The effects of acupuncture facilitating neural plasticity for treating diseases have been identified by clinical and experimental studies. In the last two decades, the application of neuroimaging techniques in acupuncture research provided visualized evidence for acupuncture promoting neuroplasticity. Recently, the integration of machine learning (ML) and neuroimaging techniques becomes a focus in neuroscience and brings a new and promising approach to understand the facilitation of acupuncture on neuroplasticity at the individual level. This review is aimed at providing an overview of this rapidly growing field by introducing the commonly used ML algorithms in neuroimaging studies briefly and analyzing the characteristics of the acupuncture studies based on ML and neuroimaging, so as to provide references for future research.
Neuroplasticity usually refers to brain plasticity, which means the ability of the brain to modify its organization to the altered demands and environments [
In the past two decades, studies on acupuncture promoting brain plasticity were greatly enhanced with the development of neuroimaging techniques. Several studies focused on investigating acupuncture-induced brain structural and functional plasticity by magnetic resonance imaging (MRI), positron emission tomography (PET), and other neuroimaging methods [
Currently, most neuroimaging findings of acupuncture facilitating neuroplasticity were obtained by the standard univariate analysis. It means the results were only significant at the group level, which limited their clinical translation to a certain extent. So, it is of great value to investigate how acupuncture promotes neuroplasticity and how the specific neuroplasticity affects the responses to acupuncture from the individual level. The application of multivariate pattern analysis (MVPA) and machine learning (ML) in neuroimaging studies provides an attractive method to this issue [
Numbers of publication on neuroimaging and machine learning in the last decade (from January 1, 2010, to June 1, 2020). The data was obtained by searching at the PubMed database with the items (Neuroimaging) AND (Machine Learning).
Therefore, we conducted this review by introducing the most widely used ML algorithms in neuroimaging studies briefly and analyzing these applications in the fieldof acupuncture promoting neural plasticity, aiming to provide an overview of this rapidly growing field and new approaches in future research.
ML is a subfield of artificial intelligence which is aimed at investigating how computers can improve decisions and predictions based on data and ongoing experience [
The SVM is so far the most popular supervised learning algorithm in neuroimaging studies and is widely utilized in classification and prediction [
DT is the rooted directed tree that predicts the output based on a sequence of splits in the input feature space. The nodes split at each step by optimizing a metric, which indicates the consistency between the estimates and truth values. When the node has no subordinate to split, the traversal of this tree generates the target outcome prediction. As a typical classification algorithm with high interpretability, DT is applied predominantly for classification and disease diagnosis in neuroimaging studies [
RF is generally the ensembles of DTs [
The concept of ANN is derived from the biological neural network. Similar to the synaptic connection in the brain, an ANN is composed of several layers of interconnected artificial neurons that make up the input layer, hidden layer, and output layer. As an ultracomplex ML algorithm, ANN establishes the computational units of multiple layers by simulating signal transmission and learning the architecture of synapse [
The diagrams of the above algorithms are summarized in Figure
Diagrams of the commonly used machine learning algorithms in neuroimaging studies. SVM: support vector machine; DT: decision tree; RF: random forest; ANN: artificial neural network.
In this review, we focused on the application of neuroimaging and ML in acupuncture promoting neuroplasticity. For a comprehensive summary of the field, we systematically searched papers in PubMed (pubmed.ncbi.nlm.nih.gov), Web of Science (
These ten studies were published from 2008 to 2020. Generally, for participant selection, these studies were performed on healthy subjects [
The detailed characteristics of these included studies were displayed in Table
The detailed characteristics of the included studies.
