Cognitive and neuroimaging outcomes in individuals with benign and low-grade brain tumours receiving radiotherapy: a protocol for a prospective cohort study

Abstract

Introduction

Radiation-induced cognitive decline (RICD) occurs in 50%–90% of adult patients 6 months post-treatment. In patients with low-grade and benign tumours with long expected survival, this is of paramount importance. Despite advances in radiation therapy (RT) treatment delivery, better understanding of structures important for RICD is necessary to improve cognitive outcomes. We hypothesise that RT may affect network topology and microstructural integrity on MRI prior to any gross anatomical or apparent cognitive changes. In this longitudinal cohort study, we aim to determine the effects of RT on brain structural and functional integrity and cognition.

Methods and analysis

This study will enroll patients with benign and low-grade brain tumours receiving partial brain radiotherapy. Patients will receive either hypofractionated (>2 Gy/fraction) or conventionally fractionated (1.8–2 Gy/fraction) RT. All participants will be followed for 12 months, with MRIs conducted pre-RT and 6-month and 12 month post-RT, along with a battery of neurocognitive tests and questionnaires. The study was initiated in late 2018 and will continue enrolling through 2024 with final follow-ups completing in 2025. The neurocognitive battery assesses visual and verbal memory, attention, executive function, processing speed and emotional cognition. MRI protocols incorporate diffusion tensor imaging and resting state fMRI to assess structural connectivity and functional connectivity, respectively. We will estimate the association between radiation dose, imaging metrics and cognitive outcomes.

Ethics and dissemination

This study has been approved by the Research Subjects Review Board at the University of Rochester (STUDY00001512: Cognitive changes in patients receiving partial brain radiation). All results will be published in peer-reviewed journals and at scientific conferences.

Trial registration number

ClinicalTrials.gov NCT04390906.

STRENGTHS AND LIMITATIONS OF THIS STUDY.

• Limiting the study to participants with only benign and low-grade brain tumours helps mitigate confounding factors such as variation in tumour biology and normal tissue infiltration.

• Prospective design with baseline evaluation prior to radiation allows capture of longitudinal changes in imaging and cognitive outcomes.

• Use of open-source software assures transparency, reproducibility and implementation of the proposed protocol by other investigators.

• Inclusion of patients receiving hypofractionated radiation, which has increasingly been used in benign and low-grade brain tumours, is highly important since much of the data focuses on conventionally fractionated radiation.

• Heterogeneity of patient population including tumour type, size and location, radiation techniques, patient clinical factors including age, other cancer treatments including chemotherapy and surgery is a significant limitation; however, these factors will be adjusted for in analysis and improve generalisability of results.

Introduction

Rationale and evidence gaps

Cognitive impairment in patients with brain tumours has a major impact on quality of life and on the ability to function at work and in daily life.^1–4 Deficits manifest clinically as impairments in multiple cognitive domains including memory, attention and executive function.^{5 6} The aetiology is often multifactorial; contributing factors may include anxiety and/or depression, tumour location and pathology, comorbidities and age, as well as effects from treatment (chemotherapy, surgery and/or radiation therapy (RT)⁷). Notably, radiation-induced cognitive decline (RICD) is observed in more than 30% of patients at 4 months after partial or whole brain RT and in more than 50% at 6 months.⁸ RICD is particularly important in patients with low-grade and benign tumours who are expected to have long-term survivals. In these patients, treatment selection to maximise quality of life and minimise cognitive deficits is imperative. Considerable efforts have been directed toward understanding and preventing RICD, an important late effect of RT.^{5 9 10} To date, multiple mechanisms underlying RICD have been elucidated, including damage to sites of neurogenesis,^{11 12} neuroinflammation,^{13 14} neuronal dysfunction¹⁵ and vascular changes.^16–18

RICD can occur in the absence of any gross anatomical changes. Advanced MRI techniques, however, may be able to detect effects from RT early on and may help elucidate mechanisms of radiation damage in RICD.¹⁹ MRI can examine volumetric and connectivity changes (both functional and structural) as well as changes in brain vasculature and perfusion. MRI may ultimately provide tools to identify patients at risk for RICD and help to direct efforts to prevent or ameliorate cognitive decline. Accordingly, accurate modelling of neurocognitive function with neuropsychological tests and correlation with in vivo imaging findings may help to identify putative biomarkers for routine quantitative evaluation of cognitive changes in patients with RICD. Novel MRI biomarkers of RICD are essential to improve understanding of how RT affects the brain structurally and functionally, to identify potential targets and therapeutics to mitigate RICD, and to improve initial RT plans to decrease complication rates.

