus pallidus to the internal globus pallidus indicate a strong correlation with the amelioration of rigidity symptom (p = 0.026*), while the attenuation of excitatory cortical projections to the STN is significantly associated with the remediation of bradykinesia (p = 0.048*). Lastly, the disparity in coupling strength pre- and post-DBS activation is underscored, implying synaptic coupling alterations found in the second step are induced by STN-DBS, which may reveal DBS has the ability to “pull” abnormal network back to a healthy functional state by the directly or indirectly restoration of the loop synaptic characteristics, therefore, normalizing these synaptic couplings. This work provides a promising approach to explore the intrinsic micro-regulatory mechanisms of DBS by interpreting the macroscopic fMRI information, offering new insights into the “push-pull” network dynamics of the CBTC and their implications for motor symptom-specific changes and treatments in Parkinson’s disease.