Lightening Anything in Medical Images

TL;DR

UniMIE tackles suboptimal illumination in medical images by introducing a training-free, diffusion-prior-based enhancer that operates without medical-domain fine-tuning. It leverages a single DDPM pre-trained on ImageNet, applying a patch-based, any-size enhancement with a learnable degradation mask and gradient-guided conditioning to align outputs with degraded inputs. Across 15 datasets and 13 modalities, UniMIE achieves superior image quality and maintains or improves downstream tasks such as segmentation and detection, often outperforming modality-specific or unsupervised baselines. The method enables broad, deployment-friendly enhancement of medical images without task-specific training, though it trades some speed for quality at high resolutions; future work will push higher-resolution diffusion models and hardware acceleration.

Abstract

The development of medical imaging techniques has made a significant contribution to clinical decision-making. However, the existence of suboptimal imaging quality, as indicated by irregular illumination or imbalanced intensity, presents significant obstacles in automating disease screening, analysis, and diagnosis. Existing approaches for natural image enhancement are mostly trained with numerous paired images, presenting challenges in data collection and training costs, all while lacking the ability to generalize effectively. Here, we introduce a pioneering training-free Diffusion Model for Universal Medical Image Enhancement, named UniMIE. UniMIE demonstrates its unsupervised enhancement capabilities across various medical image modalities without the need for any fine-tuning. It accomplishes this by relying solely on a single pre-trained model from ImageNet. We conduct a comprehensive evaluation on 13 imaging modalities and over 15 medical types, demonstrating better qualities, robustness, and accuracy than other modality-specific and data-inefficient models. By delivering high-quality enhancement and corresponding accuracy downstream tasks across a wide range of tasks, UniMIE exhibits considerable potential to accelerate the advancement of diagnostic tools and customized treatment plans.

Figures (6)

• Figure 1: UniMIE is pre-trained on ImageNet but can handle diverse enhancement tasks. The generative prior inherent in UniMIE can be generalized to any medical image modality even if it is not in the training set.
• Figure 2: Principle of UniMIE.a, The enhancement process of our UniMIE. b, An illustration of the forward process and the reverse process in an unconditional diffusion model. c, An overview of the patch-based pipeline utilized for medical image enhancement. d, An illustration of the mean estimated noise-guided sampling updates for overlapping pixels across patches, where the ratio of patch size to the overlapping region is $r = p/2$. Specifically, the grid cell highlighted with a white border is shared by only four overlapping patches. Consequently, sampling updates are performed for the pixels within this region at each time step $t$. These updates consider the average estimated noise across the four overlapping patches.
• Figure 3: a, Comparison of noise artifact of enhanced corneal confocal microscopy images from CORN dataset. b, An illustrative example to demonstrate the selection of background regions for calculating the SNR. The background regions, depicted in green, were identified by applying a disk-shaped dilation manipulation on the manually traced fibers, which are represented in red. The dilation operation was performed using radii of 1, 3, 5, 7, and 9 pixels. The top row displays the low-quality image, while the bottom row showcases the image enhanced by UniMIE. c, An exemplification of enhanced CCM medical images. and corresponding nerve fiber segmentation performance. d-h, SNR results of the low-quality and enhanced CCM medical images by various methodologies. i, Segmentation performance of enhanced CCM medical images using various methods in terms of ACC, SEN, SP, AUC, F1, IU, and DICE.
• Figure 4: Comparison of texture details of enhanced color fundus images: a, DR and b, AMD eye diseases. High-quality score performance (mean $\pm$ standard deviation) of different enhancement methods on enhanced color fundus images: c, DR and d, AMD eye diseases. e, A qualitative examination is performed to compare the distribution of illumination in enhanced Endoscene images. f-i, Comparison of image quality metrics for different enhancement methods.
• Figure 5: a, The detection results of COVID-19 after enhancement. b, Average precision of segmentation results on enhanced medical images. c, Segmentation results on enhanced images from BrainWeb and HVSMR. d, Comparisons of PA, JS, and Dice metrics on BrainWeb and HVSMR.