RT affects both grey and white matter structures, yet the functional implications of these changes are actively being investigated. Studies have shown that cranial RT is associated with dose-dependent atrophy of the cortex,²⁰ hippocampus²¹ and amygdala²² on T1-weighted (T1w) MRI. The hippocampus in particular has garnered attention as a vulnerable structure in the setting of RT; where RT has been shown to reduce neurogenesis^{23 24} and the pool of neural stem cells in the dentate gyrus.^{25 26} Additionally, radiation dose to the hippocampus has also been shown to predict Hopkins Verbal Learning Test scores after brain irradiation.²⁷ Currently, the hippocampus is the only intracranial structure for which validated dose constraints are used in standard treatment planning.^28–31 NRG Oncology CC001 showed that conformal avoidance of the bilateral hippocampi (important structures in learning and memory) during whole brain RT reduced the risk of cognitive decline at 6 months from 68.2% to 59.5%.³² Despite advances in understanding of the role of the hippocampus in RICD, however, nearly 60% of patients still experience diminished cognitive function after RT despite conformal avoidance of the hippocampus. Moreover, recent studies have shown that radiation dose to the corpus callosum and surrounding white matter tracts can impact attention and processing speed at 6 months post-RT,³³ executive function with radiation damage to the anterior cingulate cortex,³⁴ damage to perisylvian white matter can predict language dysfunction,³⁵ and damage to the hippocampus, temporal pole and entorhinal cortex can predict changes in visuospatial memory.³⁶ Thus, while the hippocampus is undoubtedly an important structure in memory formation, the singular focus on this region likely belies the complexity of structures and networks involved in memory formation and ignores the contribution of other anatomic structures to cognitive deficits seen post-therapy.

Novelty and innovation

While there have been some strides made in understanding RICD, there remain significant gaps in our knowledge, which our study hopes to address. These include applications of rs-fMRI in prediction of RICD, evaluation of cognitive outcomes after hypofractionated radiation for low-grade and benign brain tumours, evaluation of novel areas of interest that could contribute to cognitive decline, and integration of established autosegmentation software such as Freesurfer with radiation dose information.

Whole brain networks can be evaluated by analysing structural and functional connections within the brain and their connections to function and behaviour.³⁷ Structural connectivity (SC) in the brain is measured by tracing white matter tracts derived from diffusion tensor imaging (DTI).³⁸ DTI evaluates the direction and magnitude of water molecular diffusion in a three-dimensional space (diffusion tensor) and can provide information on anisotropic diffusion. Additional quantitative metrics such as the fractional anisotropy (FA), axial diffusivity (AD), mean diffusivity (MD) and radial diffusivity (RD) can be obtained from DTI and can help describe different disease states such as demyelination.^{39 40} Studies using DTI have shown that RT results in atrophy, demyelination of white matter and gliosis⁴¹ particularly in patients with a history of demyelinating diseases.⁴² Partial brain RT has also been shown to result in decreased AD and increased RD within the parahippocampal cingulum, where these changes are correlated with declines in verbal memory and fluency.⁴³ Notably, reconstruction of fibre tracts in brain tumours and surrounding tissues is confounded by false continuities within the tumour and surrounding oedema.⁴⁴ Accordingly, advanced diffusion methods have recently been developed to model and eliminate free water with single-shell diffusion weighted imaging data, to more accurately model the tissue microstructure of surrounding normal brain tissue.⁴⁵ However, similar to gross volumetric changes, apparent evidence of white matter atrophy and demyelination may not be discernable prior to 6 months or 1-year post-RT.^{46 47} Resting state functional MRI (rs-fMRI) can be used to evaluate functional connectivity (FC). In particular, graph-theory analysis of FC has revealed topological organisation of brain networks,⁴⁸ which has been used to investigate how network topology is affected in development,⁴⁹ ageing⁵⁰ and pathology.^51–53 Graph theory-based approaches treat the brain as a network of nodes and edges, where nodes can be a region of interest (ROI) or a single voxel. Edges are the connections between each node. These graphical relationships can then be modelled as a correlation matrix, in which cross correlation is performed to determine the strength between pairs of nodes. Analysis of these matrices has revealed the brain to be highly modular,^{54 55} with specific network hubs (areas of many connections to other nodes),⁵⁶ which have been shown to change in the setting of pathology and RT.⁵⁷ Nevertheless, it is not known whether functional network changes can predict RICD or precede structural changes on MRI.

