Background: Metabolic brain imaging with 2-[¹⁸F]fluoro-2-deoxy-D-glucose positron emission tomography (FDG PET) is a supportive diagnostic and differential diagnostic tool for neurodegenerative dementias. In the clinic, scans are usually visually interpreted. However, computer-aided approaches can improve diagnostic accuracy. We aimed to build two machine learning classifiers, based on two sets of FDG PET-derived features, for differential diagnosis of common dementia syndromes.Methods: We analyzed FDG PET scans from three dementia cohorts [63 dementia due to Alzheimer’s disease (AD), 79 dementia with Lewy bodies (DLB) and 23 frontotemporal dementia (FTD)], and 41 normal controls (NCs). Patients’ clinical diagnosis at follow-up (25 ± 20 months after scanning) or cerebrospinal fluid biomarkers for Alzheimer’s disease was considered a gold standard. FDG PET scans were first visually evaluated. Scans were pre-processed, and two sets of features extracted: (1) the expressions of previously identified metabolic brain patterns, and (2) the mean uptake value in 95 regions of interest (ROIs). Two multi-class support vector machine (SVM) classifiers were tested and their diagnostic performance assessed and compared to visual reading. Class-specific regional feature importance was assessed with Shapley Additive Explanations.Results: Pattern- and ROI-based classifier achieved higher overall accuracy than expert readers (78% and 80% respectively, vs. 71%). Both SVM classifiers performed similarly to one another and to expert readers in AD (F1 = 0.74, 0.78, and 0.78) and DLB (F1 = 0.81, 0.81, and 0.78). SVM classifiers outperformed expert readers in FTD (F1 = 0.87, 0.83, and 0.63), but not in NC (F1 = 0.71, 0.75, and 0.92). Visualization of the SVM model showed bilateral temporal cortices and cerebellum to be the most important features for AD; occipital cortices, hippocampi and parahippocampi, amygdala, and middle temporal lobes for DLB; bilateral frontal cortices, middle and anterior cingulum for FTD; and bilateral angular gyri, pons, and vermis for NC.Conclusion: Multi-class SVM classifiers based on the expression of characteristic metabolic brain patterns or ROI glucose uptake, performed better than experts in the differential diagnosis of common dementias using FDG PET scans. Experts performed better in the recognition of normal scans and a combined approach may yield optimal results in the clinical setting.

Metabolic brain biomarkers have been incorporated in various diagnostic guidelines of neurodegenerative diseases, recently. To improve their diagnostic accuracy a biologically and clinically homogeneous sample is needed for their identification. Alzheimer’s disease-related pattern (ADRP) has been identified previously in cohorts of clinically diagnosed patients with dementia due to Alzheimer’s disease (AD), meaning that its diagnostic accuracy might have been reduced due to common clinical misdiagnosis. In our study, we aimed to identify ADRP in a cohort of AD patients with CSF confirmed diagnosis, validate it in large out-of-sample cohorts and explore its relationship with patients’ clinical status. For identification we analyzed 2-[¹⁸F]FDG PET brain scans of 20 AD patients and 20 normal controls (NCs). For validation, 2-[¹⁸F]FDG PET scans from 261 individuals with AD, behavioral variant of frontotemporal dementia, mild cognitive impairment and NC were analyzed. We identified an ADRP that is characterized by relatively reduced metabolic activity in temporoparietal cortices, posterior cingulate and precuneus which co-varied with relatively increased metabolic activity in the cerebellum. ADRP expression significantly differentiated AD from NC (AUC = 0.95) and other dementia types (AUC = 0.76–0.85) and its expression correlated with clinical measures of global cognition and neuropsychological indices in all cohorts.

