Medical imaging datasets often exhibit heterogeneity from variations in acquisition sites, scanner, and patient demographics, leading to domain shifts that undermine the generalization of AI models. Existing approaches-such as image harmonization, domain adaptation, and domain generalization-typically rely on labeled data (e.g., segmentation masks) or explicit domain metadata (e.g., demographic attributes), which are costly and often unavailable. In this study, we propose Mixture normalization for Unified representation (MixUnify), which harmonizes pixel-level heterogeneity in input images without supervision. MixUnify models the global data distribution via a Gaussian Mixture Model (GMM), under the assumption that each domain follows a local gaussian distribution. The estimated GMM parameters are then used for Mixture Normalization. Notably, the GMM parameters are optimized through a Gaussian Matching loss that leverages feature statistics extracted from the images. We evaluate our method on three public datasets-prostate MRI, cardiac MRI, and retinal fundus images-and demonstrate that MixUnify significantly enhances segmentation performance across diverse domains. Our code is available at https://github.com/Gyeongdeok-Jo/MixUnify.

Due to privacy concerns, obtaining large datasets is challenging in medical image analysis, especially with 3D modalities like Computed Tomography (CT) and Magnetic Resonance Imaging (MRI). Existing generative models, developed to address this issue, often face limitations in output diversity and thus cannot accurately represent 3D medical images. We propose a tumor-generation model that utilizes radiomics features as generative conditions. Radiomics features are high-dimensional handcrafted semantic features that are biologically well-grounded and thus are good candidates for conditioning. Our model employs a GAN-based model to generate tumor masks and a diffusion-based approach to generate tumor texture conditioned on radiomics features. Our method allows the user to generate tumor images according to user-specified radiomics features such as size, shape, and texture at an arbitrary location. This enables the physicians to easily visualize tumor images to better understand tumors according to changing radiomics features. Our approach allows for the removal, manipulation, and repositioning of tumors, generating various tumor types in different scenarios. The model has been tested on tumors in four different organs (kidney, lung, breast, and brain) across CT and MRI. The synthesized images are shown to effectively aid in training for downstream tasks and their authenticity was also evaluated through expert evaluations. Our method has potential usage in treatment planning with diverse synthesized tumors. Our code is available at github.com/jongdoryITS-Radiomics.

White matter hyperintensities (WMHs) are critical indicators of cerebral small vessel disease and are linked to stroke and cognitive decline. Accurate WHM segmentation is challenging due to limited contrast and small size, particularly for deep WMHs where T1-weighted and FLAIR MRI are routinely used. Image fusion offers a method to combine two modalities into one fused modality useful for WHM segmentation. We propose DeAFusion: Detail-Aware Image Fusion, a novel image fusion framework. This framework introduces a Pixel-wise Information Preservation Degree map to enable detailed control of fusion at the pixel level and incorporates segmentation loss to embed semantic information. Our model outperforms existing fusion methods in enhancing lesion visibility and preserving structural details, as demonstrated on both public and in-house datasets. DeAFusion shows promise for improving clinical assessments of WMHs. Our code is available at https://github.com/Gyeongdeok-Jo/DeAFusion.

Multi-modal images play a crucial role in comprehensive evaluations in medical image analysis providing complementary information for identifying clinically important biomarkers. However, in clinical practice, acquiring multiple modalities can be challenging due to reasons such as scan cost, limited scan time, and safety considerations. In this paper, we propose a model based on the latent diffusion model (LDM) that leverages switchable blocks for image-to-image translation in 3D medical images without patch cropping. The 3D LDM combined with conditioning using the target modality allows generating high-quality target modality in 3D overcoming the shortcoming of the missing out-of-slice information in 2D generation methods. The switchable block, noted as multiple switchable spatially adaptive normalization (MS-SPADE), dynamically transforms source latents to the desired style of the target latents to help with the diffusion process. The MS-SPADE block allows us to have one single model to tackle many translation tasks of one source modality to various targets removing the need for many translation models for different scenarios. Our model exhibited successful image synthesis across different source-target modality scenarios and surpassed other models in quantitative evaluations tested on multi-modal brain magnetic resonance imaging datasets of four different modalities and an independent IXI dataset. Our model demonstrated successful image synthesis across various modalities even allowing for one-to-many modality translations. Furthermore, it outperformed other one-to-one translation models in quantitative evaluations. Our code is available at https://github.com/jongdory/ALDM/