Effects of Hippocampal Sparing Radiotherapy on Brain Microstructure—A Diffusion Tensor Imaging Analysis

Hippocampal-sparing radiotherapy (HSR) is a promising approach to alleviate cognitive side effects following cranial radiotherapy. Microstructural brain changes after irradiation have been demonstrated using Diffusion Tensor Imaging (DTI). However, evidence is conflicting for certain parameters and anatomic structures. This study examines the effects of radiation on white matter and hippocampal microstructure using DTI and evaluates whether these may be mitigated using HSR. A total of 35 tumor patients undergoing a prospective randomized controlled trial receiving either conventional or HSR underwent DTI before as well as 6, 12, 18, 24, and 30 (±3) months after radiotherapy. Fractional Anisotropy (FA), Mean Diffusivity (MD), Axial Diffusivity (AD), and Radial Diffusivity (RD) were measured in the hippocampus (CA), temporal, and frontal lobe white matter (TL, FL), and corpus callosum (CC). Longitudinal analysis was performed using linear mixed models. Analysis of the entire patient collective demonstrated an overall FACC decrease and RDCC increase compared to baseline in all follow-ups; ADCC decreased after 6 months, and MDCC increased after 12 months (p ≤ 0.001, 0.001, 0.007, 0.018). ADTL decreased after 24 and 30 months (p ≤ 0.004, 0.009). Hippocampal FA increased after 6 and 12 months, driven by a distinct increase in ADCA and MDCA, with RDCA not increasing until 30 months after radiotherapy (p ≤ 0.011, 0.039, 0.005, 0.040, 0.019). Mean radiation dose correlated positively with hippocampal FA (p < 0.001). These findings may indicate complex pathophysiological changes in cerebral microstructures after radiation, insufficiently explained by conventional DTI models. Hippocampal microstructure differed between patients undergoing HSR and conventional cranial radiotherapy after 6 months with a higher ADCA in the HSR subgroup (p ≤ 0.034).


Introduction
Although a crucial pillar of modern tumor therapy, improving morbidity and mortality, cranial radiotherapy (CR) can be associated with severe adverse reactions. While symptoms of acute (during CR) and early delayed (weeks to a few months after CR) radiation reactions are usually mild, self-limiting, and/or responsive to corticoid treatment, late-delayed reactions (several months to years after CR) can be progressive and irreversible [1]. Along Practice guidelines. It was registered by the University of Erlangen-Nürnberg Medical School as a clinical trial (Identifier NCT01849484) on 8 May 2013 and is ongoing at the time of publication of this article. Study participants provided their written informed consent. Diagnosis, recruitment, and radiation planning were conducted by the Department of Radiation Oncology at our clinic. Briefly, HIPPO-SPARE 01 is a prospective randomized controlled trial (RCT) in which study participants were randomly assigned to two groups, one receiving CCR, and the other HSR. Inclusion criteria were a minimum age of 18 years, diseases indicating neurocranial radiotherapy confirmed by histology or imaging ([skull base] meningioma, pituitary adenoma, craniopharyngioma, or brain metastases), and Karnofsky-State ≥ 50%. Exclusion criteria included, among others, persistent drug and/or alcohol abuse, prior neurocranial radiotherapy, more than 3 brain metastases, and gross tumor volume in the hippocampal region or hippocampal avoidance zone.
The aim of this sub-investigation was to evaluate the potential of DTI to detect microstructural changes in the brain following CR and to evaluate whether these effects can be mitigated using HSR.

Follow-Up and Imaging Protocol
Follow-up intervals of 6 months (±3 months) after completion of CR were chosen (Figure 1). At each interval, a standardized MRI protocol, including DTI, was conducted. Primary outcome measures were the following DTI Parameters: Fractional Anisotropy (FA), Mean Diffusivity (MD), Axial Diffusivity (AD), and Radial Diffusivity (RD).
Brain Sci. 2022, 12, x FOR PEER REVIEW 3 Medical Association (Declaration of Helsinki) and adhered to national Good Clinical tice guidelines. It was registered by the University of Erlangen-Nürnberg Medical S as a clinical trial (Identifier NCT01849484) on 8 May 2013 and is ongoing at the tim publication of this article. Study participants provided their written informed con Diagnosis, recruitment, and radiation planning were conducted by the Department o diation Oncology at our clinic. Briefly, HIPPO-SPARE 01 is a prospective random controlled trial (RCT) in which study participants were randomly assigned to two gr one receiving CCR, and the other HSR. Inclusion criteria were a minimum age of 18 y diseases indicating neurocranial radiotherapy confirmed by histology or imaging ( base] meningioma, pituitary adenoma, craniopharyngioma, or brain metastases) Karnofsky-State ≥ 50%. Exclusion criteria included, among others, persistent drug a alcohol abuse, prior neurocranial radiotherapy, more than 3 brain metastases, and tumor volume in the hippocampal region or hippocampal avoidance zone. The aim of this sub-investigation was to evaluate the potential of DTI to detec crostructural changes in the brain following CR and to evaluate whether these effect be mitigated using HSR.

