Functional Connectivity in Adult Brain Tumor Patients: A Systematic Review

Rationale

Using tools such as positron emission tomography and magnetic resonance imaging (MRI), the brain can be examined as a series of temporally coherent networks, with both adjacent and distal regions functioning in concert as the brain works to maintain stability or accomplish tasks (Damoiseaux and Greicius, 2009; Fox and Raichle, 2007). Efforts have been made to map these networks in healthy individuals over the last decade, but many questions persist regarding each network's existence, structure, and function within clinical populations. In the United States, brain and central nervous system (CNS) tumors affect 22.36 out of every 100,000 people (Ostrom et al., 2016). Brain tumor treatments over recent decades have included surgical resection, radiation therapy, and chemotherapy. Such developments alongside improvements in diagnostic tools have led to an increased likelihood of survival and therefore a larger population of long-term survivors (Ostrom et al., 2016). That tumors may vary by location, type, World Health Organization (WHO) grade, age of onset, and other variables creates challenges with respect to analyses, but identifying trends in dysfunction both within and across groups may inform treatment, prognosis, and rehabilitative intervention (Lannering et al., 1990; Louis et al., 2016).

Prior neuroimaging research has demonstrated a wide range of neural changes following diagnosis of a brain tumor (BT). Structural neuroimaging has demonstrated impacts, including damage at the site of the tumor itself, the effects of a resection, and more downstream impacts (Baris et al., 2016). Researchers have historically used blood oxygen level-dependent (BOLD) percent signal change, an indirect measure of changes in neuronal activity based on the inference that oxygenation of brain regions correlates with the presence of neurological activation, to assess brain activity in healthy and clinical groups (Raichle, 1998). Studies have used BOLD to demonstrate distinct patterns of activation during rest and various cognitive tasks in BT patients compared with healthy individuals both immediately following diagnosis and long term (King et al., 2015; Petrella et al., 2006; Robinson et al., 2014). Differences are not only seen within the regions physically affected by the tumor but are often present in distal regions of the brain as well, suggesting that tumors impact broader systems. Therefore, functional connectivity (FC) approaches provide a critical next step to understanding how a BT affects spatially distributed but temporally correlated network activity throughout the brain.

FC is a measure of statistical temporal dependencies among multiple neurophysiological events (Friston, 2011). More broadly, it is the measurement of how regions of the brain coactivate together over a given time frame, either during resting state or when performing a task (Biswal et al., 1997). This coactivation is interpreted as these regions working in concert toward a goal during a task or spontaneous activations when at rest. A number of functionally connected networks are routinely observed across studies of healthy adults (Table 1). Proper function of these networks is believed to represent optimal architecture for information exchange across the brain (Stam, 2014). Patterns of FC in various networks have been shown to relate to cognitive task ability and predict task-response properties of brain regions (Fox and Greicius, 2010), and changes in both structural and FC are observed throughout childhood and across the life span (Betzel et al., 2014).

FC is often disrupted in individuals with neurological or psychological disorders, and these changes frequently correlate with and may underlie functional impairments. For example, reduced FC strength in the default mode network (DMN) is observed in autism spectrum disorders, and persistent DMN FC during goal-directed activity, a time when it is expected to be decreased, has been observed in disorders related to attention deficits such as attention-deficit/hyperactivity disorder and schizophrenia (Broyd et al., 2009). Neurological insult such as traumatic brain injury (TBI) has also been shown to result in altered connectivity patterns compared with healthy controls (HCs) (Nathan et al., 2015; Shumskaya et al., 2012), and relationships between attention task performance and resting-state network connectivity have also been observed (Shumskaya et al., 2017). Patterns of alterations vary widely and may include decreased or increased connectivity within a network or topographical changes: physical shifts in location of network regions due to a lesion.

In addition, some clinical populations demonstrate abnormal patterns where stronger FC is observed between different networks than within a network (Broyd et al., 2009). Aberrant resting-state FC is theorized to be a biomarker for various outcomes (Castellanos et al., 2013). DMN connectivity abnormalities have been established as a possible precursor to cognitive or behavioral changes in Alzheimer's disease and may precede amyloid deposition (Krajcovicova et al., 2014; Sheline et al., 2010). Recent research has also utilized FC patterns to successfully predict schizophrenic patients' responses to antipsychotic medications (Sarpal et al., 2016) and has theorized that patterns of FC in the DMN, salience network, and executive control networks (ECNs) may be used as biomarkers for susceptibility to nicotine addiction (Fedota and Stein, 2015). Consolidation of research on BT patients' FC may ultimately similarly be used as predictors for short- or long-term cognitive and behavioral outcomes.

