Evidence of Neuroplasticity Due to Acupuncture: An fNIRS Study

Background: Mild cognitive impairment (MCI) is a precursor to the critical disease known as Alzheimer’s. It is imperative to develop a proper treatment for this neurological disease in the aging society. This study investigates the effects of acupuncture therapy (AT) on MCI patients. Methods: Eleven healthy individuals and eleven MCI patients were recruited for this study. Oxy- and deoxy-hemoglobin signals in the prefrontal cortex during working-memory tasks were monitored using functional near-infrared spectroscopy. Before the AT, working-memory experiments were conducted for healthy control (HC) and MCI groups (MCI-0), followed by 24 sessions of AT for the MCI group. The AT sessions were initially carried out for six weeks (two sessions per week), after which experiments were performed again on the MCI group (MCI-1). This was followed by another set of AT sessions that also lasted for six weeks, after which the experiments were repeated again on the MCI group (MCI-2). Statistical analyses of the signals and classications based on activation maps as well as temporal features were performed. Results: The highest classication accuracies obtained using binary connectivity maps were 85.7% HC vs. MCI-0, 69.5% HC vs. MCI-1, and 61.69% HC vs. MCI-2. The classication accuracies using the temporal features mean (i.e., mean(5:28 s)) and maximum (i.e., max(5:28 s)) values were 60.6% HC vs. MCI-0, 56.9% HC vs. MCI-1, and 56.4% HC vs. MCI-2. Conclusions: The results reveal that there was a change in the temporal characteristics of the hemodynamic response of MCI patients due to acupuncture. This was reected by a reduction in the classication accuracy after the therapy, indicating that the patients’ brain response improved and became comparable to those of healthy subjects. Similar trend was reected in the classication using image feature. These results indicate that acupuncture can be used for the treatment of MCI patients.


Introduction
Neurological injuries and diseases are the major causes of the degradation of people's mental capabilities, and they can be treated via rehabilitation. One of the signi cant brain diseases is Alzheimer's disease (AD) [1]. AD starts with mild memory loss and eventually becomes severe, and it can develop into a neurodegenerative brain disease. According to a recent study, 70%-80% of dementia cases are caused by AD worldwide [2]. Mild cognitive impairment (MCI) is a neurological condition that occurs as a prior stage of AD. MCI is de ned by a loss of vocabulary, thought, perception, and judgment-related memory [3][4][5]. It progresses more rapidly than other diseases associated with age and can lead to AD. Therefore, early diagnosis and treatment of MCI-involving either therapy or medication-based care-is necessary, as it can easily develop into advanced AD. Thus, it is necessary to examine the effects of therapy (acupuncture) on MCI patients and evaluate its effectiveness for treating MCI.
Acupuncture is a non-pharmacological procedure based on the placement of tiny needles at particular locations on the human body that are known as acupoints or acupuncture points [6]. The primary objective of this therapeutic approach is to enhance the safety and quality of life for the patient. Different diseases like colds, fevers, and cerebral pains have been treated using this traditional technique of acupuncture [7]. While the early phases of acupuncture are still being investigated, it has been recognized as part of traditional Korean medicine. Acupuncture has been part of traditional medication for more than 2000 years [8]. Acupuncture therapy also has neurological outcomes consistent with pain relief and anesthesia [9].
Several studies have utilized fNIRS for examining the changes in the hemodynamic signals of MCI and AD patients while they perform cognitive tasks. A study focusing on right parietal region of brain found that the HbO concentration changes of MCI patients were signi cantly smaller as compared with healthy controls (HCs) [20]. A recent study indicated a signi cant loss of lateralization in MCI patients [21]. Another study indicated comparable task performance between MCI and HCs in behavior results.
