arXiv:2511.14649v1 Announce Type: new 
Abstract: Accurate airway segmentation from chest computed tomography (CT) scans is essential for quantitative lung analysis, yet manual annotation is impractical and many automated U-Net-based methods yield disconnected components that hinder reliable biomarker extraction. We present RepAir, a three-stage framework for robust 3D airway segmentation that combines an nnU-Net-based network with anatomically informed topology correction. The segmentation network produces an initial airway mask, after which a skeleton-based algorithm identifies potential discontinuities and proposes reconnections. A 1D convolutional classifier then determines which candidate links correspond to true anatomical branches versus false or obstructed paths. We evaluate RepAir on two distinct datasets: ATM'22, comprising annotated CT scans from predominantly healthy subjects and AeroPath, encompassing annotated scans with severe airway pathology. Across both datasets, RepAir outperforms existing 3D U-Net-based approaches such as Bronchinet and NaviAirway on both voxel-level and topological metrics, and produces more complete and anatomically consistent airway trees while maintaining high segmentation accuracy.

RepAir هو إطار جديد مصمم لت segmenting مجرى الهواء بدقة من أشعة الصدر المقطعية، حيث يتناول قيود التوصيف اليدوي والأساليب الآلية الحالية التي غالبًا ما تنتج مكونات مفصولة. يجمع هذا النهج المكون من ثلاث مراحل بين شبكة تعتمد على nnU-Net وتصحيح الطوبولوجيا المعتمدة على التشريح، مما يسمح بتحديد الانقطاعات واقتراح إعادة الاتصال. تم تقييم الإطار على مجموعتين من البيانات، مما يوضح فعاليته في كل من الموضوعات الصحية وتلك التي تعاني من أمراض شديدة في مجرى الهواء.

RepAir es un nuevo marco diseñado para la segmentación precisa de las vías respiratorias a partir de escaneos CT torácicos, abordando las limitaciones de la anotación manual y los métodos automatizados existentes que a menudo producen componentes desconectados. Este enfoque de tres etapas integra una red basada en nnU-Net con corrección topológica anatómica, permitiendo la identificación de discontinuidades y la propuesta de reconexiones. El marco fue evaluado en dos conjuntos de datos, demostrando su efectividad tanto en sujetos sanos como en aquellos con patología respiratoria severa.

RepAir est un nouveau cadre conçu pour une segmentation précise des voies respiratoires à partir de scans CT thoraciques, répondant aux limites de l'annotation manuelle et des méthodes automatisées existantes qui produisent souvent des composants déconnectés. Cette approche en trois étapes intègre un réseau basé sur nnU-Net avec correction topologique anatomique, permettant d'identifier les discontinuités et de proposer des reconnections. Le cadre a été évalué sur deux ensembles de données, démontrant son efficacité tant chez les sujets sains que chez ceux présentant une pathologie respiratoir…

RepAir is a new framework designed for accurate airway segmentation from chest CT scans, addressing the limitations of manual annotation and existing automated methods that often produce disconnected components. This three-stage approach integrates an nnU-Net-based network with anatomical topology correction, enabling the identification of discontinuities and the proposal of reconnections. The framework was evaluated on two datasets, demonstrating its effectiveness in both healthy subjects and those with severe airway pathology.

RepAir: A Framework for Airway Segmentation and Discontinuity Correction in CT

arXiv:2511.18232v1 Announce Type: new 
Abstract: Magnetic Resonance Imaging (MRI) acquisitions require extensive scan times, limiting patient throughput and increasing susceptibility to motion artifacts. Accelerated parallel MRI techniques reduce acquisition time by undersampling k-space data, but require robust reconstruction methods to recover high-quality images. Traditional approaches like SENSE require both undersampled k-space data and pre-computed coil sensitivity maps. We propose an end-to-end deep learning framework that jointly estimates coil sensitivity maps and reconstructs images from only undersampled k-space measurements at 4x acceleration. Our two-module architecture consists of a Coil Sensitivity Map (CSM) estimation module and a U-Net-based MRI reconstruction module. We evaluate our method on multi-coil brain MRI data from 10 subjects with 8 echoes each, using 2x SENSE reconstructions as ground truth. Our approach produces visually smoother reconstructions compared to conventional SENSE output, achieving comparable visual quality despite lower PSNR/SSIM metrics. We identify key challenges including spatial misalignment between different acceleration factors and propose future directions for improved reconstruction quality.