Two commonly utilized methods of assessing FC are seed-based correlation analysis (SBCA) and blind source separation, such as independent component analysis (ICA). In general, SBCA requires an a priori hypothesis about what networks will be present during the evaluated time frame, as the researcher must use software to create a seed of voxel in the brain. Seeds may be chosen based on prior BOLD analyses indicating activity in the region or by an atlas. It is against the activity of this seed region that activity throughout the rest of the brain is compared to identify other coactive regions, that is, parts of the seed's network.

In contrast, ICA can be used to explore patterns of FC when an a priori model is not available (Calhoun et al., 2003). Given BOLD (spatial) data over the course of many time points (temporal data) for the entire brain, programs use algorithms to find a linear representation of the data in which the components are as statistically independent as possible. These maximally independent components are groups of voxel clusters throughout the brain that follow similar time courses, and each of these groups of clusters can be considered a potential network.

A recent review evaluated the research on effects of various lesion types on the connectome, the overall complex network of which the brain is comprised as assessed by graph theory (Aerts et al., 2016). Graph theory is the study of mathematical structures that model relationships (edges) between objects (nodes), and brain connectomics specifically assesses connectivity of regions across the brain. It is used to characterize properties such as patterns of segregation, integration, and centrality. These mathematical evaluations of network topology are distinct from physical topography and are conceptually distinct from ICA- or SBCA-based assessment of individual networks in the brain, often instead focusing on quantity and length of connections over FC strength. Clustering algorithms can also be used to assess these topological values (Lee et al., 2013). Differences in the types of results in topological studies from ICA and SBCA studies, as well as a recent review's inclusion of graph theory findings in BT samples (Aerts et al., 2016), prompted the exclusion of these connectomics studies from the current review. However, the complementary and distinct evaluations of how BTs impact the global organization of networks provided by topology are invaluable to a broader understanding of both short-term and long-term outcomes.

The current science suggests that the brain must be conceptualized as numerous interwoven networks (Biswal et al., 2010; Damoiseaux and Greicius, 2009; Lv et al., 2015) and that BTs and their treatments may have both proximal and distal effects on invaluable functional networks (King et al., 2015; Robinson et al., 2014). Amassing and analyzing results from across studies incorporating both seed-based and blind source separation analyses to evaluate FC in BT patients will help provide an understanding of how individual networks present in this disease state. It will allow for comparison across methodologies and illuminate gaps and inconsistencies in the present corpus of knowledge that will need to be addressed in the future. Ideally, this understanding will ultimately allow for extrapolation of expected neurological outcomes as more BT patients survive for many years past their illness and treatment.

Objectives

This review aims to summarize and conceptualize SBCA and ICA findings from resting-state and task-based functional MRI (fMRI) FC assessments of adult BT patients to better understand the immediate and longitudinal effects of a tumor on various functional networks of the human brain. Healthy control or other clinical comparison group results are described when available. Relationships between connectivity and cognitive and behavioral functioning are examined when data are available. Studies evaluating FC both pre- and postsurgery or adjuvant therapy treatments are included and described independently, as are both cross-sectional and longitudinal studies. Limitations across studies are also addressed, and recommendations for future research are made.

Eligibility criteria

Studies of multiple participants, including a group of adult BT patients or survivors using a seed-based connectivity analysis method or blind source separation method and published in English, were evaluated. Studies of children were not included given expected differences in FC due to known patterns in brain development (Betzel et al., 2014) as well as significant differences in tumor types compared with those found in adults (Ostrom et al., 2016). Participants could be at any stage of treatment. Primary outcome measures were required to include evaluation of at least one network of interest.

Information sources and search strategy

The search was conducted in the PubMed electronic database on April 20, 2017, April 26, 2017, May 8, 2017, and May 17, 2017. Search terms included brain tumor OR glioma AND functional connectivity, independent component analysis, ICA, psychophysiological interaction, OR PPI.