There has been limited evaluation of whether rs-fMRI can be used to predict early radiation changes. This may be partially due to difficulty in using rs-fMRI in high-grade glioma.³³ Other studies have consistently demonstrated that IDH wild type gliomas (ie, gliomas with more aggressive histology) have greater impact on FC metrics such as global FC derived from rs-fMRI⁵⁸ as well as impact baseline cognitive status to a greater degree prior to any treatment.⁵⁹ When limited to select patients, this modality may be useful as an MRI biomarker in early grade and benign brain tumour patients receiving radiation. Further studies focused on that population and excluding high-grade glioma patients, such as this one, are needed. Whole brain metrics such as FC may provide early identification of participants who are at risk of decline and can be targeted with novel therapeutics, either by using advanced RT techniques to improve RT plans or use of radioprotective pharmaceuticals. Preliminary studies are limited but suggest that rs-fMRI and FC represent a promising modality with which to develop dose constraints and mitigate cognitive decline after RT.^60–62

Study aims

Investigation into structures outside the hippocampus that can be spared in order to improve cognitive outcomes remains an area of active study. We have the most data for RICD related to radiation dose to the corpus callosum^{33 63} and hippocampus.^{31 32} However, we currently have no valid dose constraints for the structures outside the hippocampus including the corpus callosum. Development of dose constraints and investigation of which structures can be avoided and lead to improvement in clinical outcomes is an area of active research as we seek to understand the complex structural and functional relationships that lead to RICD.⁶⁴

Additionally, radiosurgery and fractionated radiosurgery are important modalities used frequently in the treatment of benign brain tumours. With the increased use of hypofractionation and radiosurgery, it is important to establish dose constraints that are valid in the setting of high dose per fraction.⁶⁵ As of now, we have little data on dose constraints for cognitive avoidance structures in the setting of hypofractionation, and much is extrapolated from studies of conventional fractionation.

Accordingly, this study will evaluate the effects of RT in patients with benign and low-grade brain tumours using multimodal neuroimaging and a battery of neurocognitive tests. We hypothesise that radiation induced damage will manifest prior to gross anatomical changes via alterations in network topology and microstructural integrity. Ultimately, we aim to establish structures beyond the hippocampus that are vulnerable to RT and develop dose constraints to minimise the risk and progression of RICD in this vulnerable patient population.

Methods and analysis

Study design

A total of 75 patients with benign and low-grade brain tumours planned to receive partial brain RT, either hypofractionated (>2 Gy/fraction) or conventionally fractionated (1.8–2 Gy/fraction), will be enrolled at the Wilmot Cancer Institute. All participants provide written informed consent according to the Institutional Review Board (IRB) approved protocol prior to any evaluation. Participants are followed for 12 months.

Key inclusion criteria include (1) age ≥18 years; (2) patients with benign or low-grade brain tumours including grade 2 IDH-mutant astrocytoma, grade 2 oligodendroglioma, grade 1 and 2 meningiomas, vestibular schwannomas, pituitary adenomas, craniopharyngiomas, haemangiopericytomas or other benign or low-grade brain tumours; (3) planned to receive either conventional or hypofractionated RT and (4) no contraindication to gadolinium-enhanced MRI. Surgical excision and/or chemotherapy prior to enrolment is permitted.

Key exclusion criteria include: (1) prior cranial RT; (2) inability to participate in neurocognitive testing; (3) intractable seizures; (4) non-English-speaking and (5) aphasia limiting ability to participate in neurocognitive testing.

Participants will undergo three comprehensive evaluations (baseline, 6-month and 12-month time points) that include clinical evaluation, MRI, a battery of neurocognitive tests, and questionnaires which evaluate patient-reported cognition, fatigue, anxiety and depression. An additional 3-month time point includes questionnaires and neurocognitive testing only (figure 1).

Figure 1.

Study schema.

Patient and public involvement

Patients will be involved in the design and conduct of this research as follows: After completion of this study, we will plan to further tailor the study design for a larger study by conducting interviews with participants. Once the results have been published, participants will be informed via email.