Signi cantly enhanced activation was not observed in MCI patients compared with HCs in a high-load working-memory task [22]. Furthermore, Uemura et al. reported a reduced activation pattern in the bilateral dorsolateral prefrontal cortex in a retrieval task [23]. Improved cortical activation was observed in a dual-task (walking and verbal frequency task), and reduced activation was observed during normal walking in MCI patients [24]. In a comparative study on HC, MCI, and AD patients, it was revealed that the time taken to achieve the activation level of HbO in the prefrontal cortex of an HC was shorter than those for MCI and AD patients [25]. They also observed a steeper slope of HbO for MCI patients compared with AD patients. Another study involving MCI patients revealed no activation changes with pre-and postcomputer-based memory training; however, the behavioral performance was improved [26]. These results indicate the potential of fNIRS to be used as an imaging modality for the detection of diseases such as MCI and AD. However, the effects of pharmacological (medication) and non-pharmacological (i.e., acupuncture) intervention on MCI patients using fNIRS is still in the developing stage.
Regarding the acupuncture therapy, the initial study by our research group had investigated the longitudinal effects on the MCI patients: The analyses were primarily based on the Korean version of the Montreal Cognitive Assessment score (MoCA-K) and hemodynamic responses [27]. In this study, the functional connectivity and graph theory were analyzed. The results with and without acupuncture therapies were compared between the HC group and the MCI group. The improvement in mean MoCA-K test scores and averaged hemodynamics responses depicted that the cognitive function of the MCI group was enhanced upon acupuncture. After acupuncture therapy, the functional connectivity and graph theory parameters of the MCI group were comparable to those of the HC group. The increase in the MoCA-K score was positively correlated with fNIRS data that showed its capability as a feasible tool for the evaluation of cognitive functions in the clinical setting. Nevertheless, the conclusion on the averaged increase in fNIRS data and mean cognitive test scores were deduced on the basis of statistical analyses. In contrast, to further validate the impact of acupuncture, we aim to explore the changes at the trial-level (not averaged) temporal data, spatial activation patterns, and functional connectivity by employing advanced classi cation algorithms.
Therefore, in this study, we investigate the effects of acupuncture on MCI patients via classi cation. We aimed to determine whether acupuncture can improve the mental health of MCI patients in terms of brain activation. Additionally, we attempted to determine whether there are changes in the hemodynamic responses of the patients due to acupuncture with regard to the classi cation accuracies. In order to check our hypothesis, two different classi cation methods were adopted in the study, i.e. classi cation using temporal features and through a convolutional neural network (CNN).

Participants
For the experiment, 11 HC and 11 MCI patients were recruited. For consistency, the average age, sex, and number of years of education were identical between the groups. The demographic information for all the subjects has been summarized in Table 1. MCI patients were recruited at Dunsan Korean Medical Hospital. Newspapers were also used for advertisement. All the participants were informed that they could stop participating at any time. The medical history of the participants, along with demographic and drug-intake information, was obtained before the experiment. Additionally, on-spot assessments of the height, weight, and systolic blood pressure were performed. None of the participants had indulged in a brain signal acquisition experiment before this study. To evaluate the cognitive performance of the participants, MoCA-K test was performed. For the details of the test interested readers can refer to [27]. In the experiment, seven detectors and eight emitters were positioned over the region of interest. Figure 1 shows the optode con guration on the prefrontal cortex of the human brain. A total of 20 channels were formed using all the emitters and detectors. The optodes were placed by taking FpZ as a reference point. The reference points were set in accordance with the International 10-20 System of Electrode Placement. The separation between the emitter and detectors was ≤3 cm. Fig. 1 Emitter-detector con guration on the prefrontal cortex: FpZ is taken as a reference point in accordance with the International 10-20 System [27].
