Unveiling the phantom: What neuroimaging has taught us about phantom limb pain

Abstract Phantom limb pain (PLP) is a complicated condition with diverse clinical challenges. It consists of pain perception of a previously amputated limb. The exact pain mechanism is disputed and includes mechanisms involving cerebral, peripheral, and spinal origins. Such controversy limits researchers’ and clinicians’ ability to develop consistent therapeutics or management. Neuroimaging is an essential tool that can address this problem. This review explores diffusion tensor imaging, functional magnetic resonance imaging, electroencephalography, and magnetoencephalography in the context of PLP. These imaging modalities have distinct mechanisms, implications, applications, and limitations. Diffusion tensor imaging can outline structural changes and has surgical applications. Functional magnetic resonance imaging captures functional changes with spatial resolution and has therapeutic applications. Electroencephalography and magnetoencephalography can identify functional changes with a strong temporal resolution. Each imaging technique provides a unique perspective and they can be used in concert to reveal the true nature of PLP. Furthermore, researchers can utilize the respective strengths of each neuroimaging technique to support the development of innovative therapies. PLP exemplifies how neuroimaging and clinical management are intricately connected. This review can assist clinicians and researchers seeking a foundation for applications and understanding the limitations of neuroimaging techniques in the context of PLP.

notable for advocating formal scientific investigations (Finger & Hustwit, 2003). This mystery continues to draw curiosity and interest, and it now leverages modern technology to uncover its truth.
More than merely quenching scientific curiosity, further PLP research is needed to improve the lives of amputees. The amputee population in America is estimated to increase from 1.6 million in 2005 to 3.6 million by 2050 (Ziegler-Graham et al., 2008). PLP, which is the pain of the amputated limb that is variable in timing, is experienced by up to 79.9% of amputees (Ehde et al., 2000;Ephraim et al., 2005). The severity of phantom limb sensations ranges from nonpainful to disabling and, in some instances, can be physically and psychologically debilitating . For surgical amputees, chronic preoperative pain and acute postoperative phantom pain are risk factors for PLP (Hanley et al., 2007;Larbig et al., 2019). Still, correlations for the severity of pain have been inconsistent (Sherman et al., 1984). These variations in the clinical presentation of PLP continue to burden amputees, warranting a deeper understanding of its mechanism to improve diagnosis and efficacious clinical approaches. As the demand for a conclusive understanding grows, so does the controversy among scientists and clinicians.
PLP is a particularly challenging syndrome to diagnose and treat, which may be related to the fact that, by mechanistic nature, it is challenging to understand. Nonspecific and highly varied symptoms can make diagnosing PLP difficult, requiring a comprehensive history, examination, tests, and exclusion of other possible neuropathies (Ferraro et al., 2016). The variety of PLP therapies, including pharmacologic, cranial stimulation, and sensory therapies, have been inconsistent (Aternali & Katz, 2019;Richardson & Kulkarni, 2017) with some potential being demonstrated among integrative approaches (Subedi & Grossberg, 2011). These challenges parallel the equally complex range of mechanistic explanations of PLP, which involve various combinations of cerebral, spinal, and peripheral nervous system pathologies (Flor et al., 2006). These mechanistic, diagnostic, and management inconsistencies underscore the importance of foundational tools for analyzing PLP. Due to the brain's role in interpreting, processing, and modulating pain, neuroimaging may fulfill this need. In addition to aiding understanding of PLP, neuroimaging may assist the development of PLP therapies.
This article will review current noninvasive imaging modalities for PLP research in mechanistic and therapeutic investigations. It will focus on diffusion tensor imaging (DTI), functional magnetic resonance imaging (fMRI), electroencephalography (EEG), and magnetoencephalography (MEG). Researchers and clinicians utilize each imaging modality in distinct ways to complement dynamic research involving diagnosing, characterizing, and treating PLP. By examining how DTI, fMRI, EEG, and MEG have impacted the understanding of PLP, we aim to summarize a baseline of fundamental imaging techniques to foster further research.

