Our integrated multimodal neuroimaging framework might allow us to systematically investigate the neural signatures of metabolic, perfusion, and microstructural changes in patients with rare NDDs. Multimodal neuroimaging showed characteristic regional decrease patterns (predominantly in the GM) in PET-derived glucose metabolism (i.e., SUVR) and ASL-derived rCBF, and alteration patterns (preferably in the white matter [WM]) in DTI-derived FA. At the individual level, an initial estimation of subject-specific changes in defined brain regions (hypothetically disease-related) could thus be demonstrated and quantified. However, given the small sample size, rather than showing disease-specific imaging patterns or diagnostic utility in representative cohorts, this study suggests feasibility and allows generation of hypotheses rather than demonstrating final clinical applicability. In contrast to a previous related study at the group level [17], this study may pave the way from studies at the group level to the possibility of individual imaging-based categorization/analysis by comparing single-subject data to a group of subjects/patients.

A timely and accurate differential diagnosis between NDD subtypes is highly important in the light of the constantly evolving therapeutic portfolio, and this may be facilitated by neuroimaging techniques [11,12,13]. Especially quantitative and multi-parametric approaches may be suited to capture the multifaceted alterations of NDD subtypes, as represented by the herein used framework integrating metabolic, perfusion, and microstructural parameters from simultaneous FDG-PET/MRI acquisitions. Notably, the patients investigated showed a clinical presentation that was not clearly attributable to a specific disease entity in the first place, and they were finally diagnosed with an atypical parkinsonian syndrome (CBD, NDD002) or an atypical motor neuron disease (PLS, NDD001 and NDD003), respectively. In this regard, for CBD, hypometabolism in perirolandic regions and the thalamus and markedly in the basal ganglia can be observed by FDG-PET according to previous work, with a characteristic dominance towards the clinically more affected hemisphere [27,37]. For the ALS variant PLS, typically primary motor cortex hypometabolism can be observed on FDG-PET, together with hypometabolism in projection on the course of the CST [38,39,40]. For perfusion imaging, ASL has been used only rarely, with previous work indicating CBF asymmetry in the perirolandic area and parietal cortex in patients with corticobasal syndrome [41]. Regarding motor neuron diseases, hypoperfusion was found especially for motor regions, but also for non-motor regions such as fronto-temporal regions [34]. For WM microstructure as evaluated by DTI, patients with corticobasal syndrome showed FA decreases in the sensorimotor projections of the cortical hand areas, among other regions [42,43]. In ALS and also in PLS, DTI has been used to predominantly investigate the CST, with FA being typically reduced at baseline or progressively over time [44,45]. Yet, previous work mostly used one imaging modality in isolation, thus being unable to reveal disease-related alterations in multiple domains at the same time [27,38,40,41,42,43,44,45]. Our study, however, applied combined FDG-PET, ASL, and DTI to rare and atypical variants of NDDs, thereby paving the way for the identification of hypothetical disease-related changes. In a general context, the definition of ROIs—which in this study was based on established areas of hypometabolism, hypoperfusion, and/or microstructural changes identified in the literature and conceptualized by WBSS analyses—cannot yet be considered final and should be refined in studies involving large patient cohorts; further definitions of pathologically indicated ROI locations can be found in previous related work [17].

For NDD001, in whom diagnosis remained initially vague with spastic and rigid muscle tone versus PLS in the end, our multi-parametric framework revealed decreased SUVR in one cluster close to the midline in the central and postcentral gyrus, while rCBF maps were in agreement with the location of the central gyrus bilaterally, but showing more laterally expanding alterations overlapping with the “hand knob” as the anatomical location of primary motor function. Downstream effects of those impairments seem to be reflected by bilateral FA decreases along parts of the course of the CST. In this case, all three methods from simultaneous acquisition point to a marked disease involvement of the motor system, both on the GM and WM level. Consideration of ASL and DTI in addition to FDG-PET may enhance diagnostic confidence with high complementary yield. Notably, changes within the motor system were most markedly highlighted by ASL decreases, and the rather symmetric alterations favored a PLS diagnosis over CBD from a neuroimaging point of view. Yet, also relatively large clusters of bihemispheric rCBF increases were found, primarily located within subcortical areas and WM. While large-scale studies on ASL specifically in patients with PLS are mostly lacking, those changes may indicate complex patterns in the disease course and pathophysiology of PLS as expressed by rCBF increases, which may potentially represent compensatory mechanisms at the subcortical and WM level that are related to the marked rCBF decreases predominantly at the cortical level in NDD001.

In NDD002 with a final diagnosis of CBD, SUVR was decreased at the right central gyrus, partially also overlapping with the rCBF maps, which showed a large right-hemispheric and smaller left-hemispheric cluster at the central gyrus. Furthermore, FA maps showed bihemispheric decreases along parts of the CST. In this regard, the findings from ASL and DTI seem remarkable in that they highlight bihemispheric yet asymmetric changes in the motor system, with bihemispheric changes being relatively widespread. While CBD is characteristically asymmetric (especially early in the disease course), it is bihemispheric in the underlying pathology, involving both sides of the brain along the progression trajectory [46,47]. Even in the case of unilateral symptoms, the underlying tau protein accumulation is commonly widespread across both hemispheres and the basal ganglia, also affecting WM structures [48,49,50].