تم اقتراح إطار جديد للتعلم العميق لتصوير الرنين المغناطيسي (MRI) يمكّن من إعادة البناء المتوازي من عمليات الاستحواذ المعجلة بمعدل 4x، مما يقلل بشكل كبير من أوقات المسح مع الحفاظ على جودة الصورة. تستخدم هذه الطريقة بنية ذات وحدتين تقدر خرائط حساسية الملف وتعيد بناء الصور من بيانات k-space غير المأخوذة بعين الاعتبار، مما يعالج قيود التقنيات التقليدية مثل SENSE.

Se ha propuesto un nuevo marco de aprendizaje profundo para la imagen por resonancia magnética (IRM) que permite la reconstrucción paralela a partir de adquisiciones aceleradas 4x, reduciendo significativamente los tiempos de escaneo mientras mantiene la calidad de la imagen. Este método utiliza una arquitectura de dos módulos que estima mapas de sensibilidad de bobinas y reconstruye imágenes a partir de datos k-space submuestreados, abordando las limitaciones de técnicas tradicionales como SENSE.

Un nouveau cadre d'apprentissage profond a été proposé pour l'imagerie par résonance magnétique (IRM) qui permet la reconstruction parallèle à partir d'acquisitions accélérées 4x, réduisant considérablement les temps de scan tout en maintenant la qualité de l'image. Cette méthode utilise une architecture à deux modules qui estime les cartes de sensibilité des bobines et reconstruit des images à partir de données k-space sous-échantillonnées, répondant aux limitations des techniques traditionnelles comme SENSE.

A new deep learning framework has been proposed for Magnetic Resonance Imaging (MRI) that enables parallel reconstruction from 4x accelerated acquisitions, significantly reducing scan times while maintaining image quality. This method utilizes a two-module architecture that estimates coil sensitivity maps and reconstructs images from undersampled k-space data, addressing the limitations of traditional techniques like SENSE.

Parallel qMRI Reconstruction from 4x Accelerated Acquisitions

arXiv:2503.18414v2 Announce Type: replace 
Abstract: Representation Alignment (REPA) that aligns Diffusion Transformer (DiT) hidden-states with ViT visual encoders has proven highly effective in DiT training, demonstrating superior convergence properties, but it has not been validated on the canonical diffusion U-Net architecture that shows faster convergence compared to DiTs. However, adapting REPA to U-Net architectures presents unique challenges: (1) different block functionalities necessitate revised alignment strategies; (2) spatial-dimension inconsistencies emerge from U-Net's spatial downsampling operations; (3) space gaps between U-Net and ViT hinder the effectiveness of tokenwise alignment. To encounter these challenges, we propose \textbf{U-REPA}, a representation alignment paradigm that bridges U-Net hidden states and ViT features as follows: Firstly, we propose via observation that due to skip connection, the middle stage of U-Net is the best alignment option. Secondly, we propose upsampling of U-Net features after passing them through MLPs. Thirdly, we observe difficulty when performing tokenwise similarity alignment, and further introduces a manifold loss that regularizes the relative similarity between samples. Experiments indicate that the resulting U-REPA could achieve excellent generation quality and greatly accelerates the convergence speed. With CFG guidance interval, U-REPA could reach $FID<1.5$ in 200 epochs or 1M iterations on ImageNet 256 $\times$ 256, and needs only half the total epochs to perform better than REPA under sd-vae-ft-ema. Codes: https://github.com/YuchuanTian/U-REPA

تهدف U-REPA، وهي نموذج لتوافق التمثيل، إلى محاذاة شبكات U-Nets للتشتت مع مشفرات ViT البصرية، مع معالجة التحديات الفريدة التي تطرحها هياكل U-Net. يعتبر هذا التطور مهمًا لأنه يعزز كفاءة تدريب نماذج التشتت، والتي تعتبر حاسمة لمجموعة متنوعة من تطبيقات الذكاء الاصطناعي، خاصة في توليد ومعالجة الصور.