Background on potential pain mechanisms
In order to understand the impact of the major imaging modalities used in this field, it is helpful to briefly review the prevailing discussed potential mechanisms behind PLP. In the properly functioning human nervous system, peripheral noxious stimulus generates a sensation of pain via a cascade of neuronal events . The pathway consists of primary   afferent pain fibers which carry the afferent signals from the peripheral nociceptors to the spinal cord where they synapse directly or indirectly via interneuron with the secondary neurons at the dorsal horn of the spinal cord. The afferent signals then ascend via the secondorder neurons to the brain via either spinothalamic or spinoreticular tracts (Steeds, 2009). Passage of this nociceptive information through the brainstem triggers modulatory signals back through the dorsal horn. These modulatory signals alter primary afferent neuron propagation which can facilitate or inhibit further peripheral nociceptive information (Renn & Dorsey, 2005 are also important sensory discriminatory, affective, and modulatory regions, respectively (Leung, 2020). Furthermore, supraspinal modulatory functional connectivity deficits have been associated with white matter tract deficits, emphasizing the vital role of supraspinal processing (Leung et al., 2016(Leung et al., , 2018. These normal pain mechanisms involve the intricate relationship between peripheral, spinal, and supraspinal regions. PLP draws attention because there is still no consensus on its mechanism. Cerebral, spinal, and peripheral explanations each bear scientific evidence, perpetuating the controversy (Collins et al., 2018). Mechanisms within these groups are not mutually exclusive and PLP may be explained by some combination. Furthermore, researchers speculate if PLP may be a cluster of pain disorders, rather than a single disorder (Griffin & Tsao, 2014). Researchers have prioritized this mechanistic puzzle as it is essential for providing quality care to these patients.
Cerebral mechanisms consist of cortical reorganization, alterations in sensory and motor feedback, and pain memory (Flor et al., 2006). PLP is commonly correlated with reorganization and furthermore related to the self-perception of one's own body (Subedi and Grossberg, 2011).
In support of cerebral mechanisms, a 1998 study found hemispheric differences in cortical representation in traumatic amputees absent in subjects with congenital absence of limb (Montoya et al., 1998).
Spinal mechanisms relate to amputation-related nerve injury causing spinal cord hypersensitization and further reorganization of spinal cord areas formerly occupied by functioning afferent nerves (Flor et al., 2006). Connections between the proximal sections of amputated nerves can form disruptive connections with receptive spinal nerves.
Additionally, distorted neuronal activity, hyperexcitability, and central nociceptive neuron firing pattern changes may also contribute to PLP (Subedi and Grossberg, 2011).
The peripheral mechanisms involve nerve ending and dorsal root ganglion reorganization following amputations (Flor et al., 2006).
Efficacious pre-and postoperative peripheral interventions for PLP support this explanation. Patients receiving peripheral nerve interfaces before surgery have had lower rates of peripheral neuromas and PLP (Kubiak et al., 2019). Additionally, minimally invasive percutaneous peripheral nerve stimulation programs improved functionality in patients with chronic pain postamputation (Gilmore et al., 2019).
Peripheral nervous system treatment has addressed PLP functionality and pain, which validates this mechanism.
Cerebral, spinal, and peripheral PLP mechanisms have each endured scientific evaluation with no distinct victor. These are also not mutually exclusive and PLP may be a product of a combination of these mechanisms. Makin and Flor further expand upon the multifactorial nature through a review of factors beyond remapping that may come together to contribute to PLP (Makin & Flor, 2020). Broad consideration of mechanism and dynamic changes warrants a comprehensive analysis of this complex disease. Investigators continue to explore this scientific question using several specialized neuroimaging techniques.

METHODS
A literature search was conducted using the PubMed database between January 2020 and August 2021. The literature search was organized using the following keywords/keyword combinations: "phantom limb pain and diffusion tensor imaging (DTI)," "phantom limb pain mechanism," "phantom limb pain and electroencephalography (EEG)," "phantom limb pain and functional magnetic resonance imaging (fMRI)," "phantom limb pain and amputation," "phantom pain," "phantom limb pain and mirror therapy," "phantom limb pain and magnetoencephalography (MEG)," "phantom limb pain and therapeutics," "diffusion tensor imaging (DTI)," "electroencephalography (EEG)," "magnetoencephalography (MEG)," "functional magnetic resonance imaging (fMRI)." The articles generated from the search were then screened and additional articles referenced by the searched articles were also utilized. Articles were selected based on the inclusion of amputees with PLP or phantom sensations along with the utilization of DTI, fMRI, EEG, or MEG to investigate mechanism or response to therapy.

