Methodological implications for measuring structural changes related to cognitive dysfunction in breast cancer survivors over time:
An investigation of voxel-based morphometry, quantitative MRI, and study design
November 28, 2016
Yasmin Mzayek
11116773
Supervisor: Gilles de Hollander
Co-assessor: Max Keuken
MSc in Brain and Cognitive Sciences
Abstract
Neuroimaging advances have allowed latent brain data to come to the forefront of pathological research. Structural brain imaging techniques using magnetic resonance imaging (MRI) have provided evidence for the underlying neural changes that contribute to cognitive dysfunction. Brain imaging data acquired through voxel-based morphometry (VBM), a conventional neuroimaging method, have conferred fundamental information about the structural brain changes associated with many disorders. This has benefitted clinical research by providing information about the progression of diseases and informing treatment options. In breast cancer survivors (BCS) in particular, VBM has been widely used for comparing anatomical brain differences among survivors and controls. However, VBM presents some drawbacks that could constrain study design and effect the reproducibility of data. Relatively newer approaches, namely quantitative MRI (qMRI) methods, can be applied in a similar setting to add information to and improve the reliability of data derived through conventional methods. qMRI can facilitate longitudinal, multi-center neuroimaging studies with consistent structural brain imaging data and lead to better comparability across studies. VBM and qMRI will be evaluated in this review, with the aim of providing recommendations for improving the study design of BCS research.
1.1 Introduction
Studies using magnetic resonance imaging (MRI) of the brain in clinical research are becoming more and more common. Different types of brain MRI techniques are being implemented in order to discover the structural and functional neural underpinnings of neuropsychological disorders. Two of these techniques will be systematically reviewed in this paper. The first is a methodological framework called voxel-based morphometry (VBM) and the second is a contrast acquisition technique termed quantitative MRI (qMRI). In general, the former gives information about the macrostructure of gray matter tissue, while the latter gives information about the microstructural features of white matter. By using such tools, researchers are taking an essential step for diagnosis, disease progression, and treatment purposes. For example, breast cancer survivors (BCS) have been shown to have negative cognitive effects due to the cancer itself as well as its treatments, namely chemotherapy. Structural brain imaging has added to these findings by revealing differences in brain tissue composition in this clinical population as compared to healthy controls. The most widely used structural neuroimaging method in these studies is VBM.
Both VBM and methods using qMRI can increase knowledge about the nature of the structural changes seen due to a disease. Each tool has its advantages and disadvantages, which will be laid out in what follows. VBM is widely used because it is an established methodological neuroimaging technique. qMRI is a relatively new contrast yielding technique that has not been used as much, especially in the context of clinical research. Given some of the drawbacks involved in conducting VBM and the limitations it imposes on study design, qMRI can be a useful complementary tool.
Currently, there is an increasing amount of research on the neurotoxic effects of cancer and its treatment on cancer survivors. This is especially important for BCS because they represent the largest population of cancer survivors (Boykoff et al., 2009). Also, BCS neuroimaging studies often employ VBM to measure structural changes making them appropriate for assessing its methodological features. However, such studies can benefit considerably by using quantitative approaches. qMRI can be used alongside, as well as in methods based on, VBM. Thus, in this review, VBM and qMRI will be evaluated with the aim of showing how careful considerations when using these tools together can improve current clinical research on cognitive problems among cancer survivors specifically in BCS neuroimaging studies.
1.2 Cancer-related and treatment-related effects on cognition
With the declining mortality rate in BCS, attention is shifting to the psychosocial impact of cancer- and treatment-related factors on survivors (Boykoff, Moieni, & Subramanian, 2009). Patients have reported cognitive problems after treatment, which may be due to the cancer itself as well as the cancer treatment(s) received by the patients. For example, adjuvant chemotherapy, a treatment referring to chemotherapy received after surgery, is received by up to 60% of high-risk BCS and is necessary in order to prevent any latent effects of the disease (de Ruiter et al., 2012). However, it can give rise to cognitive deficits. It is hypothesized that cancer- and treatment-related factors, including inflammatory response and poor DNA repair mechanisms, have negative effects on neuropsychological functioning (Ahles, Root, & Ryan, 2012). For example, Pomykala et al. (2013) showed significant
associations between brain metabolism, pro-inflammatory cytokines, and memory complaints that were related to dysfunction in frontal and temporal brain regions in BCS who received chemotherapy
compared to a non-chemotherapy group. Additionally, increased oxidative DNA damage was associated with decreased gray matter density in BCS giving support to the possible mechanisms through which chemotherapy-related brain damage is acquired (Conroy et al., 2013). The cognitive impairment seen in breast cancer patients has been termed “chemobrain” due to the harmful outcomes attributed
specifically to chemotherapy (Simó, Rifa-Ros, Rodriguez-Fornells, & Bruna, 2013; Boykoff et al., 2009). Chemobrain is currently a central focus of research in BCS.
Learning and memory issues have been most consistently reported across studies and are significant contributing factors to patients’ psychosocial impairment (de Ruiter et al., 2012; McDonald & Saykin, 2013). Wefel and Schagen (2012) reviewed BCS literature and reported that cognitive dysfunction stemming from either cancer- or treatment-related factors has been seen in 19% to 78% of survivors. Such a broad range may be due to the heterogeneous nature of the studied population, but also to inconsistent findings resulting from limitations in study design (i.e., age of BCS, small sample sizes, testing batteries used, inclusion of control groups) (Scherling & Smith, 2013). Because of this, the effects of chemobrain can sometimes be overlooked. Cancer-associated cognitive deficits, however, may not be reflected in neuropsychological evaluations alone.
Chemobrain, even when subtle, can have a large impact on quality of life and in turn can affect survivors’ personal and professional lives, especially in a high functioning environment. Boykoff et al. (2009) showed that, based on quality of life and daily functioning assessment scores in a sample of BCS, cognitive impairment due to chemobrain can lead to many everyday challenges. It was also reported that
BCS experience difficulties in social and work environments (Von Ah, Habermann, Carpenter, and Schneider, 2013). Difficulties such as these are not usually associated with objective neuropsychological measures, but rather with depression and anxiety (Scherling & Smith, 2013; Biglia et al., 2012). Given the impact of distress resulting from this mild cognitive impairment, the mechanisms behind chemobrain need to be revealed through well-founded studies of both objective and subjective impairments seen in BCS.
Though some patients show long-term recovery, there is a subgroup of patients that still exhibit cognitive difficulties later in life (Conroy et al., 2013). The course and the extent of recovery are not yet clearly understood. It is still difficult to pinpoint the differences between BCS who do recover and those who experience cognitive problems in the long-term. Thus, validation studies are needed for the purpose of providing reliable and meaningful results that can confirm the complaints of BCS. This can lead to an improved picture of how cognitive and neural changes come about and progress and how they can be treated. One way to add to the existing knowledge base is through acquiring brain imaging data and analyzing its associations with cognitive, behavioral, and psychosocial problems.
1.3 Magnetic resonance imaging
Magnetic resonance imaging (MRI) is an imaging technique used to visualize the structural features and functional activity of the brain. This is done by generating contrast images based on the polarity of water molecules, specifically hydrogen protons (McRobbie, 2007). These images are used in healthy and patient populations for research and clinical purposes. In research settings studies using MRI can examine anatomical and physiological features of the brain. In clinical settings MRI can be used diagnostically, to look at disease progression, and to assess the effects of clinical trials.