For the prefrontal cortex, a working-memory task was employed to detect neuronal activation. Nine trials were conducted during the experiment. Each trial consisted of a 24-s task period and a 14-s rest period. A "ready" cue of 2 s was given in each trial, before the start of the task period. This was done to alert the subject that a task was to be performed soon. During the task period, for the initial 8 s, 3-7 images randomly appeared on a screen placed in front of the subjects. After these 8 s, there was a period of 14 s during which the subjects were allowed to hold images that they selected (as per di culty). This was followed by a probe period of 2 s in which the participants had to respond, indicating whether the image displayed was among the images shown previously. To respond, the subjects used two push buttons to reply "yes" or "no." Before the commencement of the experiment session, there was a pre-rest period of 4 min. Similarly, at the end of the experiment, there was a 30-s post-rest period. Figure 2 shows the experimental paradigm. A black screen was displayed during the rest period. Throughout the experiment, the subjects were advised to keep their eyes open.

Intervention detail
The intervention details were presented in detail in a recently published work [27]. Brie y, the patients underwent several acupuncture sessions lasting 10 min each. A total of 24 sessions were conducted over twelve weeks (i.e., for brevity, before acupuncture: MCI-0, after 12 acupuncture sessions (6 weeks): MCI-1, and after 24 acupuncture sessions (12 weeks): MCI-2). During this time, the participants were instructed not to participate in any other acupuncture. The acupuncture was performed under the rules of Standards for Reporting Interventions in Clinical Trials of Acupuncture (STRICTA) [28]. Table 2 summarizes the intervention details. The study was performed in compliance with the Institutional Review Board of Dunsan Korean Medicine Hospital, Daejeon University [29]. Before the experiment, all the participants were given a complete description of the experimental procedure and provided written informed consent. For the MCI group, additional consent was obtained for the acupuncture. The entire study was in compliance with the latest Declaration of Helsinki [30].

Acquisition and preprocessing of fNIRS signals
In this study, the signals generated in the brain were sampled at a frequency of 7.81 Hz. A single-phase continuous wave fNIRS system NIRScout (NIRx Medical Technologies, USA) was used for the acquisition of fNIRS signals. Two different wavelengths were utilized by the system: 760 and 850 nm. The Modi ed Beer-Lambert's law was used to convert the raw intensities into changes in HbO (ΔHbO) and HbR (ΔHbR) [31]. This conversion was performed using the NIRSlab software.
After the data were converted into ΔHbO and ΔHbR, they were preprocessed to remove the effects of physiological noises. For this, a 4th-order Butterworth lter was utilized. A low-pass cutoff frequency of 0.15 Hz was used to remove cardiac, respiratory, and low-frequency drift signals [32,33]. The cutoff frequency for the high-pass lter was selected according to the longest time period of a single trial (i.e., 38 s (1/38 s = 0.026 Hz)) [34,35]. For analysis, the software MATLAB™ 17.0 was used (MathWorks, USA).
For analysis using images, the data for all the trials were rst converted into pictorial form. Activation maps and connectivity maps were used as image features. For the representation of the activation maps, a t-test was performed. The t-values for all the channels were calculated using the built-in function robust t of MATLAB ® . In the test, the statistical signi cance level was set as 0.05. To obtain connectivity matrices, the Pearson correlation coe cients were calculated for the data of each channel. These connectivity matrices were then used to create connectivity maps. The obtained images were classi ed using CNN. In this study, the neural network consisted of two convolutional layers and two fully connected layers. Again, vefold cross-validation was used. Additional details regarding the method were presented in a previous work [36]. For detecting cortical activation, the time-series signal for each trial was tted to the desired hemodynamic response (dHRF) signal [38][39][40].

