The underlying mechanism of brain glucose hypometabolism, an invariant neurodegenerative feature that tightly correlates with cognitive impairment and disease progression of Alzheimer’s disease (AD), remains elusive. Positron emission tomography with 2-[18F]fluoro-2-deoxy-d-glucose (FDG-PET) was used to evaluate brain glucose metabolism, presented as the rate of 2-[18F]fluoro-2-deoxy-d-glucose standardized uptake value ratio (FDG SUVR) in patients with AD or control subjects and in mice with or without thiamine deficiency induced by a thiamine-deprived diet. Brain amyloid-β (Aβ) deposition in patients with clinically diagnosed AD was quantified by performing assays using 11C-Pittsburgh compound B PET. The levels of thiamine metabolites in blood samples of patients with AD and control subjects, as well as in blood and brain samples of mice, were detected by high-performance liquid chromatography with fluorescence detection. FDG SUVRs in frontal, temporal, and parietal cortices of patients with AD were closely correlated with the levels of blood thiamine diphosphate (TDP) and cognitive abilities, but not with brain Aβ deposition. Mice on a thiamine-deprived diet manifested a significant decline of FDG SUVRs in multiple brain regions as compared with those in control mice, with magnitudes highly correlating with both brain and blood TDP levels. There were no significant differences in the changes of FDG SUVRs in observed brain regions between amyloid precursor protein/presenilin-1 and wild-type mice following thiamine deficiency. We demonstrate, for the first time to our knowledge, in vivo that TDP reduction strongly correlates with brain glucose hypometabolism, whereas amyloid deposition does not. Our study provides new insight into the pathogenesis and therapeutic strategy for AD.

Alzheimer’s Disease (AD) and Mild Cognitive Impairment (MCI) are closely associated with Tau proteins accumulation. In this study, we aimed to implement radiomics analysis to discover high-order features from pathological biomarker and improve the classification accuracy based on Tau PET images. Two cross-racial independent cohorts from the ADNI database (121 AD patients, 197 MCI patients and 211 normal control (NC) subjects) and Huashan hospital (44 AD patients, 33 MCI patients and 36 NC subjects) were enrolled. The radiomics features of Tau PET imaging of AD related brain regions were computed for classification using a support vector machine (SVM) model. The radiomics model was trained and validated in the ADNI cohort and tested in the Huashan hospital cohort. The standard uptake value ratio (SUVR) and clinical scores model were also performed to compared with radiomics analysis. Additionally, we explored the possibility of using Tau PET radiomics features as a good biomarker to make binary identification of Tau-negative MCI versus Tau-positive MCI or apolipoprotein E (ApoE) ε4 carrier versus ApoE ε4 non-carrier. We found that the radiomics model demonstrated best classification performance in differentiating AD/MCI patients and NC in comparison to SUVR and clinical scores models, with an accuracy of 84.8 ± 4.5%, 73.1 ± 3.6% in the ANDI cohort. Moreover, the radiomics model also demonstrated greater performance in diagnosing AD than other methods in the Huashan hospital cohort, with an accuracy of 81.9 ± 6.1%. In addition, the radiomics model also showed the satisfactory classification performance in the MCI-tau subgroup experiment (72.3 ± 3.5%, 71.9 ± 3.6% and 63.7 ± 5.9%) and in the MCI-ApoE subgroup experiment (73.5 ± 4.3%, 70.1 ± 3.9% and 62.5 ± 5.4%). In conclusion, our study showed that based on Tau PET radiomics analysis has the potential to guide and facilitate clinical diagnosis, further providing evidence for identifying the risk factors in MCI patients.

Objective: Interpretability and reproducibility remain major challenges in applying deep neural network (DNN) to neuroimaging-based diagnosis. This study proposes a radiomics-guided dual-channel deep neural network (RDDNN) to improve feature transparency and enhance clinical understanding in the classification of Parkinsonian syndromes. Methods: In this bi-centric study, we analysed two independent cohorts comprising 1,275 patients with idiopathic Parkinson’s disease (IPD), multiple system atrophy (MSA), and progressive supranuclear palsy (PSP), alongside 223 healthy controls from Huashan Hospital and 90 patients with IPD, MSA, and PSP (34IPD, 17MSA, 39PSP) from the University Hospital Munich. It is a re-analysis of well-studied Chinese and German cohorts of ¹⁸ F-fluorodeoxyglucose Positron emission tomography (FDG-PET) imaging of parkinsonian patients and the FDG scans were of 10-min static acquisition at 60 min post FDG injection and normalized against whole brain activity. The RDDNN model combines local features extracted via dilated convolutional networks and global features derived from Transformer-based self-attention networks. Model performance was evaluated using classification metrics and compared to radiomics and DNN approaches. The model’s outputs were also compared with nuclear medicine specialists’ visual assessments to assess interpretability and time efficiency. Furthermore, SHapley Additive Explanations (SHAP), Layer-wise Class Activation Mapping (Layer-CAM), and Rollout Attention Map (RAM) were employed to evaluate which features played the most critical roles in the model’s final classification decisions after supervised training, and to examine how both networks spatially corresponded to known brain connectivity regions. Results: In the internal blind-test cohort, the RDDNN achieved high accuracy (AUC = 0.99, accuracy = 0.98). SHAP and correlation analyses jointly indicated complementary information across channels, some of which were clinically interpretable. In the external cohort, the model maintained robust performance (AUC = 0.94, accuracy = 0.81), with consistent feature patterns across populations. The model significantly reduced evaluation time compared to nuclear medicine specialists’ readings ( p < 0.001), and the heatmaps showed disease-specific activation in anatomically relevant regions for IPD, MSA, and PSP. Conclusion: The RDDNN framework provides a clinically interpretable and reproducible DNN solution for classifying Parkinsonian disorders. By integrating radiomics and attention-based modeling, it enhances lesion localization, supports clinical decision-making, and offers performance comparable to human specialists—while substantially improving diagnostic efficiency.