There are several means through which MRI can be used to measure such differences. VBM is one common and conventional method used to examine brain structure. With VBM, structural brain maps are acquired from T₁-weighted images and are used to measure gray matter volume. qMRI is a more recent and promising way to acquire specific structural maps. Both weighted and specific maps are useful to provide a better understanding of the macro- and micro-structural characteristics of brain tissue. However, the kind of information and data they provide are different. Further, each has its own advantages and disadvantages.
Voxel-based morphometry
VBM is a neuroimaging method used to segment brain tissue into gray and white matter and to measure gray matter volume and density at a macroscopic level (Ashburner & Friston, 2000). It is primarily used to find local anatomical differences between brains through a voxel-wise comparison (Henley et al., 2010).
VBM application and its pitfalls
VBM allows researchers to gather information about the structure of the brain in living
organisms. This technique makes available information that would otherwise not be accessible without using ex vivo neuroimaging or histological approaches. It is usually used to identify structural changes that correlate with behavioral and cognitive impairment and to assess disease progression and
treatment. This is done with the goal of identifying biomarkers for diagnostic purposes and to examine the effects of clinical trials testing different therapies.
In addition to being noninvasive and having a well-established procedure, VBM has several advantages in its application to structural brain imaging studies. There is no bias towards one region, but rather, with VBM researchers can analyze differences throughout the entire brain through a voxel-wise analysis of gray matter (Ashburner & Friston, 2000). VBM is also useful for analyzing the macrostructure of the brain and for finding focal differences. (Davatzikos, 2004). Nevertheless, several issues limit the utility of its use.
The challenges that come with using VBM can be delineated at three different levels: imaging data acquisition, image preprocessing, and statistical analysis. For example, VBM data are not
comparable across different time points for the same participant or across participants scanned at different centers (Deoni et al., 2008). This prevents reliable use of VBM for longitudinal, multi-center studies. The way VBM is processed can have an effect on the data, leading to inconsistent findings based solely on the approach used (Radua, Canales-Rodríguez, Pomarol-Clotet, & Salvador, 2013; Peelle, Cusack, & Henson, 2012). It also does not give enough information for understanding underlying mechanisms (Draganski et al., 2011). Finally, VBM lacks sufficient differentiation power to detect the mechanisms behind the mild and diffuse brain damage seen in BCS, especially at the microstructural level. The impact of these limitations on the data and results lead to unreliable comparisons between studies and could explain the inconsistencies found across the literature in BCS studies.
Quantitative MRI
Quantitative MRI (qMRI) is a tool used for acquiring MRI contrasts. Its purpose is to give quantitative measures of specific MRI brain tissue parameters at the microscopic level (Weiskopf, Mohammadi, Lutti, & Callaghan, 2015; Draganski et al., 2011). Thus, unlike the weighted images that are used in VBM, qMRI yields images based on specific tissue features found in a voxel. Its application can improve design and reproducibility of brain imaging studies, as well as provide new information about diseases.
qMRI application and its pitfalls
qMRI is a noninvasive way to obtain structural brain images that complements and supports invasive histological and non-invasive macrostructural findings (Callaghan et al., 2014). With this technique, researchers can acquire multiple tissue parameter maps reflecting white matter
microstructural properties, making for more specific and sensitive data. They can thus use these maps as markers of myelination and iron content, which give information about the integrity of inter-neuron communication and brain metabolism, respectively. qMRI has also shown smaller variation in data acquired from different scanners compared to traditional approaches, making the technique a good candidate for longitudinal, multi-center studies (Callaghan et al., 2014; Weiskopf et al., 2013).
qMRI, however, also has some disadvantages at each level described above for VBM. Deoni et al., (2008) explain that reproducibility of certain tissue parameters can be affected by the measurement technique used. Also, if only a few parameters are used then the analyses is limited to a regions of interest (ROI’s) analysis since each tissue parameter is regionally specific (Draganski et al., 2011; See section 4.2). Finally, clinical interpretation and efficacy should be carefully examined since qMRI is not currently used in many clinical settings and rarely in BCS research (West, 2014). Essentially, it can be used to derive information about the microstructure of the brain and lead to better understanding of the specific neural mechanisms involved in the cognitive impairment seen in BCS.
2.1 MRI fundamentals
To get a better understanding of the workings of VBM and qMRI, core concepts of brain MRI will be reviewed briefly. As McRobbie (2007) explains, MRI is a tomographic imaging modality that relies on
the physical properties of hydrogen atoms. It works by an interaction between a magnetic field applied by the scanner and certain tissue properties. This interaction generates signals that are received by the scanner and can be used to produce images of the brain.
The tissue properties that are important for producing signals for imaging include relaxation times (T₁ and T₂) and proton density (PD). The relaxation times refer to the time it takes protons to get back to equilibrium after a magnetic field is applied. PD refers to hydrogen atom content per unit volume. Since different types of tissue differ in their atomic composition, different signal intensities can be produced. These signals are induced by radiofrequency (RF) pulses. Varying sequences of these pulses with preset timings and durations affect tissue properties differently and can be used to create different types of image contrasts.
The image contrasts that can be produced include T₁-weighted, T₂-weighted, PD-weighted, and T₂*-weighted images. The choice of pulse sequence can yield these weighted images by differentially affecting the signal intensity. Each can be favorable based on what is being examined (McRobbie, 2007). For example, T₁-weighted images are known to be useful in displaying the anatomy of structures because of the strong contrast they provide between gray and white matter. T₂-weighted are better scans for pathology. They display tissue with high water content at a high brightness, therefore they are suitable for displaying cerebrospinal fluid (CSF) and lesioned tissue. White matter in these scans appears dark due to its high fat content. Information provided by these images can, therefore, help elucidate the structure of the brain.
VBM fundamentals
Conventionally, VBM uses T₁-weighted images, obtained through specific scanner pulse
sequences, to segment brain tissue into gray matter, white matter, and CSF (Ashburner & Friston, 2000). This is done through obtaining a weighted average of tissue parameters per voxel, which entails that these images are based on several tissue properties instead of individual ones (Davatzikos, 2004). Because of this weighting method, T₁-weighted images are also affected by hardware and scanning artefacts. The images produced can be used to compare the structure of and to find anatomical differences between brains. Several preprocessing steps must be done before this comparison can be made. These consist of normalization, segmentation, smoothing, and modulation.
First, individual brain scans are registered onto a standard reference brain, such as a Montreal Imaging Institute (MNI) template or a native space template derived from the study subjects themselves
(Mechelli, Price, Friston, & Ashburner, 2005). This means that the individual images are normalized to the same stereotactic space in order to correct for differences in global brain shape and position. This step is also necessary to account for partial volume effects that could confound segmentation.
Second, an analytical technique is used to segment the brain into gray matter, white matter, and CSF. To achieve this, information about the spatial distribution of different tissue types contained in a
priori probability maps is convolved with information about the distributions of voxel intensity of
particular tissue (Mechelli et al., 2005). In standard VBM, segmentation is done after the normalization of brain images to a standard space.
Third, the brain data is smoothed and transformed before parametric analysis techniques are applied. Smoothing is applied in order to increase the signal-to-noise ratio (SNR), to render the data more normally distributed, and to deal with the structural variation inherent between subjects by increasing correspondence between voxels (“Spatial Smoothing”, 2010). To smooth the data, the average concentration of gray matter from around each voxel is taken and imposed on that voxel. More
specifically, a low pass filter, usually a Gaussian distribution, is applied to the imaging data resulting in blurred edges and enhanced spatial correlation (“Spatial Smoothing”, 2010). The normal distribution curve of the Gaussian function can have different kernel sizes defined by the full-width half-maximum (FWHM). To choose kernel size, it is recommended that studies use the matched filter theorem, which states that kernel size should be equal to the spatial differences expected between groups (Jones, Symms, Cercignani, & Howard, 2005). This process and a logit transformation make the data more normally distributed and suitable for parametric analysis.
