Brain Network Evaluation by Functional-Guided Effective Connectivity Reinforcement Learning Method Indicates...

Area of Science:

• Neuroscience
• Medical Imaging
• Computational Biology

Background:

• Current functional connectivity (FC) and effective connectivity (EC) methods have limitations in precisely delineating brain networks from fMRI data.
• Understanding brain network mechanisms is crucial for tinnitus treatment efficacy.

Purpose of the Study:

• To introduce a novel functionally guided effective connectivity with reinforcement learning (FGECRL) method for enhanced precision in identifying brain region connectivity.
• To investigate brain network alterations in tinnitus patients undergoing sound therapy and correlate them with treatment outcomes.

Main Methods:

• Developed an FGECRL approach utilizing an actor-critic framework with transformer and bidirectional LSTM encoder-decoder networks.
• Constructed FGEC networks from fMRI data of 65 tinnitus patients and 28 healthy controls before and after 6 months of sound therapy.
• Compared FGECRL with traditional FC and EC in discriminating between effective and ineffective treatment groups.

Main Results:

• FGECRL demonstrated superior accuracy in discriminating between patient groups compared to FC and EC.
• Effective tinnitus treatment showed FGEC network patterns similar to healthy controls, with decreased activity in motor and sensory networks.
• Persistent decreased EC in the primary auditory cortex indicated failed sound therapy, supporting Bayesian inference theory in tinnitus perception.

Conclusions:

• The FGECRL method offers enhanced precision for identifying brain network features relevant to tinnitus.
• FGEC network patterns provide biological evidence for the mechanism of sound therapy in tinnitus patients with varying treatment responses.