Participants | Intervention | Modality | Feature | Purpose (C/R) | ML | Feature selection | Validation | Model assessment | MVPA findings | Univariate analysis results | Conclusion | |
---|---|---|---|---|---|---|---|---|---|---|---|---|
López et al., 2013 | Migraine | One session of verum or sham ACU stimulation | Task-SPECT (function) | Blood perfusion | C | Linear SVM | Filter: discarding voxels with intensity values under 25% of the maximum | LOOCV | ACC | The classifier performed better when the training data was extracted from the verum ACU group than from the sham ACU group. | Verum ACU yielded greater changes in the perfusion patterns than sham ACU. Verum ACU produced a more significant decrease in blood perfusion. | SVM can distinguish the SPECT images of pre- and post-ACU acquisitions. Changes in blood perfusion following verum ACU is greater than sham ACU. |
Jung et al., 2019 | HS | ACU at L HT7 or L PC6 for 20 blocks | Task-fMRI (function) | BOLD signal | C | Linear SVM | No feature selection steps | LOOCV | ACC | The classifier got an accuracy of 58.6% for classifying HT7 and PC6 with the features extracted from SI, MI, paraCL, anterior and posterior insula, SMG, ACG, vmPFC, PPC, and IPL. Using signal of ROI as feature, the classifier got higher accuracy (MI, 65%; SMA, 64%; SMG 62%; SI, 62%; and dlPFC, 62%). | No significant difference in BOLD signal alteration following HT7 and PC6 stimulation | Spatial localization of pain perceptions to ACU needle can be predicted by the neural response patterns in the somatosensory areas and the frontoparietal areas. |
Yu et al., 2019 | HS | One session of TR or LT manipulation at ST36 | Task-EEG (function) | Graph theory | C | DT | Selecting the features of interest | 6-fold | ACC | The classifier got an accuracy of 92.14% and AUC of 0.9570 with all graph theory features as inputs. With the increase of filter number, the accuracy was gradually improved. The highest accuracy was 92.37% with 6 filters in the TSK model. | PLV of TR was stronger than the baseline, while PLV of LT was weaker than the baseline. The value of all the six graph theory features of TR was significantly lower than that of LT. | Different ACU manipulations have different effects on functional brain networks. Classification of different ACU manipulations based on EEG with network features is feasible. |
Liu et al., 2018 | MWOA | 24 sessions of sham ACU at NAP in 8 weeks | DTI (structure) | TABA | C | Linear SVM | Filter+wrapper: traversing the | LOOCV | ACC | The single FA, MD, AD, and RD of the mPFC-amygdala fiber contributed to lackluster classification accuracy. The classifier got a higher accuracy with the combined features of FA, MD, and RD (in which ACC, SEN, SPE, PPV, and NPV were 84.0%, 90.2%, 76.7%, 82.1%, and 86.8%, respectively). The external capsule, ACG, and mPFC significantly contributed to the discrimination of responders and nonresponders. | The increased FA, decreased MD, decreased AD, and decreased RD of the mPFC-amygdala fiber were detected in MWOA patients than HS. | The variability of placebo treatment outcomes in migraineurs could be predicted from prior diffusion measures along the fiber pathways of the mPFC-amygdala. |
Yang et al., 2020 | MWOA | 12 sessions of ACU at GV20, GV24, bil-GB13, bil-GB8, and bil-GB20 in 4 weeks | T1 (structure) | GMV | C | Linear SVM | Filter+wrapper+embedded: traversing the | 10-fold | ACC | Using the clusters located at the frontal, temporal, parietal, precuneus, and cuneus gyri as features, the classifier got the SEN of 73%, SPE of 85%, ACC of 83%, and AUC of 0.7871. | The baseline GMV in all predictive regions significantly differed between responders and nonresponders. Alterations of migraine days were correlated with the baseline GMV of L cuneus, R MiFG/IFG, L IPL, and SPL/IPL. The responders achieved an increase in GMV of the L cuneus after ACU. | The pretreatment brain structure could be a novel predictor for ACU treatment of MWOA. |
Tu et al., 2019 | cLBP | 6 sessions of ACU in 4 weeks, 8-12 effective acupoints were used in the real ACU group; 12 sham points were used in the sham ACU group. | Resting-fMRI (function) | ICA+rsFC | R | RBF SVR | Selecting the features of interest | 5-fold | The prediction model obtained an | Changes of pain severity correlated with baseline mPFC-SN and mPFC-AG FC in the real ACU group. Baseline mPFC-dACG FC was correlated with changes in pain severity in the sham ACU group. Changes of FC between the mPFC and insula/AG were correlated with the relief of pain severity after real treatment, while changes of FC between the mPFC and paraCL/SPL were correlated with the relief of pain severity after sham ACU treatment. | Pretreatment rsFC could predict symptom changes for real and sham treatment, and the rsFC characteristics that were significantly predictive for real and sham treatment differed. | |
Xue et al., 2011 | HS | ACU at GB40 or KI3 for 3 blocks, switching after a one-week interval | Task-fMRI (function) | BOLD signal | C | Linear SVM | Singular value decomposition | / | SDM | The performance of the classifier was not mentioned in this study. ACU stimulation at GB40 produced predominantly signal increases in the insula, red nucleus, thalamus, and amygdala. ACU at KI3 elicited more extensive decreased neural responses in the MFG, PCC, thalamus, and ACG. | ACU at GB40 and KI3 can both evoke similar widespread signal decreases in the limbic and subcortical structures. | Neural response patterns between ACU stimulation at GB40 and KI3 are distinct. Conventional GLM analysis is insensitive to detect neural activities evoked by ACU stimulation. |