An additional step called modulation can be applied when analysis needs to be done specifically on absolute volume for comparing two or more groups. This step controls for volume differences that arise due to spatial normalization (Mechelli et al., 2005). It is a correction step that is done by multiplying gray matter that has been normalized by its relative volume before and after normalization to yield an absolute volume measure of gray and white matter. Without this step, the interpretation of the results should refer to the relative concentration of gray or white matter. Thus, whether the data has been modulated or not needs to be explicitly stated in study reports because the interpretation of results and comparison of studies can be effected (Ridgway et al., 2009).
At this stage, standard parametric statistical analyses can be applied. In VBM, this is done voxel-wise so that univariate statistics are run on each voxel of the brain. This process has the advantage of providing information about local differences of brain tissue concentration between brains without bias towards any particular brain structure. However, since this analysis method requires multiple testing,
researchers and statisticians need to consider using a correction for multiple comparisons. It can be that a correction is not performed because of reduced statistical power in small sample sizes (Ridgway et al., 2009). However, without a correction, multiple testing increases the risk of type I error, which can be conceptualized by family-wise error rate (FWER) and false discovery rate (FDR). Several correction methods can be applied to control for these including parametric and permutation-based statistics. What is important is that the choices made for the type and level of correction are reported since each correction type has alternative options that can affect results.
A final point that should be made is that an optimized rather than a standard VBM technique can be applied if there are structural differences not associated with gray or white matter volumes (Mechelli et al., 2005). For example, during the normalization of brain images to a standard space, group
differences in size of the ventricles can affect the size of gray and white matter. This can lead to misleading gray matter volumetric data. This can happen with standard VBM where the preprocessing steps are executed in the order presented above. With optimized VBM, however, an additional
segmentation step is done before normalization to correct for this error. To implement this, images are first segmented into gray and white matter and then normalized to gray and white matter templates. Hence, normalization parameters that are derived can then be applied to the original brain images. The subsequent segmentation, smoothing, modulation, and analysis steps are akin to those of standard VBM, (See Fig. 1).
Fig. 1. Overview of voxel-based morphometry (VBM) preprocessing steps. (a) Standard VBM uses a template in standardized
space for normalization followed by segmentation of gray matter (GM) and white matter (WM), an optional modulation step, and smoothing before analysis. (b) Optimized VBM uses a template in native space for normalization, which requires an additional segmentation step beforehand and then follows a similar path as standard VBM. Adapted from “Voxel-Based Morphometry of the Human Brain: Methods and Applications,” A. Mechelli, C. J. Price, K. J. Friston, and J. Ashburner, 2005, Current Medical Imaging Reviews, 1, p. 3.
These steps allow for a lot of flexibility in terms of the options available to researchers. This flexibility is advantageous because it gives the opportunity to test different kinds of hypotheses. However, the choices made need to be laid out in detail to facilitate the reproducibility and consistency of data across studies that make similar predictions.
qMRI fundamentals
qMRI gives information about microstructural tissue properties (Weiskopf et al., 2015). This information is derived from distinct properties of hydrogen protons and can be affected by several factors, such as the macromolecular concentration in a voxel. They include T₁, T₂, PD, magnetization transfer saturation (MT), mean diffusivity (MD), and magnetic susceptibility among others. Each of these parameters can be quantified separately according to different MRI acquisition approaches and the biophysical model used to fit the imaging data. Two steps are required in order to derive this information (Jara, 2013). First, a qMRI pulse sequence is implemented during data acquisition. Second, a qMRI algorithm, specifically a biophysical model, is applied to the imaging data acquired in order to produce qMRI maps. Through this technique, qMRI provides voxel-wise measures of specific tissue parameters. Since these parameters are affected by the concentration of macromolecules in tissue they yield information about myelination, iron content, white matter integrity, and mean diffusivity. These measures are specific in that they quantify the parameters individually and independently of scanner effects and are thus sensitive to subtle changes in tissue microstructure.
In the same way that weighted images are derived, different RF pulse sequences can be used to act on specific tissue macromolecular content (Callaghan et al., 2014). The signal strength produced as a result of the pulses will differ depending on the amount of macromolecular content in a voxel. Therefore, tissue parameters can be quantified based on the signal produced. For example, MT causes changes in image intensity based on the concentration of macromolecular content (i.e. proteins) in a voxel, R1 (longitudinal relaxation rate, R1 = 1/T1) is sensitive to the mobility of water, and R2 (transverse relaxation
rate, R2 = 1/T2) depend on the iron content in gray and white matter (Draganski et al., 2011). Therefore,
these tissue parameter measurements can be associated with different histological factors, such as myelination for MT, axonal properties for R1, and iron concentration for R2 (Weiskopf et al., 2015).
Examples of contrast acquisitions techniques and methodologies that have employed qMRI include quantitative susceptibility mapping (QSM), quantitative multi-parametric mapping (MPM), diffusion tensor imaging (DTI), and voxel-based quantification (VBQ). QSM is a qMRI contrast that yields a measure of magnetic susceptibility (Deistung et al., 2013). MPM permits mapping of R1, R2, MT, and PD simultaneously (“Quantitative MRI and Voxel-Based Quantification (VBQ),” n.d.). DTI yields
information about white matter integrity by providing measures concerning the diffusion of water molecules, such as fractional anisotropy (FA) and MD (de Ruiter et al., 2012). VBQ is a methodological framework adapted from VBM, but instead uses MPM to map tissue parameters for group comparison
rather than T1-weighted images (Draganski et al., 2011; Callaghan et al., 2014; “Quantitative MRI and
Voxel-Based Quantification (VBQ),” n.d.).
The specificity and sensitivity of qMRI allows for comparison of tissue parameter measurements between groups, which can indicate white matter differences. These aspects of qMRI also make possible the direct comparison between different scanners and imaging centers. Lastly, qMRI can tell us about histological measurements of tissue microstructure and can complement VBM macrostructural data (Weiskopf et al., 2015).
2.2 Neuroimaging in breast cancer survivor studies
Cross-sectional, as well as longitudinal brain imaging studies are becoming increasingly prevalent in BCS as the number of survivors continues to grow due to advances in treatments. BCS neuroimaging studies that have focused on the effects of cancer and its treatment on the structure of the brain aim to identify the underlying neural correlates of negative cognitive effects. Cancer- and treatment-related factors can have neurotoxic outcomes and the mechanisms through which this neurotoxicity might work may be direct and oxidative DNA damage (McDonald & Saykin, 2013; Ahles, 2012). Since some
neurotoxic agents can cross the blood brain barrier (BBB), it is possible that they could lead to neuronal damage. Genetic factors and age at diagnosis may also have an effect with respect to the vulnerability of patients to these neurotoxic effects.
