Previous [18F]FDG-PET studies in DLB mostly focused either on the assessment of [18F]FDG-PET hypometabolism patterns at the group level, thus not allowing for measures of diagnostic accuracy, or on the relationship between focal metabolic alterations and clinical symptoms.
The [18F]FDG-PET SPM maps confirmed the presence of a specific hypometabolism pattern, both at the group and single-subject level (Fig.
2; Fig.
3), in our DLB case series. This hypometabolism pattern was characterized by occipital, parieto-temporal, and frontal involvement, consistent with previous reports on very limited case series [
6,
23,
29,
33].
Notably, the cluster analysis revealed two hypometabolism subtypes (Fig.
5), with the severity of occipital hypometabolism driving most of the difference. The reason why brain occipital regions may show different degrees of metabolic involvement in DLB is unclear. Brain glucose hypometabolism patterns, as revealed by [18F]FDG-PET, are related to the underlying synaptic dysfunction, as determined by local and long-distance effects due to underlying neuropathology [
34,
35]. The occipital hypometabolism found in DLB is independent of brain atrophy, as gross structural changes in the occipital lobe do not occur in patients with DLB [
36].
In our series, more severe occipital involvement (Fig.
5) was predictive of a higher prevalence and higher likelihood of developing VH at follow-up, in keeping with the previously reported association between VH and occipital hypometabolism [
37‐
39] and in support of the predictive ability of hypometabolism with respect to the development of clinical features. Deficits in the visuo-spatial, visuo-attention, and visuo-constructive cognitive domains, which have been suggested to promote VH [
40‐
42], were also more severe in the subgroup with a more extensive occipital involvement. Converging evidence shows that both VH and neuropsychological deficits might be ascribed to cholinergic dysfunction [
43‐
45]. The clinical and neuropsychological differences may therefore indicate different degrees of underlying cholinergic impairment, ultimately reflecting the variable occipital hypometabolism. It may also be speculated that the observed differences in severity of occipital hypometabolism could indicate the presence of a variable amount of neocortical occipital Lewy body pathology, supporting previous claims of possible DLB pathological subtypes [
46,
47]. Accordingly, it has been suggested that Lewy body pathology could be the main determinant of occipital hypometabolism, whereas temporo-parietal hypometabolism in DLB might be driven by underlying AD pathology [
48]. Consistently, and in keeping with our findings, pure Lewy body pathology has been associated with greater impairment in visuospatial and visual perceptive abilities [
49,
50], whereas comorbid AD pathology was related to worse performance in other cognitive domains, such as memory and orientation [
51].
Here, in a large series of cases, we have shown that [18F]FDG-PET alone, when combined with optimized semi-quantitative approaches at the single-subject level, is able to
predict the diagnosis of DLB 3 years before the final clinical diagnosis, with extremely high accuracy (≈ 90%). This provides insight into the sensitivity of [18F]FDG-PET to define pathological changes when clinical diagnosis alone is still uncertain. Considering the high clinical variability [
1] and lack of a clear clinical picture in the initial phases of DLB, an early diagnosis based on clinical data alone may be challenging. Consistently, as shown in Table
1, at entry (i.e., clinical classification at entry), 10 out of 72 patients were classified as possible DLB, and 34 out of 72 as probable DLB. The remaining clinical presentations were attributed to other dementias, such as ADD, or classified as atypical parkinsonisms or MCI. The possible diagnostic misclassifications at entry may be partially due to the clinicians’ uncertainty about patients’ clinical presentations. Several factors may have influenced the low level of confidence for the presence of core symptoms, e.g., very doubtful presence of visual hallucinations and cognitive fluctuations, only reported by the relatives and not evaluated by standard sequential neuropsychological exams. The high power of the [18F]FDG-PET hypometabolism signatures in predicting the final clinical diagnosis provided a ≈ 50% increase in accuracy compared to the first clinical assessment alone (clinical classification at entry). In cases of diagnostic uncertainty, the use of the [18F]FDG-PET biomarker in the flowchart may help to specify the clinical diagnosis, predicting the clinical scenario years in advance (Fig.
1). It is important to note that [18F]FDG-PET raters were blinded to clinical diagnoses at entry and follow-up, in accordance with the main objective of the study. In reality, however, the final clinical diagnosis was obtained with, and reinforced by, available core clinical features and supportive biomarkers, consistent with consensus guidelines [
5].
As for the differential diagnosis, the misdiagnosis of DLB as ADD is usually due to the overlap of both cognitive and behavioral symptoms [
4]. Previous studies aimed at assessing [18F]FDG-PET as a biomarker for the
differential diagnosis between DLB and ADD reported overall good, although not always optimal, accuracy values (cfr. [
4]). When considering all available studies (see Additional file
1: Table S1 for an overview) with a fair sample size (
N = 30) [
52,
53], simple visual assessment produced accuracy values of around 70% (range 69–72%) and quantitative metrics produced an accuracy of 85%. Here, we adopted an optimized semi-quantitative SPM-based approach, which has previously shown excellent performance in discriminating dementia conditions [
6,
54], outperforming visual qualitative assessment [
6]. Our results indicate that occipital lobe hypometabolism, found here in 91.7% of DLB patients and yielding the greatest ability to discriminate between DLB and ADD, is the dysfunctional signature of DLB (Table
4; Fig.