Neurocognitive testing

Assessments of neurocognitive and functional performance are performed to evaluate neurocognitive changes post-RT. The components of the neurocognitive testing battery are described in table 1. The battery includes the standard tests recommended by the International Cognition and Cancer Task Force⁶⁶ testing verbal memory (Hopkins Verbal Learning Test-Revised, HVLT-R),^{67 68} verbal fluency (Controlled Oral Word Associated Test)⁶⁹ and executive function (Trail Making Test).^{70 71} However, the battery additionally includes the Brief Visuospatial Memory Test-Revised, which has a similar format to the HVLT-R but focuses on visuospatial learning and memory,³⁶ as well as additional iPad-based tests from Cambridge Cognition which have been shown to be valid and sensitive in the assessment of cancer-related cognitive impairment.^72–74 Neuropsychological testing is administered by trained study coordinators using a standardised testing manual; study coordinators are supervised by the study team with expertise in neurology, neuropsychology and cognitive science. Raw scores will be used in analysis and adjusted for covariates such as age. Testing is performed in a quiet, comfortable room without distractions.

Table 1.

Description and platform of tests in cognitive battery

Test	Description	Cognitive domain	Platform
Wide Range Achievement Test-4	Word reading	Cognitive reserve, education level	Paper based
Hopkins Verbal Learning Test-Revised	Immediate and delayed recall of a word list	Verbal learning and memory	Paper based
Controlled Oral Word Associated Test	Number of words the participant can provide in a category over 1 min	Verbal fluency	Paper based
Trail Making Test A and B (TMT-A and TMT-B)	Connect circles containing a series of numbers (A) or numbers and letters (B) in a pattern	Executive function	Paper based
Brief Visuospatial Memory Test-Revised	Immediate and delayed recall of a series of shapes and designs	Visuospatial learning and memory	Paper based
Emotional Recognition Task	Identification of the emotion indicated by a facial expression	Emotional and social cognition	Cambridge cognition
Spatial Working Memory	Use of strategy to find a yellow token behind coloured boxes	Executive function, visuospatial working memory	Cambridge cognition
Paired Associates Learning	Match the pattern to the box where it was previously displayed	Visuospatial episodic memory and new learning	Cambridge cognition
Delayed Matching to Sample	Matching of complex visual patterns	Visual matching ability and short-term visual recognition memory	Cambridge cognition
Reaction Time Task	Select a circle in which a yellow dot appears	Assessment of motor and mental response speed	Cambridge cognition

Patient-reported outcomes

Patient-reported outcome measures of symptoms that may influence cognition are recorded longitudinally so that they can be studied and accounted for in analyses. These symptoms include fatigue, anxiety and depression using validated measures including the Functional Assessment of Cancer Therapy-Brain (FACT-Br, neurological symptoms in brain tumour patients,⁷⁵ Functional Assessment of Chronic Illness Therapy-Fatigue), symptoms of fatigue,⁷⁶ Patient Health Questionnaire-9, symptoms of depression,⁷⁷ State-Trait Anxiety Inventory⁷⁸ and Short Form of the Profile of Mood States-2, subscales of anger-hostility, confusion-bewilderment, depression-dejection, fatigue-inertia, tension-anxiety, vigor-activity and friendliness.⁷⁹ Subjective cognition is measured using the FACT-Cognition (FACT-Cog),⁸⁰ for comparison with scores on objective neurocognitive testing.

Demographic and clinical information

Patient characteristics that may affect cognitive outcomes and trajectories are recorded, including age, education level, comorbidities including diabetes, hypertension, autoimmune disease, tumour hemisphere, tumour site, tumour pathology, prior surgeries, employment status, smoking status, alcohol use, sex/gender, hypopituitarism, menopausal status, steroid use, use of medications that can affect cognition and mood, and exposure to chemotherapy.

Radiotherapy planning

Each patient is planned and treated per standard of care by their treating radiation oncologist. RT plans for single fraction or fractionated radiosurgery are created using BrainLAB Elements planning software. All other plans were created using Varian Eclipse treatment planning software. For consistency, all radiation dose maps were calculated in Eclipse for all patients using a 1 mm × 1 mm grid.

MRI acquisition

All imaging is performed on a 3T GE Discovery 750 MRI system (Milwaukee, Wisconsin, USA), equipped with an 8-channel head coil. High-resolution T1w anatomical images are acquired using a 3D BRAVO FSPGR sequence with the following parameters: repetition time (TR)=8.2 ms, echo time (TE)=3.2 ms, field of view (FOV) 256 mm², resolution 1×1×1 mm³.