Identification of relevant studies and data extraction

The eligibility assessment was performed in a standardized manner by one of the authors (M.E.F.). After deleting all duplicate articles, titles and abstracts of all articles were reviewed for suitability for inclusion, and those that were potentially eligible were reviewed in full. Therefore, 25 empirical research studies were included in this review. Information was extracted from each included study on (1) characteristics of participants (including age and sex); (2) characteristics of tumor (including type and location); (3) treatment status (pre- or post-treatment and treatment type if applicable); (4) imaging characteristics (including measure of FC, resting state, or task conducted in scanner); (5) specific networks studied; (6) brain/behavior correlates; and (7) quantitative results (see Table 3 for study details).

Risk of bias and analysis plan

Study quality was recorded across nine domains, which were chosen based on prior reviews of BT patients and survivors (Ailion et al., 2017; Wolfe et al., 2012) and recommendations for systematic reviews (Khan et al., 2003). Domains included sample size, use and size of comparison group, range of participant ages, tumor type, tumor location, treatment and disease stage, methodology, use of cognitive or behavioral correlates, and inclusion of quantitative FC data. (See Table 2 for details of quality assessment.) A systematic, narrative approach was used in this review.

Study selection

Studies were excluded if the sample was only healthy participants (5 excluded); the clinical population studied was not BT patients or survivors (11); the study used electroencephalography, magnetoencephalography (MEG), or electrocorticography and not fMRI (16); FC was studied via electrostimulation (13); ICA was used for tumor identification or grading as opposed to assessing FC (15); participants were children (<18 years) at the time of diagnosis or study (1); ICA was used for another element of the study, such as evaluating white matter or cerebral blood flow (11); the article was a review of a technique or software (6); ICA did not stand for independent component analysis, and content was therefore unrelated to the desired topic (9); the study evaluated whole-brain network topology as opposed to individual network-specific FC (4); the result was a conference abstract instead of an article (1); only case studies were presented (1); or the article was not in English (1).

Study characteristics

Twenty-five empirical studies assessing FC in BT patients were therefore included in this review (Table 3). All were written in English and published between 2008 and 2017. Studies evaluated the strength and statistical significance of FC within at least one empirically established network but used a variety of analysis methods, broadly grouped into blind source separation (ICA; 13 studies) and SBCA (16 studies); four studies incorporated both methodologies. All studies used adult participants (18 years and older in all studies that reported age ranges) and overall included 524 BT patients or survivors and 194 HCs. The majority of studies solely evaluated patients shortly after diagnosis and before treatment. Two studies included some participants who were undergoing a recurrence and had already received treatment for an earlier tumor, and one study assessed FC post-treatment for their patients for whom surgery was required (4 of 5). Eighteen studies were conducted on resting-state data only, six were conducted on task data only, and one included both resting-state and task data. Eight studies described a cognitive or behavioral correlate, only one of which provided means and standard deviations of cognitive test scores. Twelve studies indicated that preprocessing steps included normalization or coregistration to a standard template such as Montreal Neurological Institute or Talairach.

No study met all strong criteria. BT type and location varied widely across and within studies (Table 3). Only four studies included a single specific tumor type (e.g., astrocytoma). One study was not limited to a single tumor type but included only low-grade gliomas. The remaining studies had no limitations on tumor types included or included various gliomas, a broad category that does not restrict by grade. Nine studies included patients with tumors in a specific region or lobe of the brain, seven had broad restrictions (e.g., left hemisphere only) or included multiple regions, and nine had no restrictions on tumor location.

Means and standard deviations of z-transformed Pearson correlations measuring internetwork FC were reported in 10 studies. Three studies provided Dice coefficients as a measure of overlap between templates and the networks identified by the study or between networks mapped across task and rest. Two studies provided p values of comparison between BT and HC groups' FC strength but no descriptive statistics by group. Two studies provided p values of comparisons of FC between hemispheres. One study included peak z-scores from a power spectrum analysis. The remaining six studies did not provide numerical data specific to FC, instead provided qualitative descriptions or visualizations such as scatter plots and connectivity matrices without values.

Given that studies analyzed were not randomized controlled trials, risk of bias as per The Cochrane Handbook was not evaluated (Higgins and Altman, 2011).