La introducción de U-REPA, un paradigma de alineación de representación, tiene como objetivo alinear los U-Nets de difusión con los codificadores visuales ViT, abordando los desafíos únicos que presentan las arquitecturas U-Net. Este desarrollo es significativo ya que mejora la eficiencia del entrenamiento de los modelos de difusión, que son cruciales para diversas aplicaciones de IA, especialmente en la generación y procesamiento de imágenes.

L'introduction de U-REPA, un paradigme d'alignement de représentation, vise à aligner les U-Nets de diffusion avec les encodeurs visuels ViT, en abordant les défis uniques posés par les architectures U-Net. Ce développement est significatif car il améliore l'efficacité de l'entraînement des modèles de diffusion, qui sont cruciaux pour diverses applications d'IA, en particulier dans la génération et le traitement d'images.

The introduction of U-REPA, a representation alignment paradigm, aims to align Diffusion U-Nets with ViT visual encoders, addressing the unique challenges posed by U-Net architectures. This development is significant as it enhances the training efficiency of diffusion models, which are crucial for various AI applications, particularly in image generation and processing.

U-REPA: Aligning Diffusion U-Nets to ViTs

arXiv:2501.18921v2 Announce Type: replace-cross 
Abstract: The U-Net architecture and its variants have remained state-of-the-art (SOTA) for retinal vessel segmentation over the past decade. In this study, we introduce a Full-Scale Guided Network (FSG-Net), where a novel feature representation module using modernized convolution blocks effectively captures full-scale structural information, while a guided convolution block subsequently refines this information. Specifically, we introduce an attention-guided filter within the guided convolution block, leveraging its similarity to unsharp masking to enhance fine vascular structures. Passing full-scale information to the attention block facilitates the generation of more contextually relevant attention maps, which are then passed to the attention-guided filter, providing further refinement to the segmentation performance. The structure preceding the guided convolution block can be replaced by any U-Net variant, ensuring flexibility and scalability across various segmentation tasks. For a fair comparison, we re-implemented recent studies available in public repositories to evaluate their scalability and reproducibility. Our experiments demonstrate that, despite its compact architecture, FSG-Net delivers performance competitive with SOTA methods across multiple public datasets. Ablation studies further demonstrate that each proposed component meaningfully contributes to this competitive performance. Our code is available on https://github.com/ZombaSY/FSG-Net-pytorch.

قدمت دراسة جديدة شبكة التوجيه على نطاق واسع (FSG-Net) التي تعزز من تقسيم الأوعية الدموية الشبكية من خلال استخدام وحدة تمثيل ميزات جديدة وفلتر موجه بالانتباه لتحسين دقة الكشف عن الهياكل الوعائية. يعتمد هذا النهج على بنية U-Net المعروفة، مما يضمن مرونة في التنفيذ.

Un nuevo estudio ha presentado la Red Guiada a Gran Escala (FSG-Net), que mejora la segmentación de los vasos retinianos mediante un nuevo módulo de representación de características y un filtro guiado por atención para aumentar la precisión en la detección de estructuras vasculares. Este enfoque se basa en la arquitectura U-Net establecida, asegurando flexibilidad en su implementación.

Une nouvelle étude a introduit le Réseau Guidé à Grande Échelle (FSG-Net), qui améliore la segmentation des vaisseaux rétiniens en utilisant un nouveau module de représentation des caractéristiques et un filtre guidé par attention pour améliorer la précision de la détection des structures vasculaires. Cette approche s'appuie sur l'architecture U-Net établie, garantissant une flexibilité dans l'implémentation.

A new study has introduced the Full-Scale Guided Network (FSG-Net), which enhances retinal vessel segmentation by utilizing a novel feature representation module and an attention-guided filter to improve the accuracy of vascular structure detection. This approach builds on the established U-Net architecture, ensuring flexibility in implementation.

RepAir: A Framework for Airway Segmentation and Discontinuity Correction in CT

Was this article worth reading? Share it

MatosShere

Supanote

Zemith-3bda3b