Blood oxygen-level dependent (BOLD) rs-fMRI is acquired using a BOLD sensitive gradient-echo echo planar imaging (EPI) sequence with the following parameters: TR=2000 ms, TE=30 ms, FOV=192 mm², resolution 3×3×3 mm³, 150 volumes.

In order to evaluate white matter integrity and microstructural changes, we make use of a standard clinical DTI protocol with 30 diffusion directions, demonstrated to be sufficient for reconstructing white matter fibre tracts and not to affect test–retest reliability.⁸¹ DTI is acquired using a two-dimensional axial single-shot dual spin-echo EPI sequence with the following parameters: TR=10 000 ms, TE=81 ms, FOV=256 mm², resolution 2×2×2 mm³, 30 diffusion weighted directions with b=1000 s/mm² and 4 b=0 reference images.

MRI data processing

Here, we describe a comprehensive image analysis pipeline, which includes preprocessing, data cleaning and postprocessing for each imaging modality, including advanced modelling and calculation of quantitative imaging biomarkers (figure 2).

Figure 2.

MRI data processing pipeline. AD, axial diffusivity; dMRI, diffusion MRI; DTI, diffusion tensor imaging; FA, fractional anisotropy; FC, functional connectivity; FEAT, FMRI Expert Analysis Tool; FIX, FMRIB’s ICA-based Xnoisefier; GTV, Gross Target Volume; MD, mean diffusivity; MELODIC, Multivariate Exploratory Linear Optimised Decomposition into Independent Components; RD, radial diffusivity; rs-fMRI, resting state functional MRI; RT, radiotherapy; SC, structural connectivity.

All image processing is completed within URMC servers in the Centre for Integrated Research Computing, using BHWARD, a HIPAA compliant server.

RT dose calculations

The RT dose map is first scaled and mapped with CT images using pydicom (V.1.4). T1w images are registered to patient-specific CT space using FMRIB’s Linear Image Registration Tool (FLIRT).⁸² Patient-specific parcellations derived from the Desikan-Killiany⁸³ atlas using Freesurfer are registered with the RT dose map. The mean, maximum and minimum RT doses are extracted from each ROI, and the 2 Gy/fraction equivalent dose (EQD2) is calculated using the linear quadratic model,⁸⁴ with an ${\displaystyle \alpha /\beta }$ equal to 3, to model the radiosensitivity of normal brain tissue.⁸⁵

T1 weighted

T1w images are processed by first masking out the tumour using the gross target volume (GTV) as contoured by the primary radiation oncologist from the RT structure set,⁸⁶ delineated on planar MRI and CT imaging, using nibabel (https://nipy.org, V.3.1.1). This is achieved by mapping the GTV to the CT images in patient space, and then performing an affine transform to register the T1-w images to the patient specific CT images (figure 3). Thereafter, segmentation is performed using the tumour masked T1-w in Freesurfer (V.6.0.0, http://surfer.nmr.harvard.edu). The Freesurfer pipeline is run independently for each participant at each time point (pre-RT, 6 months and 12 months post-RT). Briefly, processing includes skull-stripping and removal of non-brain tissue, motion correction, intensity normalisation, automated Talairach transformation, white matter segmentation and cortical parcellation using the Desikan-Killiany atlas,⁸³ which includes cortical and subcortical ROIs. Subsequent segmentation of thalamic nuclei are also performed using a probabilistic atlas based on ex vivo MRI and histology.⁸⁷ ROIs will include whole brain grey and white matter, cerebral hemispheres and subcortical grey matter (hippocampus, amygdala, caudate, putamen, thalamus, nucleus basalis of Meynert), as well as white matter tracts including cingulum, fornix, parahippocampal white matter and corpus callosum.

Figure 3.

Representative images from participant with vestibular schwannoma. (A) RT dose map from RT structure set, mapped to CT image and scaled. (B) T1w structural image coregistered with CT image and RT dose map via affine transformation (yellow circle shows acoustic schwannoma). (C) T1w image with gross target volume (GTV, yellow circle) used to mask tumour prior to processing. (D) Subcortical and cortical structures obtained from brain parcellation, with vestibular schwannoma excluded (yellow arrow). RT, radiation therapy.