Results of individual studies

Results of individual studies presented by network of interest are described below and in Table 4.

Summary of evidence

This review aimed to summarize the effects of adult BTs on FC throughout the brain as characterized by seed-based connectivity and blind source separation analyses. The majority of studies assessed focused on newly diagnosed patients before initiation of treatment, and the DMN and SMN were most commonly analyzed. No study met all strong criteria (Tables 2 and 3). Findings were commonly presented qualitatively and were only accompanied by numerical measures of a BT sample's connectivity strength (such as Fisher z-transformed Pearson correlations indicating synchronized activity between or among network regions) in nine studies, prohibiting evaluation of overall effect sizes. Future research will benefit from using large, homogeneous samples and providing quantitative results that will allow for interpretation across studies.

From the extant research, we broadly conclude that BTs consistently correlate with changes to FC of numerous networks, but only rarely is expected network connectivity disrupted to the extent that the network cannot be identified through methods of ICA or SBCA. In general, networks showed decreased FC during times of expected connectivity in pretreatment BT populations. For example, during rest, an expected time of greater DMN activity, DMN FC was weaker, and during tasks, SMN and LN FC were weaker. These disruptions may involve a shift in the location of a network node, change in size of region or network, or decrease in FC values within the network.

If tumors were in or near the network of interest, changes in location or size of network regions may be attributed to mass effect such that the tumor physically shifted regions of the brain (Harris et al., 2014; Niu et al., 2014). Other possible mechanisms of impact include changes in blood flow to the tumor that caused reduced perfusion to other nearby areas (Zhang et al., 2016) or physical infiltration of tumor tissue into surrounding areas (Maesawa et al., 2015; Mallela et al., 2016). Unfortunately, most studies in this review recruited participants with tumors of various locations. Future studies should prioritize a homogenous tumor region in each analysis.

When studies noted tumor location, it was clear that some tumors had wholly distal effects. Reflective of early MEG studies assessing FC with BTs (Bartolomei et al., 2006a,b), a number of current fMRI studies found changes in connectivity strength occurring in the hemisphere contralateral to the tumor (Maesawa et al., 2015; Mallela et al., 2016; Wang et al., 2013). Distal effects were also observed as cerebellar tumors impacted cerebral networks (Ghumman et al., 2016), a finding that has been echoed in a study of long-term pediatric BT survivors as well (Chen et al., 2016). It was concluded that mass effect alone could not have had such significant downstream effects; studies instead implicated the aforementioned potential changes to blood flow or white matter connections due to tumor infiltration (Mallela et al., 2016). Distal changes also suggested broader plasticity of the brain. In particular, slow-growing LGGs were implicated as a cause of neural reorganization (Esposito et al., 2012) by way of sophisticated neuronal mechanisms that allow for neurite outgrowth in adults (Bavelier et al., 2010). LGG patients are also less likely to demonstrate behavioral deficits before tumor identification and have better postoperative functional prognoses than other BT patients (Desmurget et al., 2007). This suggests that these reorganizations are elements of compensation that correlate with the degree of benefits of surgical resection. Additional longitudinal studies will help determine more details of the mechanisms of disruption, as researchers may be able to better track shifts in brain structure and function over time following treatment.

Tumors in the left hemisphere generally had a greater impact on FC strength than those in the right (Ghumman et al., 2015; Harris et al., 2014), as has been observed in MEG studies (Bartolomei et al., 2006a). This may sometimes be attributed to left laterality of a network and the local physical changes caused by BTs. For example, in patients with a left hemisphere BT, the entire left-sided LN appeared to translate across the brain to the right hemisphere (Wang et al., 2013). However, this trend was also observed in other less lateralized networks, such as the DMN (Ghumman et al., 2016; Zhang et al., 2016). This suggests that more consideration of the role of asymmetry should be made, given a potentially greater vulnerability of networks within the left hemisphere (Jalili, 2014). Parietal lobe tumors also appeared to have a larger effect on DMN FC than frontal lobe tumors in one study (Harris et al., 2014), indicating a need for further investigation in that domain.

