Several studies targeting patients with lesions in the medial temporal lobe (MTL) have reported impairments in short-term memory function [1,2,3,4,5]. Recent research in rats indicates that the coordinated neural activity patterns between the dentate gyrus and the hippocampal CA3 region contribute to working memory [6]. Additionally, it was reported that damage to the right MTL may lead to a heightened propensity for impairments in spatial memory, integral to navigation and self-location, while damage to the left MTL may increase deficits in verbal memory, essential for memorising auditory and visual linguistic information [7].

In recent years, electrical stimulation to deep brain regions has been established, leading to an increase in studies aiming to enhance memory targeting the MTL [8]. However, some studies specifically targeting the hippocampus have reported a decline in memory [9,10,11,12]. This decline is hypothesised to result from an acute depolarization block leading to memory impairment [9]. Akin to electrical stimulation, neurofeedback (NF) is one of the techniques for modulating brain function. NF converts brain activity into perceptible information, such as bar length or circle size, and provides feedback, allowing individuals to self-regulate their brain activity. Through repeated NF training, it is expected that the regulation is facilitated of cognitive functions associated with brain activity. NF has gained momentum in clinical applications, with reports of symptom alleviation in various mental disorders, such as anxiety, depression, schizophrenia [13], and ADHD [14]. Additionally, there has been an increase in opportunities for NF training in healthy individuals to enhance memory [15], meditation [16], and attentional focus [17]. Since NF relies on the brain’s autonomic learning capability to regulate neural activity, it offers a potentially lower risk of interfering with memory function compared to electrical stimulation. Moreover, anatomical evidence reveals that memory function involves not only the hippocampus but also multiple regions, such as the thalamus, mammillary bodies, and the cingulate gyrus [18].

Patients with drug-resistant epilepsy, undergoing intracranial electrode placement through craniotomy for the purpose of epileptic focus diagnosis, provide an opportunity for intracranial electroencephalogram (iEEG) measurements. Recently, there has been increased interest in applying this invasive brain activity measurement method to brain-computer interfaces [19]. Signals obtained from intracranial electrodes offer superior temporal resolution compared to functional magnetic resonance imaging (fMRI) and reduce the impact of artefacts compared to scalp EEG. Additionally, the signals are not attenuated by the dura mater, skull, or scalp tissue. These advantages can be beneficial in NF studies, such as those focusing on modulating somatosensory–motor function [20,21,22] or emotional function [23]. However, these approaches are still in their early stages, and there are few NF studies specifically targeting memory function in the MTL [24].

The mechanism behind NF training is believed to be rooted in operant conditioning [25]. Operant conditioning involves changing the occurrence probability of a response under a specific condition by providing a consequence (reinforcement or punishment) in response to voluntary behaviour (operant behaviour). The “specific condition” serves as an antecedent that functions as a cue for the voluntary response. Operant conditioning involves learning the association, known as the three-term contingency, between antecedent, behaviour, and consequence.

In the context of NF, considering an illustrative scenario where the user can increase the length of a bar displayed on the screen when a specific brain activity pattern occurs during a specific cognitive condition. Through NF training, as the user engages in trial and error, they may accidentally produce the desired brain activity pattern and achieve success in increasing the length of the bar. Through repeated these experiences, the user’s probability of generating the desired brain activity pattern in a specific cognitive condition gradually increases. In other words, in NF, the user learns the contingency between the specific cognitive condition (antecedent), the brain activity pattern (behaviour), and the success in bar length adjustment (reinforcement).

The presence of consequence is crucial for the success of operant conditioning. If the reinforcements or punishments following operant behaviour are insufficient, the changes in the occurrence probability of the response will be weak, making it difficult for learning to take place. A neurophysiological theory of NF suggests the significant contribution of the striatum, a part of the reward system, to NF [26], and indeed, several NF studies have reported the observed involvement of the striatum [27,28,29]. These emphasise the importance of the consequence in NF as well.

NF studies targeting memory have extensively used scalp EEG [15,30,31,32,33,34,35,36,37]. Memory function has long been suggested to be associated with theta oscillations [38,39,40], and an increase in theta power during memory encoding has been reported in many EEG studies [38,41,42]. Based on the accumulation of such foundational research, many memory EEGNF studies have been implemented for increasing theta power. So far, previous studies have shown that upregulation of theta activity in the frontal midline improves control processes during memory retrieval [31] and accelerates and enhances memory consolidation [33,36]. Studies utilising intracranial electrodes have also reported increased theta power in the MTL during successful encoding and retrieval [43,44,45,46,47,48,49], and increased theta phase synchronization between the hippocampus and other cortical regions (rhinal cortex, prefrontal cortex) during memory encoding [50,51]. However, some studies also report a decrease in MTL theta power during encoding [52,53,54,55,56,57,58], and the causal relationship between MTL theta power changes (increases or decreases) and memory function (encoding success or failure) is inconsistent across studies.

In NF studies aiming to improve mental disorders, it is common to conduct NF during a resting state without imposing additional tasks on participants. This is because the resting state already corresponds to the “specific cognitive condition” in the learning mechanism of NF. In other words, in mental disorders, the chronic symptoms typically persist even during the resting state, which is why NF in this state enables the learning of three-term contingency. The same applies to healthy individuals training their attention, relaxation, or meditation using NF. On the other hand, when conducting NF training for memory function, the “specific cognitive condition” corresponds to the condition requiring memory function. Therefore, it is essential to create a situation that necessitates memory function, such as imposing memory tasks on participants, and conduct NF within that context to efficiently learn the three-term contingency and achieve successful memory NF. Traditional memory NF studies have often required several weeks or more for participants to gain self-regulation of brain activity [15,59]. This may be attributed to the fact that they were conducted during resting states, without creating a situation that necessitates memory function. Additionally, during the resting state, participants lack relevant clues on how to control their brain activity to adjust the feedback signal, leading to higher difficulty in learning self-regulation and requiring an extended time for the effects to appear. If NF can be conducted within a context that necessitates the desired brain function, participants will have clues on how to change the feedback signal, potentially making it easier to acquire self-regulation of brain function in a shorter period.

However, when participants perform tasks requiring a specific brain function while simultaneously receiving feedback, dual-task interference occurs, diverting attention away from the feedback, making self-regulation difficult, and vice versa. To address this issue, a task-based NF approach has been proposed [60,61], where the periods for task performance and feedback are separated, allowing for alternating short periods of task engagement and feedback. This approach enables participants to engage in tasks and feedback without divided attention. In this study, we adopt this task-based NF approach to investigate whether the neural activity in the MTL associated with memory function can be modulated.