Diffusion tensor imaging
DTI is a variant of conventional MRI that has become a standard tool in researching PLP. As a general MRI principle, tissue microstructure determines water diffusion, which translates into an image.
Anisotropy describes water diffusion that is directionally dependent while isotropy describes unrestricted water diffusion; white matter is more anisotropic than gray matter, while cerebrospinal fluid is isotropic (Hagmann et al., 2006;Pierpaoli et al., 1996). DTI capitalizes on white matter tracts to assess structural integrity and connectivity (Bandettini, 2009). In PLP, DTI has become the most common tool for evaluating anatomical changes.
Important DTI scalars include axial diffusivity (AD), radial diffusivity (RD), mean diffusivity (MD), and fractional anisotropy (FA). AD and RD characterize rates of diffusion in principal and perpendicular directions, respectively, while MD is the net displacement of water molecules (Feldman et al., 2010). FA is a ratio that describes the degree of anisotropic diffusion (Feldman et al., 2010). These scalars allow DTI to interpret structural changes within the brain. A 2019 study employed DTI to determine a connection between PLP and white matter changes. Interestingly, these researchers found symmetrically Utilizing tractography in the premotor cortices, they also found altered interhemispheric fibers (Jiang et al., 2015). reports' comparability without a validated method. The review concludes that the circular method has better repeatability, while the freehand method has less variation; these characteristics may be advantageous for studying distinct aspects of the brain (Hakulinen et al., 2012).
Furthermore, Soares et al. (2013) address the technical components of DTI interpretation at each stage of data collection and propose conformity that may serve to reduce variability among researchers.
In summary, DTI exploits water diffusion due to tissue microstructures to reveal critical structural changes due to PLP. As depicted in Figure 1, analysis of these structural changes can contribute to studying cerebral mechanisms of PLP. Technical aspects limit this imaging technique and may compromise data collection and interpretation. DTI should continue to be used to characterize how particular structural Response to mental imagery therapy Reduced cortical reorganization, which correlated with reduction in pain intensity Note: Study sample size reflects amputees with phantom limb pain unless otherwise noted. a Amputees with "painless" phantom sensations. b Compared 20 PLP amputees, 20 non-PLP amputees, and 20 controls.
in this review (EEG and MEG), fMRI has more substantial spatial resolution but lower temporal resolution (Meyer-Lindenberg, 2010).  (Foell et al., 2014). In a case report about using chronic motor cortex stimulation to treat PLP, researchers used fMRI for precise surgical electrode placement and monitored the patient's response to therapy (Roux et al., 2008). A study of 13 upper limb amputees utilized fMRI that showed reduced cortical reorganization in PLP patients following mental imagery therapy by contrasting pretraining diffuse cortical activation upon motor tasks, such as a lip purse, with post-training isolated lip area activation. This 6-week training additionally correlated reductions in pain intensity with a decrease in cortical reorganization (MacIver et al., 2008). fMRI applications and key findings are summarized in Table 2.
Logothetis explains that fMRI interpretation requires caution as this neuroimaging technique may not make key distinctions, such as topdown versus bottom-up, excitation versus inhibition, or regional differences (Logothetis, 2008). Another recent review of 10 fMRI investigations from 2001 to 2015 found that this imaging technique did not comprehensively support maladaptive brain plasticity, including the relationship between pain intensity and reorganization (Jutzeler et al., 2015). In studying fMRI during a VR movement task, Andoh et al. (2020) found that fMRI findings for PLP may vary based on methodology. These findings suggest that fMRI is inconsistent in evaluating changes due to PLP, or perhaps there are gaps in understanding of plasticity in PLP. This principle reiterates the importance of the link between PLP and neurologic adaptation and the need for collaborative techniques.
fMRI measures functional changes with strong spatial resolution but is prone to certain ambiguities in interpretation. The lack of critical distinctions may be why fMRI studies have shown mixed results in PLP research. fMRI has utility in supplementing studies of PLP therapies and interventions (Figure 1).

