Daily Arxiv

This is a page that curates AI-related papers published worldwide.
All content here is summarized using Google Gemini and operated on a non-profit basis.
Copyright for each paper belongs to the authors and their institutions; please make sure to credit the source when sharing.

CoCoLIT: ControlNet-Conditioned Latent Image Translation for MRI to Amyloid PET Synthesis

Created by
  • Haebom

Author

Alec Sargood, Lemuel Puglisi, James H. Cole, Neil P. Oxtoby, Daniele Rav i, Daniel C. Alexander

Outline

This paper presents a cost-effective, large-scale Alzheimer's disease (AD) screening method that synthesizes amyloid PET scans using widely available structural MRI. While MRI does not directly detect amyloid pathology, advanced modeling suggests that it may contain information correlated with amyloid deposition. The high dimensionality and structural complexity of 3D neuroimaging data pose significant challenges for existing MRI-PET translation methods. To address these challenges, we present a diffusion-based latent generative framework, ControlNet-Conditioned Latent Image Translation (CoCoLIT), which simplifies the learning process and enables efficient translation by modeling multimodal relationships in a low-dimensional latent space. CoCoLIT comprises three innovative components: (1) a novel weighted image space loss (WISL) that improves latent representation learning and synthesis quality; (2) a theoretical and experimental analysis of latent mean stabilization (LAS), an existing technique used to improve inference consistency across similar generative models; and (3) ControlNet-based conditioning for MRI-PET translation. We evaluated the performance of CoCoLIT on publicly available datasets and found that it outperformed state-of-the-art methods on both image-based and amyloid-related metrics. Specifically, in amyloid-positivity classification, CoCoLIT outperformed the next-best performing method by 10.5% on our internal dataset and by 23.7% on an external dataset. The code and model are available at https://github.com/brAIn-science/CoCoLIT .

Takeaways, Limitations

•
Takeaways:
◦
MRI presents the potential for cost-effective, large-scale Alzheimer's disease screening.
◦
Presenting an effective multimodal transformation technique for high-dimensional medical image data (CoCoLIT).
◦
A novel potential generation framework that achieves cutting-edge performance (utilizing WISL, LAS, and ControlNet) is presented.
◦
Improved amyloid-positive classification accuracy (+10.5% for internal dataset, +23.7% for external dataset)
◦
Publicly available code and models enable reproducibility and further research.
•
Limitations:
◦
Further research is needed on the generalization performance of the proposed model.
◦
Need for performance evaluation across diverse populations
◦
Analysis of performance variations depending on MRI data quality and acquisition method is needed.
◦
Further research is needed to demonstrate clinical utility (e.g., correlation analysis with clinical diagnoses).
👍
Made with Slashpage