Diffusion imaging

DTI data are preprocessed using FMRIB’s Software Library (FSL,^88–91 diffusion toolbox (FDT, http://fsl.fmrib.ox.ac.uk/fsl). Briefly, intervolume patient motion, brain extraction using BET⁹¹ and eddy-current induced distortion correction are performed using EDDY.⁹² The diffusion tensors are then fit on eddy-corrected data using DTIFIT.⁹³ DTIFIT fits a diffusion tensor model at each voxel and provides the three principal eigenvectors and eigenvalues of the diffusion tensor, from which the FA, MD, AD and RD can be measured.⁹³ All images are then registered to MNI standard space and interpolated to 1 mm³ voxels. Binary GTV masks are then used to mask out abnormal tissue in processed maps. The JHU white-matter tractography atlas, composed of 20 structures identified using probabilistic tractography, is then used to extract mean FA, MD, AD and RD values from ROI.⁹⁴

Functional imaging

rs-fMRI data are processed using FSL’s FMRI Expert Analysis Tool (FEAT, V.6.00).⁹⁵ Registration to high resolution structural space is carried out using a two-stage registration. First rs-fMRI data are registered to high-resolution structural space using FMRIB’s Linear Image Registration Tool (FLIRT),⁸² and then registration to MNI standard space⁹⁶ is further refined using FMRIB’s Nonlinear Image Registration Tool (FNIRT).⁹⁷ Since field maps were not acquired as part of the clinical scan protocol, a 12 df affine transformation was used for linear and nonlinear registration. All registrations are visually inspected during processing. Further, preprocessing includes skull stripping using BET,⁹¹ motion correction using MCFLIRT,⁸² slice-time correction, spatial smoothing using a Gaussian kernel of FWHM 5 mm and high-pass temporal filtering.

Single-session independent component analysis (ICA) is then performed for each participant using probabilistic ICA implemented in FSL’s Multivariate Exploratory Linear Optimised Decomposition into Independent Components (MELODIC, V.3.15). MELODIC decomposes input data into separate time courses and spatial maps using probabilistic principal component analysis. FMRIB’s ICA-based Xnoisifier (FIX)^{98 99} is then used to further denoise functional data by automatically classifying signal versus noise components from the time series data. FIX is run using the standard pretrained data (TR=3 s, 3.5×3.5×3.5 mm resolution, 6 min) which was preprocessed using default FEAT processing. All images and components are visually inspected for accuracy prior to further processing.

Once all fMRI data have been preprocessed, the denoised functional data are used to construct participant specific FC matrices. Participant-specific atlases generated using Freesurfer are then registered to the MNI standard space and used to extract the mean time series from each ROI. This is done to ensure that only functional regions outside of the tumour are used to construct FC matrices. FC matrices are then generated by computing the cross correlation between all pairs of nodes (ROIs), using the Pearson correlation coefficient (figure 4).

Figure 4.

Representative functional connectivity correlation matrices. Matrices are computed using the Pearson correlation coefficient between every time course for all pairs of nodes. Matrices are thresholded at 0.5 and normalised. The average of all patient specific correlation matrices at baseline (A) and 6 months post-RT (B). The colour bar represents the normalised correlation coefficient between pairs of nodes. RT, radiation therapy.

Participant-specific FC matrices are then analysed using the Brain Connectivity Toolbox¹⁰⁰ in MATLAB (R2020a). Participant-specific correlation matrices are thresholded to yield weighted undirected networks, and analysed using graph theory to yield measures of functional integration and segregation. Global measures of integration, including global efficiency, transitivity and modularity, are then computed for further statistical analysis.¹⁰¹ Local measures of segregation, including clustering coefficient and local efficiency (table 2), are also computed for each ROI for further statistical analysis. A more thorough review of graph theory-based measures for rs-fMRI may be found in Rubinov and Sporns.¹⁰²

Table 2.