To detect this damage, studies have looked at structural and functional brain attributes that may be associated with the cognitive problems seen in BCS (de Ruiter et al., 2012; Inagaki et al., 2007; Conroy et al., 2013; McDonald, Conroy, Ahles, West, & Saykin, 2010; Koppelmens et al., 2012; Lepage et al., 2014). These studies have shown differences in frontotemporal regions of the brain between BCS and healthy and/or non-chemotherapy treated BCS (McDonald & Saykin, 2013). These differences are evident in decreased gray matter density. Moreover, these regions are linked to working memory and executive functioning in general. In this population, deficits in these regions have been associated with poor performance on tasks testing these cognitive functions. The changes seen in these regions might be explained by the fact that temporal and frontal lobes are later-myelinating regions and are therefore more susceptible to breakdown (Callaghan et al., 2014). This means that these regions might be more vulnerable to the neurotoxic effects of chemotherapy.
Such studies have primarily used VBM for structural measurement and analysis of the brain. A few of these studies and their main findings are outlined here. A review on chemotherapy effects in
neuroimaging studies of cancer patients found that most studies report diffuse patterns of differences in gray and white matter volume (Simó et al., 2013). de Ruiter et al. (2012) found volume differences in gray matter in posterior parietal regions of the brain in BCS as compared to healthy controls. Another study showed that chemotherapy BCS have decreased gray and white matter volume in frontotemporal regions one year post-treatment compared to non-chemotherapy BCS and healthy controls (Inagaki et al., 2007). These differences, however, were not seen 3 years post-treatment implying possible recovery. VBM also showed decrease in gray matter density in BCS compared to controls in left middle temporal gyrus (Conroy et al., 2013). In this study, post-chemotherapy interval was positively associated with gray matter density revealing some recovery, though it was not complete. In a longitudinal study, gray matter reduction was seen one month after treatment with partial recovery after one year (Lepage et al., 2014). This reduction was seen mostly in frontal regions and was associated with poorer performance on working memory and executive functioning tasks. Importantly, frontotemporal regions did not fully recover.
These studies seem to show a consistent association between regions of gray matter loss and decreased cognitive performance. However, frontotemporal damage is not specific enough to pinpoint the mechanisms through which these differences come about and to explain individual differences. A detailed outline of these studies will be given later in this review to show the inconsistencies between them that might explain why some results are not reproducible.
A few BCS neuroimaging studies have applied qMRI to study structural brain changes associated with cognitive dysfunction. de Ruiter et al. (2012) used DTI in addition to VBM to measure white matter integrity in BCS. They found that FA was decreased and MD was increased in BCS who received
chemotherapy compared to a non-chemotherapy group in focal white matter tracts indicating
demyelination. Koppelmans et al. (2014) also looked at DTI and found widespread decreases in FA and increases in MD in chemotherapy-exposed BCS compared to healthy controls 20 years after treatment. A few other studies have employed DTI in similar investigations, however, no other qMRI techniques are known to be used in this clinical population.
There are several issues with the studies mentioned above in terms of the methodologies employed. For example, most have small sample sizes and/or are cross-sectional. The problem with small sample sizes is that the data do not have enough power to yield strong conclusions (Button et al., 2013) and thus make it difficult to test specific group hypotheses (Ahles, 2012). Cross-sectional studies make it difficult to assess disease progression and recovery over time and to look at individual
reliable and robust results. The main issue with running such studies is the implementation of a complex study design especially if VBM is used. For reasons that will be outlined below, longitudinal, multi-center studies using VBM, such as the one done by Lepage et al. (2014), can be problematic. However, qMRI can be a useful compatible tool that could resolve many of the limitations associated with VBM.
3.1 Shortcomings of voxel-based morphometry
Despite the popularity and frequency of use of VBM, there are some difficulties associated with its use. These can be attributed to image acquisition, preprocessing, and statistical analyses. Further, conceptual issues also exist for using this imaging method and should be considered based on the research goals of a study. Essentially, these concerns lead to challenges in acquiring reliable and comparable data for longitudinal, multi-center studies.
Image acquisition
Several problems are linked with the use of VBM during data collection. The hardware, software, and user settings can all have an effect on the outcomes of a study and render the data and results incomparable across time, sites, and studies. If two different scanners or scanner pulse sequences are used to measure group differences, then it becomes invalid to compare the groups with VBM because the differences may be attributed to the scanners themselves (Ashburner & Friston, 2000). Differences in coil loading and other hardware effects make it difficult to have consistency between scanners and scan sessions even when acquisition parameters are matched (Mechelli et al., 2005). RF coil sensitivity in particular can lead to variations in the data due to non-uniformities, such as B0 and B1 inhomogeneities,
in the signal and need to be corrected. These inhomogeneities are caused by slight variations in the position and orientation of the head in an RF coil and lead to signal and contrast biases as well as tissue misclassification in T1-weighted images (Weiskopf et al., 2013, Waehnert et al., 2016). They also limit the
capacity for T1-weighted images to be used in scanners with ultra-high field strengths (i.e. 7 tesla)
(Deistung et al., 2013). Henley et al. (2010) reported that choice of software and software version can have an effect on registration and segmentation processes as well as analysis, therefore, affecting results. They also found that there are inconsistencies across studies based on changing user-specified options.
At the preprocessing stage, issues can come about from several steps. For example, modulation of the imaging data corrects for local differences being lost due to normalization and represents absolute volume of gray matter, while unmodulated data represent concentration of gray matter (Henley et al., 2010). Though this process leads to results that should be interpreted differently, studies usually interpret them the same. Some studies also do not report whether the data is modulated or not (i.e. Conroy et al., 2013; McDonald et al., 2010).
Smoothing kernel size is also a factor to consider. Larger kernel size leads to increased
involvement of significant brain areas (Henley et al., 2010). This is only to a certain extent, however. For example, increasing the kernel size can lead to a failure to detect significant effects in small nuclei in the subcortex such as the subthalamic nucleus (STN) and the substantia nigra (SN) (Hollander, Keuken, & Forstmann, 2015). Jones et al. (2005) tested a range of kernel sizes (0 to 16 mm FWHM) that have been used in VBM studies and showed that different sizes lead to different significant results. In this study, FA values between schizophrenic patients and controls were compared and the following results were found: no differences with a kernel size range of 0 to 6-mm, differences in the superior temporal gyrus (STG) with a range of 7 to 8-mm, differences in the STG and cerebellum with a range of 9 to 14-mm, and only differences in the cerebellum with a range of 14 to 16-mm. Thus, the sensitivity of the smoothing kernel to detect differences varies with size. They also showed that the proportion of voxels with
normally distributed residuals is dependent on the smoothing kernel size, where the data is more normal with larger kernel size. Normal distribution of residuals is an assumption of the standard parametric analysis done with VBM. However, although bigger kernel size leads to more normal data, a lot of information is lost. Thus, it is crucial to base the smoothing kernel size on hypotheses concerning the expected group differences.
With regard to segmentation, signal intensity may be affected by neural tissue pathology in the patient population being measured and thus incorrectly affect morphometric segmentation of gray matter (Abbott, Pell, Pardoe, & Jackson, 2012). Aging factors have also been shown to affect tissue segmentation by altering the MR signal. One study examined such factors in order to parse out effects related to aging itself as compared to effects related to how aging differences can affect the imaging process (Jernigan et al., 2001). They found that there are age-related signal elevations in white matter structures. This can be a result of increased lightly myelinated white matter tissue voxels in elderly subjects being classified incorrectly as gray matter after segmentation. This finding could explain the inconsistencies in the reports of tissue loss in the literature concerning aging effects. Accordingly, aging
and pathological tissue properties should be taken into account given the effects they can have on the interpretation of study outcomes.