Follow-Up and Imaging Protocol
Follow-up intervals of 6 months (±3 months) after completion of CR were ch ( Figure 1). At each interval, a standardized MRI protocol, including DTI, was condu Primary outcome measures were the following DTI Parameters: Fractional Aniso (FA), Mean Diffusivity (MD), Axial Diffusivity (AD), and Radial Diffusivity (RD). All MRI exams were performed using a 1.5 T Siemens Magnetom Aera Scanner a dedicated 20-channel head/neck coil. High-resolution MRI scans of the brain were formed with a gradient field strength up to 45 mT/m (at 200 T/m/s). DTI was perfo in the axial plane with 2 mm isotropic resolution using a single-shot, spin-echo, echo nar imaging (EPI) diffusion tensor sequence (TR = 10500 ms, TE = 93 ms, readout b width = 1628 Hz/pixel, FoV 256 × 256 mm 2 , acquisition matrix size 128 × 128, and sp fat saturation). Diffusion weighting was carried out with a maximal b-factor of 1000 s along 20 icosahedral directions complemented by one scan with b = 0. All DTI mea ments were repeated with reversed-phase encoding to control for echo-planar im distortions during the postprocessing.
MRI protocol also included the acquisition of a 3-dimensional (3D) T1-MPRAG quence with 1 mm isotropic resolution (TR = 1900 ms, TE 3.0 ms, FoV = 250, acqui matrix size = 256 × 256). All MRI exams were performed using a 1.5 T Siemens Magnetom Aera Scanner with a dedicated 20-channel head/neck coil. High-resolution MRI scans of the brain were performed with a gradient field strength up to 45 mT/m (at 200 T/m/s). DTI was performed in the axial plane with 2 mm isotropic resolution using a single-shot, spin-echo, echo-planar imaging (EPI) diffusion tensor sequence (TR = 10500 ms, TE = 93 ms, readout bandwidth = 1628 Hz/pixel, FoV 256 × 256 mm 2 , acquisition matrix size 128 × 128, and spectral fat saturation). Diffusion weighting was carried out with a maximal b-factor of 1000 s/mm 2 along 20 icosahedral directions complemented by one scan with b = 0. All DTI measurements were repeated with reversed-phase encoding to control for echo-planar imaging distortions during the postprocessing.

Processing and Measurement
Diffusion data was collected in the original and reversed phase-encoding direction, resulting in an image pair with distortions in opposite directions. DICOM images were converted to NIfTI Files (Neuroimaging Informatics Technology Initiative) using dcm2nii from the MRIcron package [35]. The susceptibility-induced off-resonance field of the image pair was estimated and applied to the co-registered image stack using a method similar to that described by Andersson et al. using the FSL TOPUP tool [36,37]. After correction for Brain Sci. 2022, 12, 879 4 of 16 eddy-current artifacts, image data was reconverted to DICOM for further processing. The Software Olea Sphere fit a diffusion tensor model for each voxel. The scalar indices FA, MD, AD, and RD, were calculated. Anatomical regions of interest (ROI) were identified by coregistration of DTI and T1-MPRAGE data, as well as manual correction for possible residual motion artifacts to produce an optimum fit to the target area and DTI data sets. Due to its narrowing shape and proximity to the lateral ventricle, hippocampal DTI measurements are highly susceptible to cerebrospinal fluid (CSF) contamination, a momentous partial volume effect occurring in anatomic areas close to CSF [38]. To minimize the influence of this effect, hippocampal ROIs were placed only in their rostrocaudal proportion, where the diameter is largest in axial imaging ( Figure 2). A visual examination was conducted for surgical or pathological defects in the regions of interest, large space-occupying lesions interfering with the ROIs, and poor image quality. Mean values of the scalar indices within the ROIs were taken bilaterally in the hippocampus (CA), temporal and frontal lobe white matter (TL, FL), and genu corporis callosi (CC).