Greater effects were also seen in the tumor-ipsilateral hemisphere compared with the contralateral hemisphere regardless of whether a tumor was located in the right or left hemisphere (Agarwal et al., 2016a,b, 2017; Maesawa et al., 2015), although three of the four studies including this information utilized overlapping samples. Neurovascular uncoupling (NVU), a reduction in BOLD activation despite neuronal activity (Holodny et al., 2000; Ulmer et al., 2003), has been implicated as a basis for a lack of BOLD signal and therefore lack of implied FC near tumors (Agarwal et al., 2016a; Niu et al., 2014; Park et al., 2016). BOLD is a measure of blood flow that is believed to be a consequence of neuronal activity instead of a direct measure, but it may be that tumors cause a cascade of events that alter cerebral blood flow and volume. Therefore, a lack of network identification within the constraints of ICA or relatively weaker FC in BT samples may be a false negative with respect to neuronal activity and coordination. Greater uncoupling near the tumor may therefore explain this result of inferred decreases in tumor-ipsilateral FC strength, while some studies in this review made reference to effects of NVU as a possible explanation of results or limitation of interpretability, over two-thirds did not address it. Breath-hold cerebrovascular reactivity mapping has been proposed as a method of assessing NVU and may be important to incorporate into future studies (Zaca et al., 2014).

Assessments of the SMN showed a unique trend not reported elsewhere. Although BT-HC differences were not always observed (Kokkonen et al., 2009; Otten et al., 2012), those studies that found discrepancies in SMN FC observed a decrease in interhemispheric FC strength in their BT patients (Mallela et al., 2016; Niu et al., 2014; Posse et al., 2013). This pattern was rarely described across other networks, although this may be due to the understanding of interhemispheric function as it relates to motor tasks. However, one study observed a greater difference in intrahemispheric than interhemispheric FC in the DMN in BT patients compared with HCs (Zhang et al., 2016). This lack of interhemispheric SMN interaction regardless of the tumor's location may be a clue toward understanding the common long-term outcome of motor deficits in BT survivors (Pietilä et al., 2012), although correlations between interhemispheric FC and motor weakness were inconsistent in the studies reviewed here. Assessments of other populations have indicated that motor tasks involve more interhemispheric connectivity compared with cognitive tasks requiring more intrahemispheric connectivity (Gazzaniga, 2000; Wang and Liu, 2014), suggesting that the changes seen in BT patients may be particularly detrimental to their motor functioning. This is further confirmed by the motor network FC's association with functional skills (Otten et al., 2012; Vassal et al., 2017).

Increases in FC in BT populations were often found among unexpected pairs or groups of regions, such as stronger between-network instead of within-network connections (Kinno et al., 2014). These alterations in coordination across the brain are implicated in various neurological diseases (Shumskaya et al., 2017; Stam, 2014; Zhou et al., 2012). A study by Kinno et al. (2014) supported the idea of neurological and therefore behavioral disruption as agrammatical BT patients in their study showed significant amounts of this unexpected connectivity, described as “chaos.” If these changes are a consequence of the BT, they may reflect possible compensation via reorganization of networks as has been proposed in the context of neural hyperactivity following insult (Maruishi et al., 2007; Scheibel et al., 2007), although other literature describes hyperconnectivity as recruitment of a latent resource that does not ultimately aid in cognitive function (Hillary et al., 2015; Medaglia et al., 2012). While increases in connectivity may be functionally adaptive in the short term, the resulting elevated metabolic stress may be damaging to the primary network of impact as well as others throughout the brain (Hillary and Grafman, 2017). Tracking long-term survivors and their cognitive abilities may provide more answers as to the mechanisms and outcomes of these changes, and interactions between networks should be considered.

When researchers differentiated between high- and low-grade tumor groups in the included studies, results varied, although the general trend was for higher grade tumors to correlate with decreased FC strength (Harris et al., 2014; Mallela et al., 2016; Zhang et al., 2016). More malignant (higher grade) tumors are thought to be more infiltrative into the surrounding tissue and therefore more damaging to nearby structures (Harris et al., 2014). On the contrary, network connectomics studies of BT patients and survivors have indicated greater change with low-grade tumors and suggest that slower growth provides time for reorganization of networks throughout the brain (Aerts et al., 2016; Derks et al., 2017).

