DisQ-HNet: A Disentangled Quantized Half-UNet for Interpretable Multimodal Image Synthesis Applications to Tau-PET Synthesis from T1 and FLAIR MRI
#DisQ-HNet #tau-PET synthesis #MRI-based alternatives #Alzheimer's disease #multimodal imaging #Partial Information Decomposition #vector quantization #Half-UNet decoder
📌 Key Takeaways
- DisQ-HNet synthesizes tau-PET images from MRI scans, addressing cost and availability limitations
- The framework exposes how each imaging modality contributes to predictions through interpretable AI
- Outperforms existing methods in preserving disease-relevant signals for Alzheimer's diagnostics
- Provides modality-specific attribution through PID-based Shapley analysis
📖 Full Retelling
Researchers led by Agamdeep S. Chopra introduced DisQ-HNet (DQH), a novel AI framework that synthesizes tau-PET images from paired T1-weighted and FLAIR MRI scans on February 26, 2026, addressing the high cost and limited availability of tau-PET scans which are crucial for Alzheimer's disease pathology detection. The innovative approach combines a Partial Information Decomposition (PID)-guided, vector-quantized encoder that partitions latent information into redundant, unique, and complementary components, with a Half-UNet decoder that preserves anatomical detail using pseudo-skip connections conditioned on structural edge cues rather than direct encoder feature reuse. This methodology not only generates synthetic tau-PET images but also provides interpretability by exposing how each imaging modality contributes to the final prediction, addressing a critical gap in previous deep learning approaches for medical imaging. Across multiple baseline models including VAE, VQ-VAE, and UNet, DisQ-HNet demonstrated superior performance in maintaining reconstruction fidelity while better preserving disease-relevant signals essential for downstream Alzheimer's disease tasks such as Braak staging, tau localization, and classification. The PID-based Shapley analysis further enables modality-specific attribution of synthesized uptake patterns, offering unprecedented transparency into the AI decision-making process for medical diagnostics.
🏷️ Themes
Medical AI, Alzheimer's research, Medical imaging
📚 Related People & Topics
Partial information decomposition
Partial Information Decomposition is an extension of information theory, that aims to generalize the pairwise relations described by information theory to the interaction of multiple variables. It has been shown, however, that Partial Information Decomposition is mathematically not an extension of c...
Entity Intersection Graph
No entity connections available yet for this article.
Mentioned Entities
Original Source
--> Computer Science > Computer Vision and Pattern Recognition arXiv:2602.22545 [Submitted on 26 Feb 2026] Title: DisQ-HNet: A Disentangled Quantized Half-UNet for Interpretable Multimodal Image Synthesis Applications to Tau-PET Synthesis from T1 and FLAIR MRI Authors: Agamdeep S. Chopra , Caitlin Neher , Tianyi Ren , Juampablo E. Heras Rivera , Mehmet Kurt View a PDF of the paper titled DisQ-HNet: A Disentangled Quantized Half-UNet for Interpretable Multimodal Image Synthesis Applications to Tau-PET Synthesis from T1 and FLAIR MRI, by Agamdeep S. Chopra and 4 other authors View PDF HTML Abstract: Tau positron emission tomography (tau-PET) provides an in vivo marker of Alzheimer's disease pathology, but cost and limited availability motivate MRI-based alternatives. We introduce DisQ-HNet , a framework that synthesizes tau-PET from paired T1-weighted and FLAIR MRI while exposing how each modality contributes to the prediction. The method combines a Partial Information Decomposition -guided, vector-quantized encoder that partitions latent information into redundant, unique, and complementary components, and a Half-UNet decoder that preserves anatomical detail using pseudo-skip connections conditioned on structural edge cues rather than direct encoder feature reuse. Across multiple baselines (VAE, VQ-VAE, and UNet), DisQ-HNet maintains reconstruction fidelity and better preserves disease-relevant signal for downstream AD tasks, including Braak staging, tau localization, and classification. PID-based Shapley analysis provides modality-specific attribution of synthesized uptake patterns. Comments: 14 pages, 8 figures, 8 tables; includes PID guided vector quantized latent factorization and sobel edge conditioned Half-UNet decoder Subjects: Computer Vision and Pattern Recognition (cs.CV) ; Artificial Intelligence (cs.AI) ACM classes: I.2.10; I.4.8 Cite as: arXiv:2602.22545 [cs.CV] (or arXiv:2602.22545v1 [cs.CV] for this version) https://doi.org/10.48550/arXiv.2602.22545 Fo...
Read full article at source