Processing and Measurement
Diffusion data was collected in the original and reversed phase-encoding direction, resulting in an image pair with distortions in opposite directions. DICOM images were converted to NIfTI Files (Neuroimaging Informatics Technology Initiative) using dcm2nii from the MRIcron package [35]. The susceptibility-induced off-resonance field of the image pair was estimated and applied to the co-registered image stack using a method similar to that described by Andersson et al. using the FSL TOPUP tool [36,37]. After correction for eddy-current artifacts, image data was reconverted to DICOM for further processing. The Software Olea Sphere fit a diffusion tensor model for each voxel. The scalar indices FA, MD, AD, and RD, were calculated. Anatomical regions of interest (ROI) were identified by co-registration of DTI and T1-MPRAGE data, as well as manual correction for possible residual motion artifacts to produce an optimum fit to the target area and DTI data sets. Due to its narrowing shape and proximity to the lateral ventricle, hippocampal DTI measurements are highly susceptible to cerebrospinal fluid (CSF) contamination, a momentous partial volume effect occurring in anatomic areas close to CSF [38]. To minimize the influence of this effect, hippocampal ROIs were placed only in their rostrocaudal proportion, where the diameter is largest in axial imaging ( Figure 2). A visual examination was conducted for surgical or pathological defects in the regions of interest, large spaceoccupying lesions interfering with the ROIs, and poor image quality. Mean values of the scalar indices within the ROIs were taken bilaterally in the hippocampus (CA), temporal and frontal lobe white matter (TL, FL), and genu corporis callosi (CC).

Statistics
Statistical analysis was conducted using IBM SPSS (Version 26, SPSS Inc., Chicago, IL, USA). The significance level was set at p ≤ 0.05 (two-sided). Data analyses were of an exploratory manner. Thus, no adjustments for multiple testing were implemented. All measurements between 08/2012 and 02/2017 were included. Thus, no sample size calculation was conducted.

Analysis of Patient Characteristics
Descriptive analyses were conducted in regard to age (years), gender, tumor entity, total radiation dose (Gy), hippocampal dose (Gy), number of fractions, dose per fraction (Gy), duration of radiotherapy (days), and history of chemotherapy, surgery, or biopsy. An Independent t-test was used to test the equality of the aforementioned parameters for both groups.

Analysis of DTI Changes after Cranial Radiotherapy
Mixed linear models (MLMs; maximum likelihood method, autoregressive covariance type) were applied to evaluate all measurements of FA, MD, AD, and RD before CR, as well as in intervals up to 6, 12, 18, 24, and 30 months after CR (±3 months). MLMs use the full data set, replacing missing values by using maximum likelihood estimates. Thus, the data sets of all patients who had received MRI follow-up could be included in the analysis (n = 35).

Analysis of the Effects of Mean Hippocampal Dose on DTI Changes
An analogous mixed linear model approach with the additional inclusion of a hippocampal dose as a covariate was used to analyze its effects on DTI signal changes.

Analysis of the Effects of Hippocampal Sparing on DTI Changes
A further analogous mixed linear model was used to compare the longitudinal DTI measurements of the individual subgroups undergoing CR with and without hippocampal sparing (main effects and interaction effects between time after CCR/HSR).

Patient Characteristics
A total of 82 subjects undergoing cranial radiation were registered for the HIPPO-SPARE 01 trial at the time of analysis. Of these, 47 patients could not be included in the analysis as radiotherapy was incomplete or they had not yet received DTI after radiotherapy. The studies of the remaining 35 patients were examined visually for surgical or pathological defects in the regions of interest, large space-occupying lesions interfering with ROI, and poor image quality, in the process of which no further studies had to be excluded. Infrequent deviations in patient attendance to follow-up MRI reduced the number of data sets available for analysis at the individual time points, as detailed in Table 1.   (20 vs. 15). The majority of tumor entities indicating radiotherapy were meningiomas or pituitary adenomas. Cranial radiation was performed over an average of 41 (±6) days with a mean total dose of 49.3 (±3.4) Gy. The bilateral mean total hippocampal dose was 10.9 Gy (±9.0) and was significantly lower in the HSR subgroup compared to CCR (6.9 ± 4.3 vs. 16.2 ± 10.9; p ≤ 0.006). Both groups did not differ significantly in size, age, gender, history of chemotherapy/surgery/biopsy, or baseline DTI parameters (FA, MD, AD, and RD).