Regarding effects of BTs over time, post-treatment assessments were found to capture an immediate decrease but ultimate increase or recovery of FC strength over an extended period of time (Hart et al., 2017; Otten et al., 2012; Vassal et al., 2017). It appears that there is some degree of immediate change to FC following resection, particularly when the tumor is clearly within the network being assessed, but months of recovery may lead to return to pretreatment levels of FC strength. A study of long-term survivors of childhood BTs over 7 years past diagnosis in fact observed greater DMN, ECN, and SN FC strength in BT survivors than HCs, indicating a potential continued pattern of FC increase following recovery, although the utility of this hyperconnectivity remains debated (Chen et al., 2016; Hillary et al., 2015).

Those experiencing a tumor recurrence may show a distinct pattern of FC; when newly diagnosed treatment-naive patients were compared with individuals experiencing a tumor recurrence, the latter group demonstrated weaker DMN FC (Harris et al., 2014). An additional study included in the review utilized both treatment-naive patients and those experiencing a recurrence but did not distinguish between the groups in their results or discussion (Lang et al., 2017), a remarkable concern; patients experiencing a tumor recurrence should at least be considered in a different context than new patients, given the succession and severity of their condition. Furthermore, no mention of shunt placement was made in any longer term studies, a factor that could cause physical shifting of networks, and it remained unclear which and how many participants underwent other treatments such as radiation or chemotherapy when the aforementioned studies made reference to it.

ICA and SBCA studies rarely produced contradictory findings, instead investigating different questions. ICA indicated networks' presence and overlap or lack thereof between groups or between participants and a template. While SBCA could produce data allowing for greater description (e.g., networks observed within but not between hemispheres), it allowed for more directed analyses, inquiring as to a relationship between a given point and expected areas of functional connectedness. However, both techniques also had their limitations.

Within studies using ICA, the number of components selected varied widely, with some studies generating up to four sets of independent components to find the best network based on the literature or BOLD data. Furthermore, selection of a component to represent a given network may be based on the literature surrounding healthy populations, disregarding the anatomical changes to the brain caused by tumors. ICA inherently accounts for potential anatomical shifts in network regions, but this in turn presents the challenge of determining whether a network truly matches with the expected network from HC literature or a study's own control group.

On the contrary, SBCA requires the selection of at least one seed, and due to the current dearth of consistent, confirmatory literature on BT populations' brain activity and connectivity, these seeds were also generally selected based on HC literature. Even when templates were warped to individuals' structural scans, the anatomical changes caused by a BT prevented researchers from knowing whether their seed region encapsulated the functional region desired. However, some studies were able to draw conclusions about their regions of interest with greater confidence by selecting seeds based on preliminary ICA findings (Ghumman et al., 2016; Maesawa et al., 2015) or BOLD data (Agarwal et al., 2016a, 2017; Briganti et al., 2012; Lang et al., 2017; Niu et al., 2014). With all that has been determined about anatomical effects of BTs, it will be important for subsequent studies to utilize these confirmatory techniques. They may also be supplemented by work in graph theory.

Findings for each network were generally consistent across both resting-state and task-based scans when available. The exception was within LN studies; while most resting-state and task findings suggested that BT patients demonstrated similar FC to HCs, one ICA study used a naming task in the scanner and found that there was a potential right lateralization of the LN (Wang et al., 2013) and an SBCA study found decreased LN FC in BT patients versus controls. Studies using alternative imaging methods have demonstrated similar patterns of LN function and reorganization (Desmurget et al., 2007). However, this review's remaining results suggest that clinical use of FC in resting-state fMRI for preoperative planning may be as effective as task-based fMRI while minimizing demands on a potentially cognitively impaired patient. Task-based BOLD data may in fact be an underestimate of recruitment of a brain region due to a patient's inability to perform a task, further supporting the use of resting-state FC to determine eloquent areas (Castellanos et al., 2013; Tanaka and Stufflebeam, 2016). It may therefore be concluded that while resting-state fMRI is a useful tool in preoperative planning, task-based fMRI remains a necessary supplement for evaluating the LN.