Yin et al., 2020 | FD | 20 sessions of ACU in 4 weeks. One or two acupoints among CV12, ST36, and BL21 were used. | Resting-fMRI (function) | rsFC | C | Linear SVM | Wrapper: recursive feature elimination | LOOCV | ACC | The classifier obtained an ACC of 84.9%, SEN of 78.6%, SPE of 89.5%, and AUC of 86.8%. The FC between R insula-L precuneus, L MiOFG-L thalamus, L insula-L ACG, R ACG-R temporal pole, R SOG-R cerebellum-3 contributed crucial information for prediction. | / | The whole-brain resting-state functional brain network has good predicting potential for ACU treatment to FD patients. |
Hao et al., 2008 | HS | One session of electro-ACU at ST36 | Task-EEG/ECG (function) | BIS | R | FNN | Selecting the features of interest | Validation with an independent set | AAE | With the FNN, the AAE of the estimation and true value is 10.2278. | / | The alteration of |
Li et al., 2010 | HS | ACU at GB37 or NAP for 2 blocks | Task-fMRI (function) | BOLD signal | C | Linear SVM | Searchlight+singular value decomposition | LOOCV | ACC | The occipital cortex, limbic-cerebellar areas, and somatosensory cortex could help to differentiate the central neural response patterns induced by real or sham ACU stimulation with higher accuracy above the chance level. | Compared with the sham group, the ACU group induced higher signal intensity at some major regions of limbic-cerebellar system and small regions of the primary somatosensory cortex and supplementary motor area. | Neural response patterns of brain cortex to the ACU stimulation at GB37 and a nearby NAP could differ from each other effectively with the application of the MVPA approach. |
M/F: male/female; Y: year; C/R: classification/regression; ML: machine learning; MVPA: multivariate pattern analysis; ACU: acupuncture; SPECT: single-photon emission computed tomography; SVM: support vector machine; LOOCV: leave-one-out-cross-validation; ACC: accuracy; SPE: specificity; SEN: sensitivity; HS: healthy subjects; fMRI: functional magnetic resonance imaging; BOLD: blood oxygenation level dependent; TR: twirling-rotating manipulation; LT: lifting-thrusting manipulation; EEG: electroencephalogram; DT: decision tree; NB: naïve Bayes; KNN:
According to aims and design, these studies can be divided into three types. Among them, three studies [
Acupoint specificity refers that acupoints have different therapeutic effects and biophysical characteristics compared to sham acupoints and that different acupoints have relatively different therapeutic effects and biophysical characteristics [
Acupoint specificity is not only the core of acupuncture theory and the base of clinical practice but also the focus of acupuncture-neuroimaging research [
Two [
Acupuncture manipulation is the key in acupuncture clinical practice and significantly affects acupuncture efficacy [
The integration of ML and neuroimaging features has been extensively employed in predicting the clinical efficacy of drugs or other interventions [
These five studies on acupuncture efficacy prediction demonstrated that the specific neuroplasticity features including morphology of gray matter and white matter and cerebral functional activity patterns contained vital information for predicting the response of patients to acupuncture stimulation. The integration of ML and neuroimaging provides a new and promising approach for investigating mechanisms of acupuncture efficacy at the individual level, which has great potential for clinical translation and will be the important growth pole in acupuncture research.
In addition to the three aspects described above, there are still some other concerns that should be focused in future neuroimaging-based ML studies, for example, investigating the influences of acupuncture with different acupoint combination or different stimulation intensity on neural plasticity and predicting clinical efficacy of acupuncture with the neuroimaging features acquired under acupuncture stimulation.
The application of neuroimaging techniques in acupuncture mechanism has produced remarkable advance[
Due to difficulties in data acquisition, the sample size of neuroimaging study is generally small [
Considering that there are generally more features than samples in neuroimaging data, it is beneficial to take appropriate manners to eliminate the redundant features and reduce the dimension of data. The ten studies included in this review indicated that when using a single feature as input, the accuracy of the classifier is lackluster, whereas when multiple neuroimaging features applied, the accuracy of the model was significantly improved [
The current ML studies generally favor seeking homogeneous subjects to establish classification and prediction models [
The goal of ML is establishing mappings between training data and labels and then use the mappings as benchmarks for predicting the labels of the unseen data. Similar to other ML studies [
In summary, we provided an overview of the literature on the application of ML and neuroimaging in acupuncture promoting neural plasticity. Studies published so far have preliminarily demonstrated at the individual level that different acupoint stimulation and different acupuncture manipulations had significantly different real-time modulatory effects on functional brain plasticity and that the specific structural and functional neuroplasticity features at baseline could accurately predict the improvement of symptoms following acupuncture treatment. Although this research field is currently in its early stage and faces many challenges, we still believe that integrating ML and neuroimaging techniques will be a promising approach to understand the facilitation of acupuncture on neuroplasticity in the future.
There is no original data in this review.
The authors declare that they have no competing interests.
Tao Yin, Peihong Ma, and Zilei Tian contributed equally to this work.
This work was supported by the National Key R&D Program of China (No. 2018YFC1704600 and No. 2018YFC1704605), the National Natural Science Foundation of China (No. 81973960), and the Sichuan Science and Technology Program (No. 2020JDRC0105).
The details of data acquisition and literature selection process.