DTI Changes after Cranial Radiotherapy
DTI signal changes in the hippocampus, temporal lobes, frontal lobes, and corpus callosum of all patients over time are depicted in Figure 3. Brain Sci. 2022, 12, x FOR PEER REVIEW 7 of 16 Figure 3. Mean values of fractional anisotropy (FA), axial diffusivity (AD), radial diffusivity (RD), and mean diffusivity (MD) before cranial radiotherapy (CR) as well at prospective intervals of 6 (± 3) months after CR. Baseline measurements (before CR) are represented as "0" on the x-axis. In the corpus callosum, with its longitudinally structured white matter tracts, FA decreased after CR, driven by an increase in RD and a decrease in AD. This effect appears to attenuate over time. Similar, yet only partially significant, tendencies were observed in temporal and frontal lobe white matter.
In the hippocampus, however, driven by an unexpected increase in AD, a distinct increase in FA was observed, followed by a return to near-normal levels after 30 months. Differences compared to baseline: * = p ≤ 0.05, ** = p ≤ 0.01, *** = p ≤ 0.001. Error bars = ±1 standard deviation.
In cerebral white matter, an overall decrease of FACC (p < 0.001) and increase of RDCC (p ≤ 0.001) was observed when comparing measurements before and after CR, which was significant in all individual follow-up measurements when compared to baseline (FA after 6, 12, 18, 24, and 30 months (±3): p < 0.001, p < 0.001, p ≤ 0.009, p ≤ 0.004, p ≤ 0.009; respective , and mean diffusivity (MD) before cranial radiotherapy (CR) as well at prospective intervals of 6 (± 3) months after CR. Baseline measurements (before CR) are represented as "0" on the x-axis. In the corpus callosum, with its longitudinally structured white matter tracts, FA decreased after CR, driven by an increase in RD and a decrease in AD. This effect appears to attenuate over time. Similar, yet only partially significant, tendencies were observed in temporal and frontal lobe white matter. In the hippocampus, however, driven by an unexpected increase in AD, a distinct increase in FA was observed, followed by a return to near-normal levels after 30 months. Differences compared to baseline: * = p ≤ 0.05, ** = p ≤ 0.01, *** = p ≤ 0.001. Error bars = ±1 standard deviation.
In cerebral white matter, an overall decrease of FA CC (p < 0.001) and increase of RD CC (p ≤ 0.001) was observed when comparing measurements before and after CR, which was significant in all individual follow-up measurements when compared to baseline (FA after Brain Sci. 2022, 12, 879 8 of 16 6,12,18,24, and 30 months (±3): p < 0.001, p < 0.001, p ≤ 0.009, p ≤ 0.004, p ≤ 0.009; respective RD: p < 0.001, p < 0.001, p ≤ 0.001, p < 0.001, p < 0.001). AD CC tended to decrease after CR, which was significant after 6 months (p ≤ 0.007). MD CC tended to increase after CR, which was significant 12 months after radiotherapy compared to baseline (p ≤ 0.018) as well as when comparing measurements 12 months and 6 months after CR (p ≤ 0.044). In the temporal lobe white matter, AD decreased overall after CR, which was significant 24 and 30 months after CR compared with baseline (p ≤ 0.004, p ≤ 0.009), as well as when comparing 24 months to 12 months post CR (p ≤ 0.032). MD TL and FA TL showed a decreasing tendency over time, which did not reach significance (e.g., FA TL after 30 months compared to baseline: p ≤ 0.078; MD TL after 24 months: p ≤ 0.116). Similarly, AD showed a decreasing tendency in the frontal lobe, which did not reach significance (e.g., AD FL after 30 months: p ≤ 0.094). No distinct changes were observed in RD TL , AD TL , FA FL , MD FL , or RD FL .
Hippocampal FA had considerably increased 6 months (p ≤ 0.011) as well as 12 months (p ≤ 0.039) after CR when compared to baseline. After the peak increase at 6 months, FA showed a gradual tendency to decrease, with differences to baseline becoming insignificant after 18 months (18, 23, and 30 months compared to baseline: p ≤ 0.086, p ≤ 0.222, p ≤ 0.734). AD CA and MD CA were overall considerably higher after CR (p ≤ 0.005, p ≤ 0.040), with significantly higher AD CA at the individual time points 6, 12, 18, and 30 months after radiation compared to baseline (p ≤ 0.003, 0.015, 0.003 and 0.005) and higher MD CA 6, 18 and 30 months after CR (p ≤ 0.026, 0.020 and 0.008). There was an increasing tendency of RD CA after CR, which was significant 30 months after radiation (p ≤ 0.019).