Statistical analysis
Concerning the statistical analysis used with VBM data, there are three main issues that need to be addressed. First, parametric analysis can be affected by multiple comparison correction type and level (Henley et al., 2010). No correction can lead to many false positives and correction can lead to loss of power and missing true positives. Second, studies adjust for total gray matter volume without including total intracranial volume as a covariate (Henley et al., 2010). In patients with a neurodegenerative disease, this can obscure some effects associated with the disease. Third, VBM does not sufficiently model interactions (Davatzikos, 2004). Although a multivariate approach would remedy these problems, it is not possible with the univariate nature of T1-weighted imaging data. Further, differences of similar
magnitude can be seen as significant or insignificant depending on orientation of group differences (as represented in high-dimensional space), sample size, and brain region.
Conceptual issues
Conceptual issues with VBM are important when considering appropriate usage and
interpretation of the results. This is because with all the difficulties described above, results become hard to interpret and generalize. A main feature of VBM is that it provides information about the macrostructure of gray matter tissue. This means that the understanding of underlying mechanisms is limited, especially in the vein of white matter microstructure (Draganski et al., 2011). It is important to note that VBM provides images acquired on a relative scale, which could obscure group differences (Engstrom, Warntjes, Tisell, Landtblom, & Lundberg, 2014). Additionally, Davatzikos (2004) points out that though VBM is good for localized and linear differences in brain structure, it is not useful for finding subtle, complex, and distributed changes especially those of a nonlinear nature.
A concluding reflection for VBM methodology is that there are many options to choose from at every step. This has its advantages in terms of facilitating study designs specific to the hypotheses in question. However, several concerns might also arise as a result of such analytical freedom. It cannot only be problematic because of the effects such choices can have on the data, but also because it makes it difficult to compare results across studies, particularly if these choices are not explicitly delineated.
This often leads to inconsistent findings. Careful consideration needs to be implemented when setting up a protocol involving VBM as well as explicit reporting of the image acquisition, preprocessing, and analysis techniques used with reasoning behind the choices made (Ridgway et al., 2009). In addition, these considerations should be made a priori based on clearly defined hypotheses. Here, data sharing can play a role. Poldrack and Gorgolewski, (2014) outline the benefits of sharing of neuroimaging data for the purposes of enhancing reproducibility of results and improving research practices among other goals. Data sharing can facilitate replication studies with specified hypotheses in order to uphold the robustness of results.
3.2 Issues with BCS studies using VBM
Six BCS neuroimaging studies (four cross-sectional and two longitudinal studies) will be analyzed closely in this review in terms of the design and methods used in the experiment, (See Table 1). All of these studies used VBM for examining structural changes in the brain. Also, they all used T₁-weighted contrasts for image acquisition.
Inagaki et al. (2007) compared chemotherapy (C+) and non-chemotherapy (C-) treated BCS with healthy controls using two neuroimaging experiments in a cross-sectional setting. The first experiment included patients a year after they had received chemotherapy and the second included those who are 3 years post-treatment. They found gray and white matter volume differences in superior and middle frontal gyri, parahippocampal gyrus, cingulate gyrus, and precuneus in the C+ group only 1 year post treatment. No differences were found 3 years after treatment. However, this conclusion might be erroneous given that the authors based it only on a comparison of statistical significance of both effects according to a 0.05 p-value (Nieuwenhuis, Forstmann, & Wagenmakers, 2011). They should instead have looked at the statistical significance of the difference between the effect seen at 1 year and the effect seen at 3 years post-treatment. Though this study included a relatively large sample size of about 50 to 70 patients and two time points, it still had a cross-sectional design making it difficult to assess the progress of brain damage and evaluate recovery.
Another cross-sectional study looked at structural differences between BCS who received chemotherapy and those who did not 9 years after treatment (de Ruiter et al., 2012). They did not use a healthy control group. They found that the C+ group had decreased in gray matter in left lateral posterior parietal cortex, bilateral precuneus, left occipital cortex, and bilateral cerebellum compared to the C- group. VBM does not allow for standardized reference data of healthy population, therefore, not having
control group makes it difficult to assess anatomical changes. They also had a small sample size of 17 to 19 patients, limiting the power of the study.
In a study done on a large sample of 184 BCS exposed to chemotherapy 21 years ago, VBM was used to look at focal tissue volume loss at one time point (Koppelmans et al., 2012a). A large healthy control group of 368 individuals was also used, but a C- control group was not included. It was found that the C+ group had less gray matter volume compared to controls, but no differences were found in white matter volume. These significant differences cannot be attributed to chemotherapy itself because a C- control, which is necessary for the parsing out of cancer-related effects, was not used.
Conroy et al. (2013) compared chemotherapy-treated BCS 3 to 10 years post-treatment with healthy controls. They had a sample size of 24 to 27 patients. Gray matter density reduction was seen in the left temporal lobe, right midbrain, left thalamus, right cerebellum, and right insula in C+ compared to controls. Again, a non-chemotherapy control group was not used making it difficult to link chemotherapy effects to structural differences found.
Using a longitudinal design, McDonald et al., (2010) examined C+ and C- treated BCS and healthy controls. They used three different time points: before chemotherapy treatment, 1 month following treatment, and 1 year following treatment. They had a small sample size of 12 to 18 patients. After 1 month, differences were found in C+ compared to controls. They had reduced gray matter density in the bilateral middle frontal gyri and left cerebellum. Also, compared to baseline, within-group differences in the C+ group were seen in bilateral frontal, temporal, and cerebellar regions and right thalamus. Partial recovery was seen after 1 year.
Another longitudinal study looked at the same time points as the one just mentioned, but only tested a chemotherapy treated group compared to healthy controls (Lepage et al., 2014). They also had a small sample size of 19 patients. 1 month after treatment, C+ showed diffuse reductions in gray matter volume in frontal, temporal, parietal, and occipital regions compared to controls with partial recovery after 1 year. VBM has not been shown to have good reproducibility therefore rendering the longitudinal data problematic due to the possibility of inconsistency of the acquired data and results across time points (Weiskopf et al., 2013).
Table 1.
Outline of methodology of structural MRI studies using voxel-based morphometry (VBM) in breast cancer survivors (BCS) Study Design Time (years) Groups n (C+ ) Scanne r VBM implementation Modulatio n Smoothin g (FWHM) Multiple compariso n correction Interpretation Inagaki et al., (2006) Cross-sectional 1, 3 a C+, C-, healthy 51,
controls de Ruiter et
al., (2012)
Cross-sectional >9 C+, C- 19 3T Standard Yes 7-mm Yes Volume
Koppelman s et al., (2012)a Cross-sectional 21 C+, healthy controls
184 1.5T Optimized Yes 8-mm Yes Volume
Conroy et al., (2013) Cross-sectional 3-10 C+, healthy controls 27 3T Optimized Not reported 10-mm No Density McDonald et al., (2010) Longitudina l Baselineb, 1 month follow-up, and 1 year follow-up C+, C-, healthy controls 17 1.5T Optimized Not
reported 10-mm Yes Density
Lepage et al., (2014) Longitudina l Baselineb, 1 month follow-up, and 1 year follow-up C+, healthy controls
19 1.5T Standard Yes 3-mm Not
reported Volume
a
The 1 year and 3 year time points refer to two separate samples of BCS bBaseline refers to the time point before chemotherapy treatment
Some of these studies showed that there might be recovery of structural damage in later years. However, there are several reasons why subtle differences might remain hidden in the context of VBM analysis especially if there is partial recovery. The range of smoothing Gaussian kernel used in these studies is 3-mm to 12-mm FWHM. None of the studies justified the reason for the width chosen for the kernel. Since this has been shown to affect the sensitivity of the analyses, it is possible that certain effects were not apparent due to the smoothing kernel chosen. Aging is another factor that needs to be considered in these studies since aging-like effects are evident in BCS. Particularly, age of breast cancer patients can play a role in how vulnerable these patients are to the neurotoxic effects of chemotherapy (McDonald et al., 2013, Ahles et al., 2012).
