arXiv:2511.15968v1 Announce Type: new 
Abstract: Multi-task learning can suffer from destructive task interference, where jointly trained models underperform single-task baselines and limit generalization. To improve generalization performance in breast ultrasound-based tumor segmentation via multi-task learning, we propose a novel consistency regularization approach that mitigates destructive interference between segmentation and classification. The consistency regularization approach is composed of differentiable BI-RADS-inspired morphological features. We validated this approach by training all models on the BrEaST dataset (Poland) and evaluating them on three external datasets: UDIAT (Spain), BUSI (Egypt), and BUS-UCLM (Spain). Our comprehensive analysis demonstrates statistically significant (p<0.001) improvements in generalization for segmentation task of the proposed multi-task approach vs. the baseline one: UDIAT, BUSI, BUS-UCLM (Dice coefficient=0.81 vs 0.59, 0.66 vs 0.56, 0.69 vs 0.49, resp.). The proposed approach also achieves state-of-the-art segmentation performance under rigorous external validation on the UDIAT dataset.

تم اقتراح نهج جديد للتعلم متعدد المهام لتجزئة الأورام في تصوير الثدي بالموجات فوق الصوتية، حيث يتناول التدخل المدمر بين المهام. تستخدم هذه الطريقة تنظيم التناسق مع ميزات مورفولوجية قابلة للاشتقاق مستوحاة من BI-RADS. تم التحقق من صحة هذا النهج باستخدام مجموعة بيانات BrEaST في بولندا، مع تقييمات على مجموعات بيانات خارجية من إسبانيا ومصر، مما أظهر تحسينات ذات دلالة إحصائية في أداء التجزئة.

Se ha propuesto un nuevo enfoque de aprendizaje multi-tarea para la segmentación de tumores en ultrasonidos mamarios, abordando la interferencia destructiva entre tareas. Este método utiliza regularización de consistencia con características morfológicas diferenciables inspiradas en BI-RADS. La validación se realizó utilizando el conjunto de datos BrEaST en Polonia, con evaluaciones en conjuntos de datos externos de España y Egipto, mostrando mejoras estadísticamente significativas en el rendimiento de segmentación.

Une nouvelle approche de l'apprentissage multi-tâches pour la segmentation des tumeurs par ultrasons mammaires a été proposée, abordant l'interférence destructrice entre les tâches. Cette méthode utilise une régularisation de cohérence avec des caractéristiques morphologiques différentiables inspirées de BI-RADS. La validation a été effectuée à l'aide du jeu de données BrEaST en Pologne, avec des évaluations sur des ensembles de données externes d'Espagne et d'Égypte, montrant des améliorations statistiquement significatives des performances de segmentation.

A novel approach to multi-task learning for breast ultrasound tumor segmentation has been proposed, addressing destructive task interference. This method employs consistency regularization with differentiable BI-RADS-inspired features. Validation was conducted using the BrEaST dataset in Poland, with evaluations on external datasets from Spain and Egypt, showing statistically significant improvements in segmentation performance.

Externally Validated Multi-Task Learning via Consistency Regularization Using Differentiable BI-RADS Features for Breast Ultrasound Tumor Segmentation

arXiv:2512.02198v1 Announce Type: new 
Abstract: Multifractal analysis has revealed regularities in many self-seeding phenomena, yet its use in modern deep learning remains limited. Existing end-to-end multifractal methods rely on heavy pooling or strong feature-space decimation, which constrain tasks such as semantic segmentation. Motivated by these limitations, we introduce two inductive priors: Monofractal and Multifractal Recalibration. These methods leverage relationships between the probability mass of the exponents and the multifractal spectrum to form statistical descriptions of encoder embeddings, implemented as channel-attention functions in convolutional networks.
  Using a U-Net-based framework, we show that multifractal recalibration yields substantial gains over a baseline equipped with other channel-attention mechanisms that also use higher-order statistics. Given the proven ability of multifractal analysis to capture pathological regularities, we validate our approach on three public medical-imaging datasets: ISIC18 (dermoscopy), Kvasir-SEG (endoscopy), and BUSI (ultrasound).
  Our empirical analysis also provides insights into the behavior of these attention layers. We find that excitation responses do not become increasingly specialized with encoder depth in U-Net architectures due to skip connections, and that their effectiveness may relate to global statistics of instance variability.

تقدم دراسة جديدة طرق إعادة ضبط متعددة الكسريات لتحسين الشبكات العصبية المستخدمة في تقسيم الصور الطبية، وخاصة ضمن إطار U-Net. تتناول هذه الطريقة القيود المفروضة على التقنيات متعددة الكسريات الموجودة التي تعتمد غالبًا على التجميع الثقيل، مما يحسن التمثيل الإحصائي لتضمينات المشفر من خلال وظائف الانتباه القنوي.

Un nuevo estudio presenta métodos de recalibración multifractal para mejorar las redes neuronales utilizadas en la segmentación de imágenes médicas, específicamente dentro de un marco U-Net. Este enfoque aborda las limitaciones de las técnicas multifractales existentes que a menudo dependen de un fuerte pooling, mejorando así la representación estadística de las incrustaciones del codificador a través de funciones de atención por canal.

Une nouvelle étude présente des méthodes de recalibrage multifractal pour améliorer les réseaux neuronaux utilisés dans la segmentation d'images médicales, spécifiquement au sein d'un cadre U-Net. Cette approche répond aux limitations des techniques multifractales existantes qui reposent souvent sur un fort pooling, améliorant ainsi la représentation statistique des embeddings d'encodeur grâce à des fonctions d'attention par canal.

A new study introduces Multifractal Recalibration methods for enhancing neural networks used in medical imaging segmentation, specifically within a U-Net framework. This approach addresses limitations in existing multifractal techniques that often rely on heavy pooling, thereby improving the statistical representation of encoder embeddings through channel-attention functions.

Externally Validated Multi-Task Learning via Consistency Regularization Using Differentiable BI-RADS Features for Breast Ultrasound Tumor Segmentation

Was this article worth reading? Share it

LabEx

LangWatch

Attentive AI