Effects of Mean Hippocampal Dose on DTI Changes
There was a pronounced positive correlation between mean hippocampal radiation dose and FA CA after radiotherapy (p < 0.001)-FA CA values were 0.001024 higher per Gy of radiation (0.68% of mean baseline FA CA per each Gy of radiation dose). No significant interactions between mean hippocampal dose and hippocampal AD, MD, or RD were found.

Effects of Hippocampal Sparing on DTI Changes
Before radiation, there were no significant differences in hippocampal FA, MD, AD, or RD of subjects in the CCR and HSR subgroups. After CR, AD CA of the HSR subgroup was higher overall (p ≤ 0.034) as well as in the individual follow-up measurements 6 months after radiation (p ≤ 0.025) when compared to the CCR subgroup ( Figure 4). MD CA was higher overall in the HSR subgroup with a tendency to significance (p ≤ 0.072). There were no significant differences between FA CA and RD CA of both groups.

Figure 4. Hippocampal Sparing Radiotherapy (HSR) vs. Conventional Cranial Radiotherapy
Higher hippocampal AD was noted in the HSR subgroup and 6 months after CR. MD was in the HSR subgroup with a tendency to significance (p ≤ 0.072). Further differences were in cant. Differences compared to baseline: * = p ≤ 0.05. Error bars = ±1 standard deviation.

Discussion
To the knowledge of the authors, this is the first study based on a prospectiv domized controlled trial (1) analyzing multiple DTI measurements of the hippoca over an extended period in a comparatively large data set and (2) evaluating HSR in comparison to conventional CR using this technique.
Our patient cohort consisted predominantly of meningioma and pituitary ade patients. Cognitive impairment has been described even after the moderate rad doses used in radiotherapy of benign CNS tumors, tumors of the nasopharynx, prophylactic indications [21,[39][40][41]. Benign entities, however, are associated with confounders than malignancies (i.e., growth dynamic, edema, mass effect, medic This implies particular suitability of the studied cohort for evaluating the cerebral tion response pathophysiology, as well as improved comparability to preclinical mostly derived from healthy individuals [42]. This study focuses on the late-delayed phase of radiation reactions (more th weeks after CR), during which cognitive decline is usually observed. To limit th founding influence of the distinctly different, likely edema-driven, pathophysiol processes underlying acute and early delayed injury, follow-up intervals of 6 mont months) after CR were chosen, thus excluding measurements in the first 3 month CR.