Comparison of hemodynamic responses
The hemodynamic response function (HRF) is a typical response that is detected for healthy subjects and patients. To describe the shape of the HRF, the dHRF was used to identify the activation in each trial. After the active trials were extracted for all the subjects, the trials were averaged. The same procedure was followed for both the groups. Visual inspection revealed credible activation for both the patients and the healthy subjects. However, the responses of the MCI patients obtained before and after acupuncture sessions differed signi cantly. Figure 3 shows the maximum amount of activation observed for the MCI and HC groups, along with the standard deviation. For both groups, the amounts of activation were represented using bars, and the standard deviations were represented as dotted lines above them. To evaluate the statistical difference between the MCI group and the healthy group, t-tests were performed. While comparing the average amount of activation between the healthy subjects and MCI patients, two independent sample t-tests were performed. Whereas paired t-tests were used for the comparison of the activation between the MCI patients before and after acupuncture sessions. A difference was considered to be statistically signi cant if the following two conditions were satis ed: i) tvalue > critical t-value (t crt ) and ii) p-value < 0.05. The average amount of activation for healthy subjects was signi cantly higher as compared to MCI-0 (p-value < 0.001). Interestingly, after the rst twelve acupuncture sessions, there was no longer a statistically signi cant difference in the responses between the HC and MCI-1 groups (p-value = 0.377). However, statistically signi cant differences were observed when the data of the MCI-0 were compared with MCI-1 (p-value < 0.001). The activation in case of MCI-2 was lower than HC group (p-value < 0.001).
Similarly, statistically signi cant results were obtained while comparing the MoCA-K scores of HC and MCI-0 (p-value < 0.001). Interestingly, no statistical signi cance was found while comparing the MoCA-K score of HC with MCI-1 and MCI-2 (p-value = 0.074 and p-value = 0.099, respectively). Figure 4 shows the obtained average MoCA-K scores for both HC and MCI. The standard deviations in the score are represented by lines in the gure.
t-maps and connectivity maps are widely used in both fMRI and fNIRS for indicating the areas of activation. In this study, these maps were used for classi cation. Figure 6 shows the t-maps for each subject in HC and MCI groups.  Figure 7 shows the connectivity maps whereas the corresponding binary matrices are shown in Figure 8. For converting the connectivity maps into binary matrices, a threshold value of 0.8 was used. The trend of the connectivity maps was similar to that of the t-maps. These results support the claim that the acupuncture made the patients more normal. To verify the results based on functional connectivity, independent sample and paired t-tests were performed. The values of correlation for both healthy and MCI group were converted from r-values to z-values. This was done using Fisher's r-to-z transform. Signi cant difference was found while comparing the HC with MCI-0 groups (p-value < 0.001). Same was the case for the comparison of functional connectivity maps of MCI-0 and MCI-1 (i.e., p-value < 0.001). CNN was used for the classi cation of the images. Figure 9 presents the classi cation accuracies obtained using t-maps. Whereas Figure 10 and Figure 11 shows the classi cation accuracies obtained using connectivity maps and corresponding binary maps, respectively. As shown, the classi cation accuracies were changed due to the acupuncture. After the therapy, the classi cation accuracy decreased, which is similar to trend of the temporal features. However, among the three different types of maps used in the CNN-based classi cation, the binary matrices yielded the highest classi cation accuracy i.e., 85.7% for HC vs MCI-0, 69.5% for HC vs MCI-1, and 61.69% for HC vs MCI-2.

Discussion
The objective of this study was to investigate the effects of acupuncture on MCI patients and to validate it as a method for improving their mental capabilities. It was hypothesized that acupuncture would change the hemodynamic response. Our second hypothesis was that the acupuncture would change the activated area, resulting in making the patients response like healthy subjects. To the best of the author's knowledge, this was the rst study in the eld of fNIRS to investigate the effects of acupuncture with regard to classi cation. We aimed to determine whether acupuncture can make patients normal, by using fNIRS as a brain imaging tool. Our ndings are summarized below.
1. i) As a rst step, the brain activation area of the MCI group was compared with that of the healthy group. This was done both before and after the acupuncture. The brain activation patterns for the working-memory task were compared using a t-test. Initially, there were signi cant differences between the hemodynamic responses of the two groups (p-value < 0.001). After six weeks, over which 12 acupuncture sessions were conducted, interestingly, there was no signi cant difference between the hemodynamic responses of the MCI and healthy groups (p-value = 0.377). Thus, the statistical test validated our hypothesis, i.e., the acupuncture made the patients more normal.