Inconsistencies can also arise due to the kind of brain differences that are usually seen in BCS. These differences are said to be subtle and diffuse. Moreover, different morphological profiles might lead to different paths of the same disease as a consequence of the nonlinear features of brain structures (Davatzikos, 2004). Thus, VBM might not be the best way to examine anatomical changes in this population.In terms of inconsistencies due to statistical analysis, it should be noted that not all studies used multiple comparisons corrections. Conroy et al. (2013) explained that the sample size was too small to apply this kind of correction. Lepage et al. (2014) were unclear as to whether they used any correction for multiple comparisons, since none was reported. These studies risk finding outcomes that are due to chance rather than true effects.
Importantly, all of these studies were single center studies. Due to the difficulties of comparing VBM data across different sites, especially when it comes to the hardware, software, and user settings effects of using different scanners, it is not practical to implement multi-center studies. Further, it is hard to establish the reliability of results in the longitudinal VBM studies given poor reproducibility of T₁-weighted images. Longitudinal, multi-center studies with consistent data are needed to alleviate some of the issues found in these studies such as small sample sizes, insufficient control groups, and
comparability of data across different time points.
Despite all of the drawbacks that come with applying VBM methodology to a study concerned with structural neuroimaging, VBM is still a useful tool and can be improved. Attentiveness at every step discussed is essential and can by itself improve the reliability and validity of results. However, some issues cannot be improved based solely on careful reflection on the design of the study, but concern technical and conceptual issues that remain despite such efforts. For these, qMRI might be a suitable extension to VBM in that it could resolve such issues.
4.1 Benefits of qMRI
As mentioned before, qMRI allows for the quantification of microstructural changes in the brain. It reflects individual properties of tissue, meaning that specific measures of each property (i.e. T₁, T₂, PD, MD, MT, and magnetic susceptibility) can be simultaneously obtained. These specific measures are differentially sensitive to microstructural changes in tissue (Draganski et al., 2011). Thus, qMRI can lead to a better parcellation by providing the independent contribution of these measures, which reflect iron, myelin, and mean diffusivity in MRI among other properties (Stuber et al., 2014). In turn, analysis of these quantified parameters can be used to give information about the white matter changes associated with a disease.
qMRI provides the opportunity for performing large scale studies. Multi-center and longitudinal studies lead to better identification of inter-individual differences, small effects, and estimation of neuroanatomical population variance. Studies have shown that variation of qMRI data from different scanners is small relative to other structural brain neuroimaging techniques (Callaghan et al., 2014). Weiskopf et al. (2013) tested inter-site and intra-site reliability of quantitative tissue parameter images as compared to T₁-weighted images. They showed, using an MPM protocol, that quantitative imaging parameters have a higher reproducibility than T₁-weighted images. T₁-weighted images were shown to be more biased in signal and contrast due to these RF field inhomogeneities. qMRI mapping acquisition
tools such as MPM can correct for B0 and B1 inhomogeneities. MPM has the advantage of mapping tissue
parameters separately from the RF transmit and receive field inhomogeneities, therefore, allowing these nonuniformities in the signal to be removed or reduced (Waehnert et al., 2016, “Quantitative MRI and Voxel-Based Quantification (VBQ),” n.d.). Thus, quantitative imaging approaches remove hardware-dependent factors and can be used with no further corrections. Good reproducibility of qMRI parameters can help ensure more consistent results.
Another study also showed good inter- and intra-site reliability of T₁ and T₂ (Deoni et al., 2008). The imaging approach they used, driven equilibrium single pulse observation of T₁ and T₂ (DESPOT1 and DESPOT2), showed good inter- and intra-site reproducibility of relaxation times allowing comparison of voxel values across sites and sessions. They reported that the test-retest coefficient of variation (CoV), a relative variability measure, between sites for T₁ and T₂ were 6.8% and 8.3%, respectively. Within the same site, but across different time points, the CoV values were 6.4% and 7.9%. This means that the data is reproducible between different scanners and at different measurement times. Further, this approach is clinically relevant given the short acquisition time of about 15 minutes.
qMRI can provide an important contribution to BCS research because it can be used to develop reference values of MRI tissue parameters in a healthy population. This would greatly benefit studies that do not have access to a healthy control group for testing. qMRI parameter values of a patient population can, thus, be compared to a standardized set of values derived from a healthy population. These values can also be correlated with clinical measures of disability as a way of finding underlying mechanisms behind disease symptomology (Engstrom et al., 2014). Since qMRI can capture the biochemical processes of white matter changes due to a disease, it might be better for finding early changes resulting from that disease (Deoni et al., 2008).
Further, qMRI contrasts can be acquired at ultra-high field strengths such as 7 tesla because they can be controlled for RF field inhomogeneities (Deistung et al., 2013). Such field strengths provide better spatial resolution as increased SNR can yield smaller voxels. This can lead to better segmentation and, thus, anatomical specificity. For example, QSM has provided reliable contrasts that allow for the identification of subcortical structures that are difficult to distinguish using weighted imaging or less sensitive qMRI contrasts (Deistung et al., 2013). QSM is highly sensitive to local iron content and can be very beneficial to clinical research, given the association of elevated iron with neurodegenerative diseases. Thus, the benefits of qMRI, in terms of the information it can yield and the reproducibility of its measures among other things, make it an appropriate tool for clinical research.
4.2 Issues with using qMRI
Even though qMRI has a lot of advantages and could add to VBM, some issues still hold. Some of these can be seen at the image acquisition level and include measurement techniques. These can affect reproducibility of certain tissue parameters, such as relaxation times (Deoni et al., 2008). Sufficient repeated samples are required in order to produce reliable data. Conventionally, this led to long scan times. However, new techniques have cut the scanning time to a comparable level with clinical scans (Setsompop, Feinburg, & Polimeni, 2016; Deoni et al., 2008).
Also, it is not clear whether data from different magnetic field strengths are comparable because they have been shown to effect relaxation rates (West, 2014). This could present as a problem for multi-center studies. Clinically, however, this might not be significant since most recent BCS studies use either 1.5T or 3T scanners making it possible to match based on scanner field strength.
Since the quantification of tissue parameters is specific and sensitive, using one or a few parameters limits analyses to ROI (Draganski et al., 2011). In other words, different regions are represented based on the parameter used because each parameter is sensitive to different tissue properties. For example, some are more sensitive to water content, while others are more sensitive to iron content of a voxel, leading to parameter maps with different information regarding the micro-architecture of tissue. This is problematic for an analysis that needs to be unbiased to any region of the brain, but can be useful if there are expectations of where the changes should be in the brain. A voxel-wise analysis can be done with qMRI if multiple parameters are used. Draganski et al. (2011) and Callaghan et al. (2014) applied this approach using the voxel-based quantification (VBQ) technique. VBQ is a method used to quantify multiple tissue parameters and uses standard parametric mapping for voxel-wise data analysis of the whole brain. Thus, it is based on VBM, but provides specific instead of weighted images. However, the problems of multiple comparisons correction and spatial smoothing in whole brain analysis are still a concern in this approach.