In NDD003 with a final diagnosis of PLS, small parietal clusters of decreased SUVR were observed, while rCBF maps showed one cluster with decreases in the right central gyrus, and FA maps indicated discrete bihemispheric decreases especially in the upper/central CST. In this case, rather subtle changes were revealed, which might go unnoticed or classified as unspecific when solely using information from one imaging modality or visual image readings. The findings were more subtle than in NDD001, hypothetically reflecting the earlier disease course. Notably, findings can also be negative for FDG-PET in PLS, which typically has a slower disease course as compared to ALS [51]. Overall, while from PET/MRI acquisitions the structural standard sequences are oftentimes used for mere anatomical co-registration, the adjunct of advanced sequences such as ASL and DTI may facilitate correct diagnoses also in cases where SUVR alterations are minor or not entirely specific, and they could also improve diagnostic confidence by delivering complementary information.

Given that perfusion imaging by ASL is a modality with a limited number of applications to ALS/PLS as well as CBD so far, the results of the current comparative study add to the previous literature, which has revealed abnormalities in perfusion (especially including hypoperfusion) in primary motor areas [34,41]. These findings could be confirmed in our study, and it could be demonstrated that the PET- and ASL-derived significant clusters slightly differed from each other, although they were located adjacent to each other or partially overlapped (mostly along the motor cortices). Yet, distinct individual spatial patterns and also bihemispheric alterations have been revealed in our patients, which may further highlight the complementary potential. In addition, DTI demonstrated significant reductions in FA in WM regions predominantly adjacent to the hypometabolic and hypoperfused GM areas, so that WM alterations could be considered to be closely tied to GM impairment. Those results were obtained in comparison to individuals with SCDs. Statistically, the group of individuals with SCDs represents a group of subjects with no conspicuousness on PET/MRI and a relatively normal distribution of PET/MRI-derived analysis parameters; even if one subject would be in a very early stage of a potential NDD, the contribution to the group-level parameter would be in the order of ~6% and, thus, should not significantly influence the statistical comparison of the respective patient with NDD versus the SCD group.

The present ASL analyses processed each PLD separately and subsequently averaged the resulting CBF maps, that way ensuring stability within the context of this study; however, this does not fully utilize the potential of multi-delay acquisitions. Multi-PLD perfusion imaging would make it possible to determine the most efficient multi-PLD acquisition protocol in terms of the number of label/control pairs and could further refine ASL results. Model-based approaches that jointly estimate CBF and arterial transit times (ATTs) have been shown to provide more physiologically accurate perfusion estimates, particularly in samples with prolonged or heterogeneous ATTs, such as older patients or patients with cerebrovascular diseases [21,52]. Thus, multi-PLD perfusion imaging that includes rather short PLDs (<1.0 s) and rather long PLDs (>2.5 s) is necessary to provide sufficient information to estimate ATT for calculating adjusted CBF maps [53].

Regarding limitations, our study included only a very limited number of exemplary cases, diagnosed with rare and atypical NDD entities. Of note, the main intention of this methodological investigation was to propose a modular framework for parameterized multimodal neuroimaging, which could facilitate differential diagnosis especially in clinically complex cases and rare NDDs. By use of this methodological approach, a detailed association assessment of neuroimaging patterns and clinical phenotypes could be performed in the future. Although observations from the present study are not representative, the findings suggest that a combined PET/MRI framework could help to improve phenotype characterization in NDDs at an individual level. Since this study was a retrospective investigation, a group of individuals with SCDs was used to determine the location of the ROIs and for the comparisons, rather than a group of completely healthy control subjects. This SCD group may have included individuals with preclinical neurodegenerative changes or other neurological/psychiatric comorbidities (though without detectable pathological findings on PET). However, entirely healthy subjects typically do not undergo PET examinations in clinical routine, due to its invasiveness and radiotracer-related radiation exposure. From a statistical point of view, even if one subject would be in a very early stage of a neurodegenerative disease, the contribution to the group-level parameter would be in the order of ~6%. Nevertheless, the methodology of this study should ideally be validated in the future using age- and sex-matched healthy controls. Furthermore, the volumetric assessments from T1-weighted images based on commercially available software were derived from comparisons to control datasets as normative data. However, these datasets may not necessarily include balanced geographic, ethnic, or demographic composition, and the educational status was unknown. Thus, local controls with homogeneous sequences from the same MRI scanner and elaborate clinical information may have been used for more exact reference comparisons. As a final limitation, it has to be noted that when the framework is applied to patients whose neuroanatomy diverges substantially from the template, normalization errors could systematically affect parameter quantification in ways that the current analysis cannot detect.