Discussion
To the knowledge of the authors, this is the first study based on a prospective randomized controlled trial (1) analyzing multiple DTI measurements of the hippocampus over an extended period in a comparatively large data set and (2) evaluating HSR in direct comparison to conventional CR using this technique.
Our patient cohort consisted predominantly of meningioma and pituitary adenoma patients. Cognitive impairment has been described even after the moderate radiation doses used in radiotherapy of benign CNS tumors, tumors of the nasopharynx, or in prophylactic indications [21,[39][40][41]. Benign entities, however, are associated with fewer confounders than malignancies (i.e., growth dynamic, edema, mass effect, medication). This implies particular suitability of the studied cohort for evaluating the cerebral radiation response pathophysiology, as well as improved comparability to preclinical data, mostly derived from healthy individuals [42].
This study focuses on the late-delayed phase of radiation reactions (more than 12 weeks after CR), during which cognitive decline is usually observed. To limit the confounding influence of the distinctly different, likely edema-driven, pathophysiological processes underlying acute and early delayed injury, follow-up intervals of 6 months (±3 months) after CR were chosen, thus excluding measurements in the first 3 months after CR.
We detected similar DTI changes in cerebral white matter after CR, especially in the corpus callosum, with a FA decrease and RD increase at all time intervals after CR, an MD increase after 12 months and AD decrease after 6 months, as well as in the temporal lobe with a decrease in AD after 24 and 30 months. The regional variability of DTI changes following CR detected in our patient cohort bears further similarities to previous reports, where most prominent changes have been reported (along with cingulum and fornix) in the corpus callosum, while various other white matter regions had not shown significant alterations (or even opposite parameter movements) [46]. Previous studies proposed topographical variability in radiosensitivity to explain this phenomenon, with a possible predilection for late myelinating neural cells [46,51]. It stands to reason that the pronounced DTI changes measured in certain white matter regions, particularly the corpus callosum, result at least in part from the intrinsically high absolute DTI signal of these regions (Figure 3), likely due to their large proportion of closely and parallelly aligned white matter tracts.
Sparse data are available regarding the radiation response of grey matter and the hippocampus, with conflicting results in preclinical and clinical trials. Clinical studies reported either decreasing or insignificant changes in FA and MD [50,52,53], restricted, however, by small case numbers and lacking measurement of further diffusion parameters. In an investigation of 25 primary brain tumor patients, an increase in hippocampal FA was associated with verbal and memory decline and an increase in hippocampal MD with verbal decline [54]. Preclinical studies of acute to early delayed radiation response reported either constant and decreasing FA values [55][56][57][58], either decreasing or increasing AD and MD [55,56,58], and an increase of RD [55]. One study showed significantly higher FA values in rats 1-9 months after irradiation with 39 Gy compared to controls [59].
In our patient cohort, a significant increase in hippocampal FA was noted 6 and 12 months after CR, followed by a tendency to return to near baseline levels, with differences to baseline becoming insignificant 18 months after CR. The FA increase was dose-dependent (0.68% increase of FA per Gy of radiation). It was driven by a strong and early increase of AD (significant at months 6, 12, 18, and 30 after CR) contrasted by a relatively late and less pronounced increase in RD (significant only 30 months after CR). While DTI changes in white matter may be explained to a large extent by axonal damage and demyelination, the demonstrated hippocampal DTI changes may indicate further, hitherto insufficiently understood, pathomechanisms of radiogenic microstructural change influencing cerebral DTI signal.
Alongside direct DNA damage and apoptosis of glial/neuronal cells, the cerebral radiation reaction consists of a complex interplay of microglial activation, astrocyte proliferation, oligodendrocyte loss and demyelination, stem cell depletion, neuronal receptor alterations, and vascular damage [3,60]. These pathophysiological phenomena are assumed to be linked by the process of chronic neuroinflammation [4,42,60]. Radiogenic cell damage leads to the activation of microglia and astrocytes [61]. Pro-inflammatory cytokines, chemokines, and reactive oxygen/nitrogen species maintain inflammation yet may lead to further oxidative stress and cell damage [62]. Furthermore, cytokines increase the penetrability of the blood-brain and the blood-liquor barrier for immigrating immune cells, leading to an unusually high number of antigen-presenting cells in the brain parenchyma [4]. They also activate local dendritic cells, which migrate to cervical lymph nodes to activate T-cells. Insufficient inhibition of this state of inflammation, i.e., due to inadequate penetration of T-lymphocytes and macrophages limiting inflammation, can lead to chronic neuroinflammation [4]. Preclinical models have shown such persistent glial activation even months after CR [63,64].
