Domain-Specific Latent Representations Improve the Fidelity of Diffusion-Based Medical Image Super-Resolution
Fuente:
arXiv
Salvato in:
| Autori principali: | , , , , , , , , , , |
|---|---|
| Natura: | Preprint |
| Pubblicazione: |
2026
|
| Soggetti: | |
| Accesso online: | |
| Tags: |
Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
|
| _version_ | 1866914470808256512 |
|---|---|
| author | Cajas, Sebastian Judith, Ashaba Gorijavolu, Rahul Kapadia, Sahil Kasimbazi, Hillary Clinton Kinyera, Leo Kwesiga, Emmanuel Paul Manthena, Sri Sri Jaithra Varma Nakayama, Luis Filipe Doreen, Ninsiima Celi, Leo Anthony |
| author_facet | Cajas, Sebastian Judith, Ashaba Gorijavolu, Rahul Kapadia, Sahil Kasimbazi, Hillary Clinton Kinyera, Leo Kwesiga, Emmanuel Paul Manthena, Sri Sri Jaithra Varma Nakayama, Luis Filipe Doreen, Ninsiima Celi, Leo Anthony |
| contents | Latent diffusion models for medical image super-resolution universally inherit variational autoencoders designed for natural photographs. We show that this default choice, not the diffusion architecture, is the dominant constraint on reconstruction quality. In a controlled experiment holding all other pipeline components fixed, replacing the generic Stable Diffusion VAE with MedVAE, a domain-specific autoencoder pretrained on more than 1.6 million medical images, yields +2.91 to +3.29 dB PSNR improvement across knee MRI, brain MRI, and chest X-ray (n = 1,820; Cohen's d = 1.37 to 1.86, all p < 10^{-20}, Wilcoxon signed-rank). Wavelet decomposition localises the advantage to the finest spatial frequency bands encoding anatomically relevant fine structure. Ablations across inference schedules, prediction targets, and generative architectures confirm the gap is stable within plus or minus 0.15 dB, while hallucination rates remain comparable between methods (Cohen's h < 0.02 across all datasets), establishing that reconstruction fidelity and generative hallucination are governed by independent pipeline components. These results provide a practical screening criterion: autoencoder reconstruction quality, measurable without diffusion training, predicts downstream SR performance (R^2 = 0.67), suggesting that domain-specific VAE selection should precede diffusion architecture search. Code and trained model weights are publicly available at https://github.com/sebasmos/latent-sr. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2604_12152 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | Domain-Specific Latent Representations Improve the Fidelity of Diffusion-Based Medical Image Super-Resolution Cajas, Sebastian Judith, Ashaba Gorijavolu, Rahul Kapadia, Sahil Kasimbazi, Hillary Clinton Kinyera, Leo Kwesiga, Emmanuel Paul Manthena, Sri Sri Jaithra Varma Nakayama, Luis Filipe Doreen, Ninsiima Celi, Leo Anthony Computer Vision and Pattern Recognition Artificial Intelligence Latent diffusion models for medical image super-resolution universally inherit variational autoencoders designed for natural photographs. We show that this default choice, not the diffusion architecture, is the dominant constraint on reconstruction quality. In a controlled experiment holding all other pipeline components fixed, replacing the generic Stable Diffusion VAE with MedVAE, a domain-specific autoencoder pretrained on more than 1.6 million medical images, yields +2.91 to +3.29 dB PSNR improvement across knee MRI, brain MRI, and chest X-ray (n = 1,820; Cohen's d = 1.37 to 1.86, all p < 10^{-20}, Wilcoxon signed-rank). Wavelet decomposition localises the advantage to the finest spatial frequency bands encoding anatomically relevant fine structure. Ablations across inference schedules, prediction targets, and generative architectures confirm the gap is stable within plus or minus 0.15 dB, while hallucination rates remain comparable between methods (Cohen's h < 0.02 across all datasets), establishing that reconstruction fidelity and generative hallucination are governed by independent pipeline components. These results provide a practical screening criterion: autoencoder reconstruction quality, measurable without diffusion training, predicts downstream SR performance (R^2 = 0.67), suggesting that domain-specific VAE selection should precede diffusion architecture search. Code and trained model weights are publicly available at https://github.com/sebasmos/latent-sr. |
| title | Domain-Specific Latent Representations Improve the Fidelity of Diffusion-Based Medical Image Super-Resolution |
| topic | Computer Vision and Pattern Recognition Artificial Intelligence |
| url | https://arxiv.org/abs/2604.12152 |