Overview of graph theory measures used to analysis resting state functional connectivity and structural connectivity obtained from diffusion tractography

Measurement	Definition	Equation
Node degree	The no of connections between one node and the rest of the network	${\displaystyle k_i=\mathrm{\Sigma }\left(a_{ij}\right)}$
Clustering coefficient	The no of connections between the neighbours of a node	${\displaystyle 2t/\left(k\left(k-1\right)\right);}$ k is the node degree; t is the fraction of triangles around a node.
Efficiency	Inverse of path length (minimum no of edges to traverse from one node to another)	${\displaystyle \frac{1}{N\left(N-1\right)}\mathrm{\Sigma }\left(\frac{1}{d_{ij}}\right)}$ Where ${\displaystyle d_{ij}}$ is the shortest path length between nodes i and j
Modularity	Areas of highly interconnected nodes, with few connections to nodes in other modules	${\displaystyle Q=\frac{1}{l}\mathrm{\Sigma }\left(a_{ij}-\frac{\left(k_i{k}_j\right)}{l}\right)\delta \left(m_i,m_j\right)}$

A more in depth review of graph theory and graph theory measures can be found in Rubinov and Sporns.¹⁰²

We estimate that 52 evaluable patients (a total of 75 participants allowing for 30% of participants with missing or incomplete data) will have ≥80% power to detect at least 0.4 SD change on the delayed recall measure of the HVLT-R post-RT. The power analysis is based on a paired t-test with a two-sided significance level of 0.05.

Analysis plan

Graphical methods will be used to explore the cognitive test and imaging data, to visually describe and compare distributions of continuous variables, and to visualise results of statistical analyses. Quantitative imaging metrics (cortical thickness, subcortical volume, FA, MD, AD, RD, local efficiency and clustering coefficient) will be analysed to investigate their relationships with RT dose and cognitive measures. Comparisons between raw scores on cognitive tests and imaging metrics pre-RT and post-RT will be performed using paired t-tests and Wilcoxon signed-rank tests. Pearson and Spearman correlation analyses will be used to assess associations between pairs of continuous measures. Multivariate mixed effect regression models will be used to evaluate the relationships of cognitive tests at 6-month and 12-month visits with RT dose to ROIs known to be instrumental in the specific cognitive domain adjusting for the baseline cognitive test, imaging parameters, age, gender, tumour laterality and tumour type. During the analyses, false discovery rate method will be used to account for multiple comparisons.¹⁰³ All statistical analyses will be performed using R (R Foundation for Statistical Computing, Vienna, Austria) or SAS V.9.4.

RICD is an important target of efforts to use more sophisticated radiation techniques such as intensity modulated RT and proton therapy in order to decrease side effects.¹⁰⁴ While validated dose constraints exist for structures such as the brainstem, cochlea, optic nerves and chiasm, and pituitary gland,²⁸ development of dose constraints for intracranial structures involved in cognition is a new and exciting area of research that promises to improve radiation outcomes. Despite conformal dose reduction to the hippocampi, RICD occurs in a large percentage of patients, reflecting the complexity of memory formation and the need to identify non-hippocampal structures involved in higher cognitive functions. The pathology underlying RICD likely begins prior to any gross anatomical changes or noticeable differences in cognition observed by the patient. Accordingly, development of quantitative in vivo biomarkers is essential for developing dose constraints and monitoring RICD.

This study uses conventional and advanced MRI, neurocognitive testing and dosimetry information to provide a comprehensive description of RICD in patients with brain tumours receiving RT. The proposed analyses will provide insight into which intracranial structures are particularly susceptible to RT and how they modulate changes in cognition via aberrant network topology. The results of this study will help to provide dose constraints to better avoid cognitive decline that can ultimately be used to create radiation plans associated with less cognitive change. Incorporation of rs-fMRI into treatment planning and monitoring has the potential to improve cognitive outcomes in the setting of RT and provide personalised treatment. Additionally, utilisation of graph theory will be able to identify specific nodes and hubs within brain networks that are susceptible to RT at the population and individual level.¹⁰⁵

This protocol and analysis pipeline will aid researchers interested in combining MRI data including segmentation of intracranial structures not used in standard radiation planning with radiation dosimetry information to advance our understanding of RICD. We provide detailed information on study design, clinical and imaging protocols and analysis pipeline with which to investigate RT-induced cognitive changes on intracranial structures that are not segmented with standard radiation planning software. This is an important and complex process which should be transparent, and one of our goals with this paper is to promote utilisation of open software packages in a useful and standardised way for the radiation oncology community.

Ethics and dissemination

Studies involving human participants are reviewed and approved by the Research Subjects Review Board at the University of Rochester. All patients/participants provide their written informed consent to participate in this study. All results will be presented at relevant conferences and published in relevant peer-reviewed journals.

Ethics statements

Patient consent for publication

Not applicable.