Brain network evaluated by functional-guided reinforcement learning effective connectivity indicates theraputic effect in tinnitus patients

Motivation: We hypothesize that the investigation of brain FC may serve as a foundational step in exploring EC. Goal(s): To enhance the precision of identifying effective connectivity between distinct brain regions. Approach: We adopts the actor-critic framework with an encoder-decoder model as the actor network. The encoder utilizes the Transformer model, and the decoder employs a bidirectional long-short-term memory (LSTM) network with attention. Results: Tinnitus patients were feature by enhanced information output of the motor network, somatosensory network, and the visual network. Effective treatment is featured by a similar pattern of FGEC network between tinnitus patients after treatment and the healthy controls. Impact: The pattern of FGEC network can be regarded as a direct evidence to indicate the effectiveness of sound therapy for tinnitus patients prior to treatment.