Dementia with Lewy bodies (DLB) is the second most common neurodegenerative dementia, that shares clinical and metabolic similarities with both Alzheimer’s and Parkinson’s disease. In this study we aimed to identify a DLB-related pattern (DLBRP), study its relationship with other metabolic brain patterns and explore its diagnostic and prognostic value. A cohort of 79 participants with DLB, 63 with dementia due to Alzheimer’s disease (AD) and 41 normal controls (NCs) and their 2-[¹⁸F]FDG PET scans were analysed for identification and validation of DLBRP. Voxel-wise correlation and multiple linear regression were used to study the relation between DLBRP and Alzheimer’s disease-related pattern (ADRP), Parkinson’s disease-related pattern (PDRP) and PD-related cognitive pattern (PDCP). Diagnostic and prognostic value of DLBRP and of modified DLBRP after accounting for ADRP overlap (DLBRP ⊥ ADRP), were explored. The newly identified DLBRP shared topographic similarities with ADRP (R² = 24%) and PDRP (R² = 37%), but not with PDCP. We could accurately discriminate between DLB and NC (AUC = 0.99) based on DLBRP expression, and between DLB and AD (AUC = 0.87) based on DLBRP ⊥ ADRP expression. DLBRP expression correlated with cognitive impairment, but the correlation was lost after accounting for ADRP overlap. DLBRP and DLBRP ⊥ ADRP correlated with patients’ survival time. DLBRP has proven to be a specific metabolic brain biomarker of DLB, sharing similarities with ADRP and PDRP, but not PDCP. We observed a similar metabolic mechanism underlying cognitive impairment in DLB and AD. DLB-specific metabolic changes were more detrimental for overall survival.

Behavioral variant of frontotemporal dementia (bvFTD) is common among young‐onset dementia patients. While bvFTD‐specific multivariate metabolic brain pattern (bFDRP) has been identified previously, little is known about its temporal evolution, internal structure, effect of atrophy, and its relationship with nonspecific resting‐state networks such as default mode network (DMN). In this multicenter study, we explored FDG‐PET brain scans of 111 bvFTD, 26 Alzheimer's disease, 16 Creutzfeldt‐Jakob's disease, 24 semantic variant primary progressive aphasia (PPA), 18 nonfluent variant PPA and 77 healthy control subjects (HC) from Slovenia, USA, and Germany. bFDRP was identified in a cohort of 20 bvFTD patients and age‐matched HC using scaled subprofile model/principle component analysis and validated in three independent cohorts. It was characterized by hypometabolism in frontal cortex, insula, anterior/middle cingulate, caudate, thalamus, and temporal poles. Its expression in bvFTD patients was significantly higher compared to HC and other dementia syndromes (p < .0004), correlated with cognitive decline (p = .0001), and increased over time in longitudinal cohort (p = .0007). Analysis of internal network organization by graph‐theory methods revealed prominent network disruption in bvFTD patients. We have further found a specific atrophy‐related pattern grossly corresponding to bFDRP; however, its contribution to the metabolic pattern was minimal. Finally, despite the overlap between bFDRP and FDG‐PET‐derived DMN, we demonstrated a predominant role of the specific bFDRP. Taken together, we validated the bFDRP network as a diagnostic/prognostic biomarker specific for bvFTD, provided a unique insight into its highly reproducible internal structure, and proved that bFDRP is unaffected by structural atrophy and independent of normal resting state networks loss.