Electroencephalography
EEG is a temporal-functional imaging technique that is useful in PLP research. EEG interprets electrical flow across membranes as neurons depolarize. It is distinct from the other imaging techniques discussed in this review by recording real-time measurements in varying cognitive states (Bunge and Kahn, 2009 method of analyzing cortical reorganization due to advantageous temporal and spatial resolution (Wiech et al., 2004). Other researchers have challenged EEG and MEG for truly assessing signal sources, and suggest employing fMRI as a tertiary, complementary component for signal localization (Bunge and Kahn, 2009;Cottereau et al., 2015). EEG, along with MEG, has higher temporal resolution but lower spatial resolution than fMRI (Meyer-Lindenberg, 2010). The unique ability to perform EEG simultaneously with fMRI further distinguishes this tool as a select method for capturing nonrepeatable events (Bandettini, 2009).
A case report of a subject with a congenitally absent limb found EEG signatures during attempted movements of the phantom limb to be similar to a cohort of healthy volunteers (Walsh et al., 2015). Another study analyzed a cohort of 22 right-hand amputees via EEG showed distinct global and local network changes in alpha and beta bands (Lyu et al., 2016). An investigation of the connection between pain catastrophizing and PLP using EEG showed that these patients had an increased response at the N/P135 dipole of the affected side, suggesting that attention to stimuli may be associated with PLP (Vase et al., 2012).
Mirror therapy has been studied as a potential PLP treatment, but recent developments in VR have enabled inventive therapeutic techniques. One such VR investigation utilized EEG and observed PLP alleviation and alpha wave coherence during stimulation of referred sensation areas (Osumi et al., 2020). EEG presents a safe and practical way to monitor the forefront of therapeutic techniques for PLP. It allows researchers to gather robust functional change data during therapies.
EEG applications and key findings are summarized in Table 3.
While it has many applications and strengths, EEG is limited by lower spatial resolution than fMRI (Meyer-Lindenberg, 2010 to interpret, and this tool should be one of many used in conjunction (Jackson & Bolger, 2014). EEG alone therefore may not provide adequate information about functional changes due to PLP.
EEG is one of the two main techniques for identifying functional changes with temporal resolution. It measures changes in electrical potential perpendicular to the direction of neuronal signal propagation.
While EEG is limited by a lack of spatial resolution, its inherent design allows it to be easily used alongside other tools to provide comprehensive results. EEG is a practical way to monitor functional changes while developing PLP therapies (Figure 1).

Magnetoencephalography
MEG is another temporal-functional imaging technique used in PLP research. It functions by measuring the small magnetic fields created by electrical currents involved in neuronal signaling. The measured magnetic dipole is 90 • off phase with the electrical one. The electrical and magnetic fields detected by EEG and MEG are generated by extracellular and intracellular currents, respectively (Singh, 2014). Both of these measured phenomena occur at directions perpendicular to that of neuronal signal propagation. Because it assesses magnetic activity at and parallel to the brain's surface, MEG is limited to the analysis of superficial sulci pyramidal cells (Bunge & Kahn, 2009). As mentioned before, EEG and MEG share a common caveat as they both have difficulty localizing signal sources (Bunge & Kahn, 2009). MEG also has a higher temporal resolution but lower spatial resolution compared to fMRI (Meyer-Lindenberg, 2010). ulation was stopped and associated these regions of the brain with pain relief. Their results demonstrate that MEG is useful for identifying response to therapy and potential surgical targets for pain relief . Another investigation of brain-machine interface training integrated MEG reading with a robotic hand. Interestingly, they found this training to intensify pain when used with the  (Yanagisawa et al., 2016). MEG applications and key findings are summarized in Table 4.
Despite its usefulness and safety, MEG has sensitivity to artifacts.
Ray et al. addressed the challenge of deep brain stimulation artifact when using MEG by focusing on the occipital lobe following a visual stimulus (Ray et al., 2009). The study shows that this tool can provide relevant information if researchers account for its limitations.
MEG is another primary technique for identifying functional changes with temporal resolution. In contrast to EEG, MEG detects magnetic activity parallel to the brain surface. This imaging modality is mostly limited by potential artifacts, which an adapted approach may control. MEG has promising future use for studying robotic and interventional therapy in PLP research (Figure 1).

DISCUSSION
Neuroimaging has proven paramount in the study of PLP (Figure 1). DTI readily outlines structural changes and has potential for surgical applications but is frequently cited for technical limitations, such as subject motion and resolution. Additionally, DTI has often been criticized for variation in measuring technique and data interpretation. fMRI captures functional changes with spatial resolution in various PLP therapies, but cannot make critical neurologic distinctions, which limits data interpretation without behavioral or structural correlation. EEG and MEG are notable for identifying functional changes with a strong temporal resolution and are differentiated by perpendicular electric and parallel magnetic activity, respectively. EEG is significant for spatial limitations, while both EEG and MEG are limited by artifact. Overall, all four imaging techniques provide unique perspectives that have shaped the modern understanding of PLP.
Accessibility and practicality are common barriers that limit PLP neuroimaging. The study scale is often resource dependent, which has restricted how much imaging data can be collected. Limited reproducibility of neuroimaging findings may also hinder the analysis of PLP in certain cases. Consistent techniques and collaboration may alleviate the burden on groups studying PLP. Additionally, the automation of imaging analysis using artificial intelligence and machine learning algorithms may generate uniformity among data interpretation (Hu et al., 2019;Vieira et al., 2017). These advancements enable the synthesis of data sets to help map neural changes. Robust data collection illustrates the key intersection of imaging and analytical technology, especially in the context of clinical disease. As this field evolves, researchers will continue to utilize neuroimaging aiming to provide fundamental insight into PLP's pathogenesis and treatment.

FUNDING INFORMATION
None.