Though qMRI is becoming more and more popular for use in research, it is still not established in terms of methods and clinical efficacy (West, 2014). Specifically, segmentation of gray and white matter can be unreliable even with qMRI. This can be due to local differences in myelination or to registration errors, for example. Also, very few neuroimaging studies in BCS have used qMRI contrasts and the ones that did have only used DTI (i.e. de Ruiter et al., 2012). This can be remedied by using methods such as VBQ, since it is based on VBM. Such an adaptation of an established methodological framework will
allow reliable acquisition of quantitative tissue parameter measures in addition to tissue volumetric information in order to compare groups at a macro- and microstructural level.
4.3 qMRI in neurodegeneration studies
Though qMRI techniques are rarely used in BCS studies, their application has been seen in several neurodegeneration studies. Of significance is a study done by Draganski et al. (2011) that uses qMRI to measure structural changes due to the aging process. In this study, both VBM and VBQ were implemented and compared. VBQ method was important in identifying age-related white matter changes that were not seen in the VBM analysis of T₁-weighted images as well as validating macrostructural changes seen in VBM. These white matter changes were reflected in patterns of decreasing FA, MT, and R1, increasing MD, and increasing and decreasing R2* with age in widespread white matter tracts including frontal, parietal, and temporal regions. They also showed that a
multivariate statistical approach could be used with VBQ given that multiple parameters are acquired simultaneously per voxel and results in comparable findings to the standard univariate analysis
commonly used. Conventional VBM only provides gray matter volume or density, making it inappropriate to apply multivariate analyses. Methods using qMRI can, therefore, complement and add to research concerned with structural brain differences between groups.
Another study looked at voxel-based qMRI measures of T1, T2, and PD to compare multiple
sclerosis (MS) patients with healthy controls (Engstrom et al., 2014). MS has been associated with diffuse and global white matter changes, which are difficult to capture using conventional MRI methods. All three quantitative parameter maps were good for discriminating between gray matter, white matter, and CSF. With voxel-based analysis, group differences were displayed by lower R1 and R2 and higher PD values in MS patients compared to healthy controls in diffuse cortical and subcortical white matter regions. In the study, two-dimensional multi-parametric visualizations were also used to represent the quantitative tissue parameters in order to detect the direction of change. These probable directions of change can be verified by longitudinal studies using qMRI and thus yield information about the early signs of disease.
In the context of breast cancer, such studies might be important to consider because BCS have shown structural changes that are akin to those of neurodegeneration and could be described as a speeding up of the deleterious effects of aging such as those described in the Draganski et al., (2011) study (Koppelmans et al., 2012a, McDonald & Saykin, 2013, Ahles et al., 2012). Thus, VBQ using
multi-parametric image acquisition protocols such as MPM could be applied to BCS studies as they have been in neurodegenerative studies. Also, since neurological tissue changes in BCS are widespread and diffuse, qMRI specificity and sensitivity features might make it a better candidate for capturing abnormal changes in the brain due to cancer and its treatment, especially those concerning white matter that are difficult to discover using conventional neuroimaging methods.
Discussion
Studies using VBM can uncover structural differences between groups based on gray matter volume changes. This method is used extensively in clinical settings and is dominant in breast cancer studies assessing neurocognitive changes. It comes with many benefits such as an established
methodology, the capacity to provide information about brain tissue macrostructure, and a non-biased voxel-wise analysis of the whole brain. VBM has provided considerable data about the effects seen in clinical populations, such as breast cancer survivors, that has helped guide research. It has allowed for the identification of structural brain changes that are associated with and complement the mild cognitive impairment seen in chemobrain. Neuroimaging studies that use VBM have attempted to provide
validation for the cognitive effects seen in BCS. However, further validation studies are needed to pinpoint the specific neural mechanisms behind the structural changes associated with chemobrain and to show how these mechanisms are affected over time. This is difficult to do without more specific and sensitive measures that can capture the subtle and diffuse effects seen in BCS. Multi-center studies are needed in order to access larger samples of BCS and in order to compare different kinds of breast cancer treatments (i.e. chemotherapy vs. endocrine therapy). Longitudinal studies are also needed to assess disease progression and recovery over time.
VBM presents issues during image acquisition, preprocessing, statistical analyses, and interpretation of results. Hardware, scanner software, and user options have been shown to have an effect on imaging data. Preprocessing options within each step also have effects on the data. These factors as well as drawbacks with the analysis can affect the outcomes of a study and, therefore, should be considered carefully. Even the interpretation of the results can be affected by the methodological specifications applied. One approach that can settle several of these issues is through substituting T1
-weighted images obtained from VBM with qMRI-derived tissue parameter maps. This can be done by fitting qMRI models to the imaging data to produce specific tissue parameter values instead of
volumetric information based on weighted images. These maps can then be compared between patients and controls to look at group differences and changes.
qMRI can complement VBM
Though qMRI seems like a promising method to use in studies looking at structural changes in clinical populations, VBM still dominates. Additionally, even though they cannot solve every problem associated with these neuroimaging study designs, qMRI parameter measures can complement conventional morphometric analysis. This has been shown in studies that applied VBQ.
Many of the advantages that qMRI offers to VBM can extend research in BCS. For example, it is possible that neurotoxic effects can lead to iron deposition and decreased white matter integrity. It is also possible that long-term effects in white matter structure are still present even though gray matter recovery is seen. Based on the research discussed in this review, several hypotheses can be made concerning white matter effects in BCS. It is expected that BCS that had received chemotherapy will show diffuse patterns of lower R1, R2, and MT and higher PD values compared to a non-chemotherapy treated group and healthy controls in widespread regions including frontal and temporal white matter tracts. These findings would suggest increased iron deposition, axonal injury, and demyelination, which might contribute to the cognitive dysfunction seen in BCS. This can be tested by looking at how tissue parameter data is associated with neuropsychological data. They would also support DTI findings of lowered white matter integrity and have implications on the underlying pathologies of gray matter structural changes. Subcortical investigations could reveal white matter changes in areas known to be involved in memory and learning, such as the hippocampus. By assessing and comparing qMRI parameter maps, researchers can uncover such effects. They can also add to and provide support for gray matter structural changes found through VBM analysis.
Subcortical segmentation is often overlooked in VBM. This is partly due to the inadequacy of this methodology for providing appropriate contrasts when comparing between groups. qMRI has been shown to improve the segmentation process (Deistung et al., 2013; Engstrom et al., 2014). Because of this, it might be more appropriate to use to examine subcortical structural changes that have not yet been assessed in BCS. With reports of diffuse and subtle brain tissue damage, it is likely that damage is not just limited to the cortex. Improved segmentation can also benefit the derivation of volumetric measures (Engstrom et al., 2014). That is, qMRI information can be used as input in VBM automatic brain segmentation procedures leading to more robust yet complementary volumetric information for group
comparison. Further, findings from VBM can be used to derive hypotheses about the location of expected cortical damage. qMRI can then be used in an ROI analysis based on such predictions. This is advantageous if small sample sizes are involved and makes multiple comparison corrections
unnecessary.
