关键词: 脑机接口, 侵入式电极, 神经调控, 闭环刺激

Abstract: Brain-Computer Interface (BCI) technology offers a transformative approach to establishing direct communication pathways between the brain and external devices, heralding a new era in neuroscience and clinical neurology. While significant progress has been made in applying BCI to restore motor function in paralyzed individuals, research is rapidly advancing into the more complex domain of decoding and modulating higher cognitive functions such as memory and emotion. We provides a comprehensive review of invasive BCI research focused on episodic memory, with particular emphasis on the neural mechanisms underlying its core component—spatial memory—and the critical influence of emotional valence. Memory dysfunction is a central pathological feature of many neurological and psychiatric disorders, as seen in the progressive memory loss characteristic of Alzheimer’s disease (AD) and the maladaptive consolidation of traumatic memories in post-traumatic stress disorder (PTSD). The current lack of effective treatments for these conditions highlights the urgent clinical need for precise neuromodulation technologies.
The first part of this review synthesizes recent breakthroughs in decoding the neural substrates of spatial memory and emotional valence, hallmarked by a critical paradigm shift in experimental approaches. Intracranial recording studies in humans based on real-world navigation tasks provide more valuable insights than constrained desktop-based virtual reality paradigms, due to their higher ecological validity. These studies indicate that hippocampal theta oscillations serve as a robust biomarker for decoding movement states and velocity, characterized by short-duration bursts whose frequency correlates with movement speed. Furthermore, the entorhinal cortex exhibits grid cell-like periodic activity, working together with "memory trace cells" in the hippocampus to encode spatial locations and goal representations. Beyond self-motion information, environmental boundaries are also distinctly represented through specific oscillatory patterns in the medial temporal lobe, a process strongly modulated by task relevance and cognitive demand. Concurrently, research on emotional valence has identified distinct neural circuits: negative emotions such as fear are linked to beta-band synchrony in the amygdala-hippocampal network, whereas positive valence and reward processing involve the dopaminergic system and are reflected in high-frequency activity and sharp-wave ripples in the hippocampus. These detailed mechanistic insights offer a foundational map of neurobiological targets for selective intervention.
Building on this advanced in decoding, the field is undergoing a strategic pivot from open-loop to closed-loop neuromodulation. Traditional open-loop deep brain stimulation (DBS), which uses fixed stimulation parameters, has shown limited efficacy and poor adaptability to the brain's dynamic states, with effects often being inconsistent across different memory tasks. In contrast, next-generation closed-loop BCIs can monitor specific neural signatures in real time—such as particular oscillatory couplings during memory encoding or slow-wave-ripple events during sleep-dependent consolidation—and deliver personalized, spatially targeted electrical stimulation at optimal moments. Recent empirical evidence from human studies confirms that this adaptive "monitoring-decoding-intervention" framework can significantly enhance both memory encoding efficiency and long-term consolidation. For instance, stimulation triggered by detected slow-wave oscillations during NREM sleep has been shown to enhance the coupling between slow waves, spindles, and ripples, thereby improving memory performance. This represents a critical evolution from static, unidirectional stimulation to a dynamic, bidirectional brain-machine dialogue.
Looking forward, the convergence of dynamic network neuroscience models—which provide the real-time 'blueprint' of brain network activity—with next-generation electrodes—which provide the tools for high-resolution interaction—paves the way for building truly adaptive systems that can dynamically interpret and respond to brain states. Future developments will need to address the challenges of profound individual variability in neural signatures and ensure the long-term stability of recording interfaces. Furthermore, advancing beyond simple classification to the reconstruction of rich memory content requires more sophisticated decoding algorithms capable of handling the nonlinear, non-stationary nature of neural signals across different brain states. The integration of deep learning models with high-density, flexible electrode arrays may enable more precise mapping of distributed memory networks. Despite these challenges, closed-loop BCI technology—grounded in precise mechanistic insights from cognitive neuroscience—holds great potential to revolutionize the treatment of memory disorders. By enabling real-time, adaptive interaction with the brain's memory networks, this technology paves the way for personalized therapeutic strategies that can dynamically respond to the changing needs of patients with neurological and psychiatric conditions, ultimately restoring cognitive function and improving quality of life.

Key words: brain-computer interfaces (BCIs), invasive electrodes, neuromodulation, closed-loop stimulation