2. ii) For classi cation using temporal features, a feature set consisting of eight features was extracted from the hemodynamic responses of both groups. These features were then used as pairs to classify the healthy and MCI groups. LDA was used for classi cation. The combination of mean(5:28 s) and max(5:28 s) features yielded the highest classi cation accuracies. According to the results, in most of the cases, the classi cation accuracy was higher before the acupuncture than after it. Validation based on temporal features was in accordance with our hypothesis and the results of the statistical test. The results obtained through classi cation are in line with MoCA-K scores. In a future study, advanced techniques can be utilized to further validate the hypothesis.
iii) The prefrontal cortex of the brain is the major contributor to most active-memory tasks, such as mental arithmetic, counting, and working-memory tasks, as it is sensitive to mental training [41,42]. With age, this area becomes weak with regard to capabilities, giving rise to dangerous diseases such as AD [43]. Such changes in mental health can be monitored using brain activation maps and brain connectivity maps. Thus, the effects of the acupuncture were validated using connectivity maps and activation maps.
Connectivity maps and activation maps were created for each trial. Using these maps, CNN-based classi cation was performed. For the MCI group, the acupuncture increased the activated area and enhanced the connectivity in the prefrontal cortex. The results indicated that with the acupuncture, the activation areas and connectivity maps of the MCI patients became more similar to those of the healthy subjects. Additionally, the classi cation accuracy decreased, supporting the aforementioned hypothesis.
These results indicate improvements in the patients and are in accordance with the literature [27,44].
The foregoing results support our hypothesis that acupuncture makes MCI patients more normal with regard to brain health. The enhancement of the patients' mental state due to acupuncture is clearly re ected by the ΔHbO signal, which is increased after acupuncture [25,45]. Although there is a decrease in the maximum value after MCI-2, but the activation is higher as compared to MCI-0. Also, the improvement in the cognitive abilities of the MCI patients is re ected in the MoCA-K test scores. The results obtained through classi cation are in line with the predecessor of the current study [27]. Also, from the results of classi cation, there is a notable difference between the classi cation accuracies obtained from temporal features and from image features. By using images of brain activation and connectivity higher classi cation accuracies were obtained. These results are in line with a recently published work by [36]. Overall, the results indicate that acupuncture is useful for enhancing the mental state of patients.
Owing to the small number of fNIRS channels available, the study was limited to the prefrontal cortex of the brain. In future research, a larger number of channels can be used, for investigating other brain areas [46]. Additionally, hybrid neuroimaging modalities can be used for brain signal acquisition [47]. This would be useful for verifying the hypothesis of the present study for the entire brain. Furthermore, it would allow observation of the activation and connectivity of the entire brain. To make the activation maps more precise, bundled optodes placement can be utilized to increase the spatial resolution [48]. Another limitation of this study was the number of subjects. Although the number of subjects was similar to those for previous studies, the study can be extended to a larger number of subjects. Moreover, only female participants participated in the study, representing another limitation, as the sex-based variability cannot be evaluated. In the future, identical research can be performed on male subjects, and the results can be compared with those of the present study.

Conclusion
This study represents a step toward analyzing the treatment of MCI for making the brain of patients more active. It was found that the traditional method of acupuncture can improve the mental health of patients, which was con rmed using classi cation methods. The results revealed that the acupuncture improved the hemodynamic responses of the patients. This claim is in line with the cognitive performance of the participants monitored through MoCA-K test scores. Additionally, the results indicated that the activated area of the MCI patients, as well as the connectivity, increased with the acupuncture. These changes in the brain have been proven statistically. Finally, the decrease in the classi cation accuracies indicated that the brain activation of the MCI patients became somewhat similar to those of the HC group due to acupuncture therapy.  Emitter-detector con guration on the prefrontal cortex: FpZ is taken as a reference point in accordance with the International 10-20 System [27].