Two major limitations in studies examining the BCS cohort are small sample sizes and limited longitudinal data. Given that qMRI has shown reproducibility across scanners and sites, it can facilitate longitudinal, multi-center studies by increasing access to a larger sample of patients and yielding consistent data over time. Also, since current studies do not always include appropriate control groups, larger sample sizes can help uncover effects specifically related to chemotherapy by parsing out other cancer- and treatment-related factors. Longitudinal studies are necessary to assess disease progression or healing. They are also beneficial for controlling for heterogeneities found between subjects. Current long-term VBM outcomes in BCS need to be validated because they have not shown good reproducibility across time points. This can be done through qMRI. Several studies have shown that qMRI data remains consistent over time, much more so than VBM data (Weiskopf et al., 2013).
In any research setting, qMRI can help with standardization of parametric measures. In breast cancer studies, specifically, qMRI can be used to obtain standard scores based on different age cohorts of BCS. These can make the recruitment process easier, especially if there is no access to a healthy control group, because BCS groups can be compared with a standardized norm based on the age-specific standard scores instead of healthy controls. They can also provide values that can be shared through open data.
Some considerations suggested for VBM should also be deliberated when using qMRI. Even though qMRI can augment VBM, there are still issues that it cannot resolve as well as issues that are inherent in the contrast tool itself. These include smoothing kernel size, modulation, voxel-wise univariate analysis with multiple testing, and interpretation of the results.
Moreover, some of these problems are not necessarily specific to VBM, but are immanent in the methodology of neuroimaging studies. Primarily, not having a priori hypothesis based predictions can result in unsupported choices when it comes to specific scanner settings, preprocessing options, analysis approaches, and reporting of results. This can be problematic because the methodological choices, such as the choice of smoothing kernel size, have been shown to affect results. This can lead to inconsistent results across different studies. It can also lead to studies that cannot be replicated or compared.
In the context of structural MRI research, it is important to consider new approaches that could advance structural neuroimaging methodology and lead to consistent data and results. With qMRI applied in a VBM framework, such as has been done with VBQ, we can improve the reliability of structural MRI imaging and learn more about the microstructural neural characteristics associated with both macrostructural changes and neuropsychological impairment in patients with cancer, including breast cancer survivors. By yielding more consistent data in longitudinal, multi-center studies, qMRI can provide the information necessary to validate findings in research and can thus lead to better profiles for diagnosis, improved tracking of disease prognosis and recovery, degree of impairment, and appropriate treatments.
References
Abbott, D. F., Pell, G. S., Pardoe, H. R., & Jackson, G. D. (2012). Selecting appropriate voxel-based methods for neuroimaging studies. NeuroImage, 59(2), 885–886.
Ahles, T. A. (2012). Brain vulnerability to chemotherapy toxicities. Psycho-Oncology, 21(11), 1141–1148.
Ahles, T. A., Root, J. C., & Ryan, E. L. (2012). Cancer- and cancer treatment-associated cognitive change: an update on the state of the science. Journal of Clinical Oncology: Official Journal of the American
Society of Clinical Oncology, 30(30), 3675–3686.
Ashburner, J., & Friston, K. J. (2000). Voxel-based morphometry--the methods. NeuroImage, 11(6 Pt 1), 805–821.
Biglia, N., Bounous, V. E., Malabaila, A., Palmisano, D., Torta, D. M. E., D’Alonzo, M., … Torta, R. (2012). Objective and self-reported cognitive dysfunction in breast cancer women treated with chemotherapy: a prospective study. European Journal of Cancer Care, 21(4), 485–492.
Boykoff, N., Moieni, M., & Subramanian, S. K. (2009). Confronting chemobrain: an in-depth look at survivors’ reports of impact on work, social networks, and health care response. Journal of Cancer
Survivorship: Research and Practice, 3(4), 223–232.
Button, K. S., Ioannidis, J. P. A., Mokrysz, C., Nosek, B. A., Flint, J., Robinson, E. S. J., & Munafò, M. R. (2013). Power failure: why small sample size undermines the reliability of neuroscience. Nature Reviews
Neuroscience, 14(5), 365–376.
Callaghan, M. F., Freund, P., Draganski, B., Anderson, E., Cappelletti, M., Chowdhury, R., … Weiskopf, N. (2014). Widespread age-related differences in the human brain microstructure revealed by quantitative magnetic resonance imaging. Neurobiology of Aging, 35(8), 1862–1872.
Conroy, S. K., McDonald, B. C., Smith, D. J., Moser, L. R., West, J. D., Kamendulis, L. M., … Saykin, A. J. (2013). Alterations in brain structure and function in breast cancer survivors: effect of
post-chemotherapy interval and relation to oxidative DNA damage. Breast Cancer Research and Treatment,
137(2), 493–502.
Davatzikos, C. (2004). Why voxel-based morphometric analysis should be used with great caution when characterizing group differences. NeuroImage, 23(1), 17–20.
de Hollander, G., Keuken, M. C., & Forstmann, B. U. (2015). The subcortical cocktail problem; mixed signals from the subthalamic nucleus and substantia nigra. PloS One, 10(3), e0120572.
Deistung, A., Schäfer, A., Schweser, F., Biedermann, U., Turner, R., & Reichenbach, J. R. (2013). Toward in vivo histology: a comparison of quantitative susceptibility mapping (QSM) with magnitude-, phase-, and R2*-imaging at ultra-high magnetic field strength. NeuroImage, 65, 299–314.
Deoni, S. C. L., Williams, S. C. R., Jezzard, P., Suckling, J., Murphy, D. G. M., & Jones, D. K. (2008).
Standardized structural magnetic resonance imaging in multicentre studies using quantitative T1 and T2 imaging at 1.5 T. NeuroImage, 40(2), 662–671.
de Ruiter, M. B., Reneman, L., Boogerd, W., Veltman, D. J., Caan, M., Douaud, G., … Schagen, S. B. (2012). Late effects of high-dose adjuvant chemotherapy on white and gray matter in breast cancer survivors: converging results from multimodal magnetic resonance imaging. Human Brain Mapping, 33(12), 2971– 2983.
Draganski, B., Ashburner, J., Hutton, C., Kherif, F., Frackowiak, R. S. J., Helms, G., & Weiskopf, N. (2011). Regional specificity of MRI contrast parameter changes in normal ageing revealed by voxel-based quantification (VBQ). NeuroImage, 55(4), 1423–1434.
Engström, M., Warntjes, J. B. M., Tisell, A., Landtblom, A.-M., & Lundberg, P. (2014). Multi-parametric representation of voxel-based quantitative magnetic resonance imaging. PloS One, 9(11), e111688.
Henley, S. M. D., Ridgway, G. R., Scahill, R. I., Klöppel, S., Tabrizi, S. J., Fox, N. C., … EHDN Imaging Working Group. (2010). Pitfalls in the use of voxel-based morphometry as a biomarker: examples from huntington disease. AJNR. American Journal of Neuroradiology, 31(4), 711–719.
Inagaki, M., Yoshikawa, E., Matsuoka, Y., Sugawara, Y., Nakano, T., Akechi, T., … Uchitomi, Y. (2007). Smaller regional volumes of brain gray and white matter demonstrated in breast cancer survivors exposed to adjuvant chemotherapy. Cancer, 109(1), 146–156.
Jara, H. (2013). Theory of Quantitative Magnetic Resonance Imaging. World Scientific Publishing Co Inc.
Jernigan, T. L., Archibald, S. L., Fennema-Notestine, C., Gamst, A. C., Stout, J. C., Bonner, J., & Hesselink, J. R. (2001). Effects of age on tissues and regions of the cerebrum and cerebellum. Neurobiology of Aging,