Background: Alzheimer’s disease (AD) is marked by accumulation of Aβ and tau protein causing neurodegeneration in the brain. AD represents a pathological and clinical continuum, ranging from mild cognitive impairment (MCI) to dementia. A specific metabolic imaging biomarker of AD ‐ Alzheimer’s disease‐related pattern (ADRP) has been previously identified from 2‐[¹⁸F]fluorodeoxyglucose positron emission tomography (2‐[¹⁸F]FDG‐PET) scans of clinically diagnosed AD patients, using scaled subprofile model based on principal component analysis (SSM/PCA) ^1–3. However the clinical diagnosis may be wrong in substantial number of cases ⁴. We aimed to identify ADRP on patients with Alzheimer’s pathological changes in cerebrospinal (CSF) fluid and to validate it in an independent group of AD and other dementia patients and compare it to the original ADRP ¹.Method: 2‐[¹⁸F]FDG‐PET scans from 20 AD1 and 20 normal controls (NC1) were used for pattern identification. Additional 110 scans were analyzed for validation: 68 from AD patients (37 AD2, 13 atypical clinical presentation AD (atAD), 18 MCI), 21 with frontotemporal dementia (FTD), 8 non‐AD MCI and 13 NC2 (Table 1). AD was defined with CSF biomarkers as: Aβ42 < 815 pg/ml, pTau > 60 pg/ml, tTau > 400 pg/ml. Topographic profile rating algorithm was used to prospectively calculate ADRP Z‐scores from subjects scans³. Mini‐mental state examination (MMSE) was performed in all patients.Results: ADRP was identified and voxel weights were found to be stable by bootstrap resampling, Figure 1. Pattern’s expression was significantly higher in AD1 than NC1 (p<0.001) and it strongly correlated (r=‐0.70) with MMSE. Pattern’s expression was also higher in AD2 than NC2 (p<0.001), MCI vs. NC2 (p<0.001), MCI vs. non‐AD MCI (p=0.02) and AD2 vs. FTD (p=0.02), Figure 2. Newly identified pattern moderately correlated (r=0.52) with the original one ¹.Conclusions: We identified ADRP on a cohort of pathologically confirmed AD patients, which was not done before. ADRP has shown to be a reliable metabolic biomarker of AD related neurodegeneration. References: (1) Mattis,PJ. et al. Neurology 87,1925–33(2016); (2) Teune,LK. et al. Curr Alzheimer Res 11,725–32(2014); (3) Spetsieris,PG. & Eidelberg,D. Neuroimage 54,2899–914(2011); (4) Beach,TG. et al. J. Neuropathol. Exp. Neurol. 71,266–273(2012).

Background: Dementia with Lewy bodies (DLB) is the second most common neurodegenerative dementia, yet it remains under-recognised and misdiagnosed, which delays treatment, causes inaccurate prognosis and limits research opportunities. Imaging with 2-[ ¹⁸ F]fluoro-2-deoxy-D-glucose positron emission tomography (FDG PET) is a supportive DLB biomarker. We evaluated a multivariate, quantifiable metabolic network biomarker, termed DLB-related pattern (DLBRP), for its further clinical translation across centres and disease stages. Methods: We analysed demographic, clinical and FDG PET imaging data of 1180 participants from 14 tertiary centres and two multicentre datasets. We included 379 DLB, 28 mild cognitive impairment-LB (MCI-LB), 195 dementia due to Alzheimer’s disease (ADD), 172 MCI-AD without α-synuclein co-pathology (MCI-AD-S–), and 73 MCI-AD with α-synuclein co-pathology (S+) patients, along with a comparative group of 333 normal controls (NCs). From the scans, we calculated the expression of DLBRP, AD-related pattern (ADRP) and Parkinson’s disease-related pattern (PDRP) and compared them across groups. DLBRP scores were correlated with clinical measurements. Results: Across independent cohorts, DLBRP robustly distinguished DLB from NCs (sensitivity >89%, specificity >90%), and scores correlated with Unified Parkinson’s Disease Rating Scale Part III and independently predicted Mini–Mental State Examination. DLBRP was elevated already in MCI-LB. In a small longitudinal dataset, we observed steady increases in DLBRP expression with scores exceeding the diagnostic threshold prior to dementia onset. DLBRP and PDRP discriminated DLB from ADD (sensitivity, 74%–90%; specificity, 80%). In MCI-AD groups, ADRP was expressed, whereas DLBRP and PDRP were increased only in MCI-AD-S+, although comparatively less than in MCI-LB. Conclusions: This study demonstrates the value of DLBRP in diagnosing prodromal and manifest DLB and distinguishing them from their AD counterparts. While overlap between patterns may reflect actual co-pathology, this possibility cannot be accepted without thorough pathological confirmation. The current findings support the use of DLBRP in patient evaluation and in future trial design.