Latent diffusion improves ct to pet image translation with limited data
Domain-Adapted Latent Diffusion Models for CT-to-PET Image Translation
Computer Vision and Pattern Recognition
Summary
PET scans give important info about tumors but are expensive and not always available, unlike CT scans. The authors developed a new computer method that helps turn CT images into PET-like images using fewer examples than before. They also found better ways to measure how accurate these converted images are, focusing on parts where tumors appear. Their method improves on previous ones by borrowing knowledge from general images and adapting it for medical use.
What this means in practice
- •For radiology software developers: Create software that generates PET-like images from CT scans when PET data is unavailable, improving diagnostic options with fewer matched training samples.
- •For medical imaging hardware teams: Incorporate latent diffusion methods into imaging devices to offer enhanced tumor visualization from CT data without extra PET scanning hardware.
Authors
Yann Bourdé, Lea Bogensperger, Arnela Hadzic, Michael Krauthammer
Abstract
Positron Emission Tomography (PET) provides important metabolic information for tumor detection and staging but remains significantly less accessible than Computed Tomography (CT) due to its high cost, infrastructure requirements, and reliance on injected radioactive tracers. CT-to-PET image translation using generative models aims to address this disparity, yet state-of-the-art diffusion-based methods are limited by the scarcity of paired training data. To overcome this, we propose a domain-adapted latent diffusion framework that transfers natural-image priors using a relatively small dataset of 21,930 paired CT/PET slices, incorporating a signed power transform to resolve extreme intensity sparsity in PET data. Compared to CPDM, a state-ofthe-art baseline for CT-to-PET image translation, our approach yields superior performance in pixel-wise reconstruction and localized tumor fidelity, demonstrating effective transfer of natural-image priors to the medical domain. We further provide a systematic analysis of evaluation metric reliability for sparse PET data, showing that standard full-image metrics can be misleading indicators of translated image quality, and propose background-filtered and mask-restricted metrics as more clinically meaningful alternatives.