FC has been an area of interest across neuroscience research due to its clinical correlates such as with Alzheimer's disease and cognitive effects of TBI (Broyd et al., 2009; Nathan et al., 2015; Raichle, 2015; Shumskaya et al., 2017; Stevens et al., 2012). While significant research has been done on BT patients' and survivors' abilities, fewer than one-third of the studies reviewed in this article presented any data on functional or cognitive abilities alongside their imaging results. Most studies, including brain/behavior correlates, used descriptors of individual participants such as “right hand weakness” or “gait instability” without providing quantifiable data, although Vassal and associates (2017) reported using a standardized metric for measurement of muscle testing. Only four studies used “gold standard” neuropsychological measures with normative data, which included the Lafayette Grooved Pegboard and tasks from the Wechsler Adult Intelligence Scale-Third Edition, Wechsler Memory Scale-Revised, and National Institutes of Health (NIH) Toolbox. Others included language or motor tasks that were unstandardized. However, results, including cognitive or behavioral correlates, were aligned with prior findings from HCs or other clinical groups; greater disruption or chaos tended to correlate with greater deficits (Kinno et al., 2014; Otten et al., 2012). Once a deeper understanding of these brain/behavior relationships is gained, it may be appropriate to consider FC changes a biomarker for functional outcomes of a BT. Furthermore, given the associations between psychological disorders and FC changes, future studies should include data on these disorders in BT samples. Ideally, future researchers will aim to delineate among effects of BTs, treatments, and comorbid disorders on phenotypes of FC.

Limitations

The most significant limitation of reviewing such a rare clinical population is the small sample size of most studies. This limits researchers' ability to consider the specific effects of tumor location, tumor pathology and grade, age at diagnosis, and other demographic variables. The few studies including long-term survivors also did not have the power to account for differences in earlier treatments or time since diagnosis. Tumor location variability greatly impacted on interpretability of results across this review. While seven studies narrowed their BT participant pool to those with tumors in specific locations, eight had only broad limitations, and nine included all cortical tumors or noted no exclusionary criteria by location at all. In the latter categories, this meant it was possible for a given participant's tumor to overlap with a network of interest, to be adjacent to it, or to be noticeably distal from it. Each of these scenarios would, if known, lead to distinct interpretations of findings. This was frequently described as a limitation in the Discussion section with little additional reverence paid to tumor locations' potential roles in results. Notably, similar challenges are encountered in analyses of FC following TBI where the location of insult is not only usually heterogeneous across participants but sometimes indeterminable from neuroimaging as well (Hayes et al., 2016).

The breadth of tumor types and grades included in a given study further limited interpretability. A sample, including tumors of varying grades, may obscure genuine results; for example, less malignant tumors will grow more slowly, potentially allowing for reorganization of brain networks before being assessed for these studies. This could result in insignificant group-wide findings due to inconsistencies among participants by grade. However, it also suggests that in the instances where conclusions could be drawn, results were that much more likely to be a significant outcome in this broad population.

Data interpretation is further limited by the possibility of impacts of study procedures. For example, 12 studies indicated that preprocessing steps included normalization or coregistration to a standard template. Lesions in the brain may impact the location of standard anatomical landmarks utilized in the normalization process, which in turn may lead to inaccurate warping and data that do not reflect patients' patterns of connectivity (Crinion et al., 2007). There also remains the concern that findings may be compromised by magnetic field inhomogeneities caused by the lesion as well as shunts or physical ramifications of surgical intervention in postsurgical analyses.

Neuropsychological measures with validated normative data were also rarely used in the few cases where authors made efforts to demonstrate brain/behavior relationships. Data on post-treatment FC and how it relates to long-term cognitive and adaptive functioning will be invaluable to researchers and clinicians, but to date, only one study identified conducted express pre- and post-treatment assessments (Vassal et al., 2017). Repeated evaluations across patients' treatment and life span will grant researchers and clinicians a clearer understanding of the impacts of BTs on FC strength in various networks so that they may provide more confident prognoses. Furthermore, understanding the trajectory of insult and recovery may inform timing and methods of intervention as well as treatment plans. Common complications such as seizures or hormone disorders should also be included, and evaluating the effects of treatments such as chemotherapy and radiation will be another invaluable step in this field of research. Post-treatment and longitudinal research must differentiate between survivors who have only undergone surgical resection and those who have also been treated with chemotherapy and radiation to parse apart treatment effects. In addition, future researchers may consider comparing BT survivors who have undergone chemotherapy with survivors of other non-CNS cancers whose treatment protocol included chemotherapy as a way to compare effects of neurotoxicity of the treatment to tumor-specific effects.