4). The substantial, although suboptimal, accuracy produced using the cingulate island sign (ROC < 80%) in comparison with the other hallmarks confirms its value in supporting differential diagnosis with AD, as acknowledged in the new DLB diagnostic criteria [
55]. Our cut-off CIS value of 0.82 to discriminate between DLB and ADD patients was consistent with a previous study showing that a CIS value of 0.81 optimally discriminated between AD and DLB pathology at post-mortem evaluation [
27]. Additionally, another study found a CIS value of 0.79 as the optimal cut-off to discriminate between clinically diagnosed DLB and ADD patients [
27]. It has been shown that a higher cingulate island sign value could be specifically associated with a lower neurofibrillary tangle (NFT) pathology Braak stage rather than with the presence of significant amyloid plaque deposition [
56]. Our results are consistent with this claim, suggesting that most of our clinically diagnosed DLB patients were likely to have lower NFT pathology as compared to the ADD sample.
Notably, the DLB vs. ADD classification was hampered by the presence of 8.3% of cases with the PCA variant in the ADD sample. The differentiation between PCA and DLB hypometabolism patterns can be challenging due to the topographical overlap in the hypometabolism patterns between the two conditions [
28] (Fig.
3). However, as shown in a recent study by Cerami et al., the focal hypometabolism of the frontal eye field regions may help in defining the PCA condition [
25], while a more diffuse involvement of DLPFC is frequently reported in DLB and not in PCA (Fig.
3).
Difficulty with diagnostic differentiation between DLB and atypical parkinsonisms is also frequent, due to ambiguous clinical presentations characterizing the initial clinical stages [
57]. Hughes et al. reported that although the clinical diagnosis at follow-up reached an 85% accuracy in predicting the post-mortem diagnosis, the clinical classification at entry was revised in 60% of patients [
58]. When the clinical scenario is still ambiguous, [18F]FDG-PET can provide patterns of brain hypometabolism able to accurately discriminate between DLB, progressive supranuclear palsy, cortico-basal degeneration, and multiple system atrophy [
23]. In a previous study, the [18F]FDG-PET single-subject procedure was applied in a more limited DLB case series (
N = 29), showing promising discriminative abilities with respect to other atypical parkinsonian disorders in the early diagnostic phase [
23].
As for the DLB vs. PD classification, it must be noted that most of the PD patients (84%) presented with the typical PD-like [18F]FDG-PET pattern [
24] (excellently discriminated from DLB by the SPM-t-map classification). These stable, cognitively normal patients with PD might present with executive-attention deficits [
59] and very limited prefrontal hypometabolism, contrasting with the prevalent posterior cortical dysfunction characterizing DLB.
On the other hand, a small number (
N = 5, 13.8%) of PD cases exhibited the pattern typically found in DLB, characterized by occipital and posterior parietal-temporal hypometabolism (Table
3). The DLB-like pattern has been previously observed in patients with Parkinson’s disease [
60,
61]. Crucially, the DLB-like pattern was detected at the individual level in PD patients, even in the early disease phase, and was associated with an increased risk of developing dementia along the disease course [
24]. The hypometabolism overlap in patients on the trajectory to Parkinson’s disease with dementia (PDD) and those with DLB is not surprising, considering the phenotypical, genetic, and pathological similarities between the two syndromes [
62,
63]. Accordingly, the PDD/DLB distinction is today considered essentially a clinical one and purely based on the chronology of symptom presentation. The use of [18F]FDG-PET hypometabolism t-maps is thus critical in distinguishing between DLB/PDD and stable PD and is of extreme importance for prognostic accuracy and treatment options.
[18F]FDG-PET hypometabolism represents a flexible biomarker, able to differentiate DLB not only from ADD, but also from other parkinsonisms, either typical (as shown here) or atypical (see [
23]). This is different from DaTSCAN, currently included as an indicative biomarker in the DLB diagnostic criteria, which only provides a dichotomous discrimination between parkinsonisms and ADD and does not allow for a more fine-grained differentiation between DLB and typical and atypical parkinsonisms [
4,
23,
55]. Notably, [18F]FDG-PET, combined with semi-quantitative methods, allows for discrimination between DLB and ADD with accuracy even superior to that of DaTSCAN, whose sensitivity (78%), is also hampered by the negligible dopaminergic neurodegeneration characterizing some DLB cases (cfr. [
4]).
The [18F]FDG-PET classification showed substantial inter-rater agreement, consistent with previous validation studies in other neurodegenerative conditions [
6,
23,
54,
64,
65]. This supports the reliability of the [18F]FDG-PET SPM maps for diagnostic classification, suggesting that the assessment of disease-specific hypometabolism patterns obtained by this [18F]FDG-PET SPM procedure is not influenced by the specific rater expertise. Our SPM-t-maps, derived from voxel-by-voxel comparisons between an individual subject and a predefined normal database, provide fast and easily accessible information about the distribution of brain hypometabolism that does not require rater expertise. Similarly, a previous comparative study showed that the SPM-t-maps of brain hypometabolism provide the clinician with a higher degree of accuracy compared to visualization of standard rainbow maps of FDG uptake [
6]. The effectiveness of this procedure in clinical routine is supported by the independence of clinicians’ expertise, as shown by comparable results among raters.