View in EDS

Generative morphological alignment of multi-view mammograms via spatially-conditioned diffusion.

Saved in:
Bibliographic Details
Title: Generative morphological alignment of multi-view mammograms via spatially-conditioned diffusion.
Authors: Tang, Qingfeng1 (AUTHOR) tqf0913@aqnu.edu.cn, Ding, Pengcheng1 (AUTHOR) Y24070017@stu.aqnu.edu.cn, Zhang, Liangliang1 (AUTHOR) 18237145173@163.com, Dai, Guowei2 (AUTHOR) daigw@stu.scu.edu.cn, Hamilton, Matthew3 (AUTHOR) mhamilton@mun.ca, An, Hui1,4 (AUTHOR) anvui@126.com
Source: Engineering Applications of Artificial Intelligence. May2026, Vol. 172, pN.PAG-N.PAG. 1p.
Subjects: Mammograms, Morphology, Generative artificial intelligence, Diagnostic imaging, Deformations (Mechanics), Artificial neural networks
Abstract: Analyzing multi-view mammograms, specifically the craniocaudal (CC) and mediolateral oblique (MLO) views, poses a significant challenge due to the complex non-rigid breast tissue deformation between acquisitions. Most computational methods overlook this spatial correspondence, limiting their ability to learn morphologically-aware representations. To address this, we introduce Morph-Gen, a novel generative framework that explicitly models the morphological alignment between mammogram views via a spatially-conditioned diffusion model. Our approach learns to synthesize a target view conditioned on a given source view and its associated radiological report. The synthesis process is meticulously guided by a trio of conditions: visual features from the source view, localized semantic embeddings extracted from the text report, and a dynamically predicted dense deformation field. This geometric condition, which captures the non-rigid spatial transformation, is learned by a dedicated Deformable Geometry Network optimized via a self-supervised objective. These multimodal conditions are seamlessly injected into the diffusion U-Net using cross-attention, providing fine-grained guidance. By compelling the model to understand and execute this complex anatomical transformation, Morph-Gen learns a deeply disentangled latent space that captures morphological invariants and pathological abnormalities, providing a powerful foundation for downstream diagnostic tasks. [ABSTRACT FROM AUTHOR]
Copyright of Engineering Applications of Artificial Intelligence is the property of Pergamon Press - An Imprint of Elsevier Science and its content may not be copied or emailed to multiple sites without the copyright holder's express written permission. Additionally, content may not be used with any artificial intelligence tools or machine learning technologies. However, users may print, download, or email articles for individual use. This abstract may be abridged. No warranty is given about the accuracy of the copy. Users should refer to the original published version of the material for the full abstract. (Copyright applies to all Abstracts.)
Database: Engineering Source
Description
Abstract:Analyzing multi-view mammograms, specifically the craniocaudal (CC) and mediolateral oblique (MLO) views, poses a significant challenge due to the complex non-rigid breast tissue deformation between acquisitions. Most computational methods overlook this spatial correspondence, limiting their ability to learn morphologically-aware representations. To address this, we introduce Morph-Gen, a novel generative framework that explicitly models the morphological alignment between mammogram views via a spatially-conditioned diffusion model. Our approach learns to synthesize a target view conditioned on a given source view and its associated radiological report. The synthesis process is meticulously guided by a trio of conditions: visual features from the source view, localized semantic embeddings extracted from the text report, and a dynamically predicted dense deformation field. This geometric condition, which captures the non-rigid spatial transformation, is learned by a dedicated Deformable Geometry Network optimized via a self-supervised objective. These multimodal conditions are seamlessly injected into the diffusion U-Net using cross-attention, providing fine-grained guidance. By compelling the model to understand and execute this complex anatomical transformation, Morph-Gen learns a deeply disentangled latent space that captures morphological invariants and pathological abnormalities, providing a powerful foundation for downstream diagnostic tasks. [ABSTRACT FROM AUTHOR]
ISSN:09521976
DOI:10.1016/j.engappai.2026.114344