The process of extracellular waste clearance from the brain interstitium, termed the "glymphatic system", consists of a directional proton motion along the perivascular spaces, through the interstitium, and into paravenous drainage pathways [65]. MRI models have shown glymphatic system activity throughout the brain, including the hippocampi [66,67]. Other processes associated with chronic neuroinflammation, such as M. Alzheimer's and ischemia, have been linked with alterations in glymphatic system activity [68,69]. The influence of this systematic fluid movement through the interstitium, constituting 12-20% of cerebral water content [70], on DTI-signal is has not been systematically evaluated. We propose the hypothesis that a proportion of the DTI signal is explained by directional interstitial proton movement due to glymphatic system activity. Alterations of this system due to chronic neuroinflammation could account for the distinct increase of hippocampal MD and AD, comparatively smaller increase in RD, and consecutive decrease in FA detected in our patients. While such changes may not be limited to the hippocampus, their effect on DTI signal would likely be more apparent in less structured tissue, i.e., the hippocampi or grey matter, in comparison to cerebral white matter with its longitudinal and parallel fiber tracts, where the traditional tubular axon model conclusively explains DTI parameter patterns during microstructural change. Targeted research is needed to further explore this concept.
The compartment of perivascular spaces (PVS) has already been shown to systematically influence DTI signals [71]. Predominant measurement of DTI signal along the PVS has been proposed as a model for detecting glymphatic activity [72]. Diurnal alterations of DTI parameters could be explained in another model by liquor-like portions of the DTI signal, presumably representing PVS [73]. While the PVS of superficial brain parenchyma near the entrance point of penetrating vessels was associated with a higher MD and lower FA [71], a continuously decreasing diameter up to capillary level in deeper laying tissue, such as the hippocampus, could result in a relatively greater increase in AD and thus FA increase.
Predominant damage to the relatively isotropic, radiosensitive neuronal stem cells, as compared to the anisotropic neuronal axons, may further contribute to a FA increase. Extensive radiogenic effects on the extracellular matrix may also lead to an increase in MD and RD, and in close proximity, the PVS, AD, and FA [74].
As described above, previous studies reported decreases in hippocampal FA in measurements conducted very early after radiation. However, acute radiation reactions are thought to be driven primarily by subtle parenchymal edema not yet detectable by conventional imaging. Edema may lead to a decrease in FA both directly [75] and indirectly by restricting fluid motion through perivascular pathways [76]. Early FA decreases would therefore be expected to diminish after edema resolution in the late-delayed reactions measured in this study.
Mean hippocampal radiation dose was associated with a stronger increase in hippocampal FA. However, hippocampal DTI changes appeared to be more pronounced in the HSR subgroup. This was significant when comparing AD of both groups after 6 months, with a higher AD detected in the HSR group. The underlying mechanism of this phenomenon is unclear. Potentially, a divergent immunocellular response above a certain threshold of cellular damage might influence hippocampal DTI characteristics. The observed differences may also result from radiogenic changes outside the hippocampus. A thus triggered generalized inflammatory response triggered could impact hippocampal DTI characteristics more strongly than local radiogenic tissue damage. Notably, in this context, hippocampal sparing radiotherapy planning may lead to differences in relevant characteristics of the three-dimensional dose distribution outside the hippocampus, including increased dose inhomogeneity and higher maximum point doses.
Additionally, tissue compression due to peritumoral mass effect, the influence of crossing fiber tracts, anisotropy increases due to gliosis, and diffusion differences in the pathophysiological stages of axonal degeneration have been discussed to account for unexpected DTI changes after radiation [46,[77][78][79][80].
Notably, a multilinear model approach without multiple testing corrections was chosen in this study to allow for a more direct interpretation of the observed data, which needs to be considered when interpreting the provided results. Albeit somewhat larger than previous DTI study samples evaluating hippocampal radiation response, this study is still limited by its relatively low number of subjects. Generally, the susceptibility of hippocampal measurement to CSF contamination may variably lead to false low FA values. Although reducing standardization compared to the automated selection, a manual anatomical approach to ROI selection was thus chosen to enable visual quality control by the researcher. As a technical limitation, direct co-registration of DTI-maps and the three-dimensional radiotherapy dose distribution was impossible.
Previous trials have consistently reported correlations between DTI signal alterations of cerebral white matter and neurocognitive decline [81,82], in some instances proposing a predictive nature of early DTI measurements [81,83], as well as between hippocampal dose and neurocognitive decline [16,[21][22][23]26,28]. Therefore, the dose-dependent hippocampal DTI changes observed in this study suggest a potential link between radiogenic DTI changes and neurocognitive impairment, which needs to be evaluated in further research.
Differentiating which proportion of DTI signal changes result from axonal injury or demyelination and which are accounted for by further pathophysiological processes (such as glymphatic system activity, edema, vascular injury, or other inflammatory processes) will pose a crucial goal in future DTI research aiming to evaluate cerebral radiation response.

Conclusions
This study adds to evidence of DTI parameter changes in the brain parenchyma after CR, underlining the feasibility of this technique in assessing cerebral radiation response. The discovered patterns of hippocampal microstructural change indicate hitherto insufficiently explained pathomechanisms of the radiation response influencing DTI signal. 6 months after CR, hippocampal microstructure differed between HSR and CCR patients.  Data Availability Statement: The data sets used and/or analyzed during this study are available from the corresponding author upon reasonable request.