arXiv:2510.22630v2 Announce Type: replace 
Abstract: Atypical mitotic figures are important biomarkers of tumor aggressiveness in histopathology, yet reliable recognition remains challenging due to severe class imbalance and variability across imaging domains. We present a DenseNet-121-based framework tailored for atypical mitosis classification in the MIDOG 2025 (Track 2) setting. Our method integrates stain-aware augmentation (Macenko), geometric and intensity transformations, and imbalance-aware learning via weighted sampling with a hybrid objective combining class-weighted binary cross-entropy and focal loss. Trained end-to-end with AdamW and evaluated across multiple independent domains, the model demonstrates strong generalization under scanner and staining shifts, achieving balanced accuracy 85.0%, AUROC 0.927, sensitivity 89.2%, and specificity 80.9% on the official test set. These results indicate that combining DenseNet-121 with stain-aware augmentation and imbalance-adaptive objectives yields a robust, domain-generalizable framework for atypical mitosis classification suitable for real-world computational pathology workflows.

تم تطوير إطار جديد باستخدام DenseNet-121 لتحسين تصنيف الأشكال الانقسامية غير النمطية، والتي تعتبر مؤشرات حاسمة لعدوانية الأورام في علم الأمراض النسيجية. تتناول هذه الطريقة التحديات مثل عدم التوازن في الفئات وتنوع الصور، مما يجعلها تقدمًا كبيرًا للباحثين والأطباء. من خلال دمج تعزيزات واعية للصبغة وتحولات هندسية وشدة، تهدف هذه الطريقة إلى تحسين موثوقية التعرف على الانقسام، مما يساعد في النهاية على تحسين تشخيص وعلاج السرطان.

Se ha desarrollado un nuevo marco utilizando DenseNet-121 para mejorar la clasificación de figuras mitóticas atípicas, que son indicadores cruciales de la agresividad tumoral en histopatología. Este método aborda desafíos como el desequilibrio de clases y la variabilidad en las imágenes, lo que lo convierte en un avance significativo para investigadores y clínicos. Al incorporar aumentación consciente del tinte y diversas transformaciones, este enfoque busca mejorar la fiabilidad del reconocimiento de mitosis, ayudando en última instancia a un mejor diagnóstico y tratamiento del cáncer.

Un nouveau cadre utilisant DenseNet-121 a été développé pour améliorer la classification des figures mitotiques atypiques, qui sont des indicateurs cruciaux de l'agressivité tumorale en histopathologie. Cette méthode répond à des défis tels que le déséquilibre des classes et la variabilité des images, ce qui en fait une avancée significative pour les chercheurs et les cliniciens. En intégrant une augmentation sensible aux teintures et diverses transformations, cette approche vise à améliorer la fiabilité de la reconnaissance de la mitose, aidant ainsi à un meilleur diagnostic et traitement du cancer.

A new framework using DenseNet-121 has been developed to improve the classification of atypical mitotic figures, which are crucial indicators of tumor aggressiveness in histopathology. This method addresses challenges like class imbalance and variability in imaging, making it a significant advancement for researchers and clinicians. By incorporating stain-aware augmentation and various transformations, this approach aims to enhance the reliability of mitosis recognition, ultimately aiding in better cancer diagnosis and treatment.

Robust Atypical Mitosis Classification with DenseNet121: Stain-Aware Augmentation and Hybrid Loss for Domain Generalization

arXiv:2601.08127v1 Announce Type: new 
Abstract: The development of robust artificial intelligence models for histopathology diagnosis is severely constrained by the scarcity of expert-annotated lesion data, particularly for rare pathologies and underrepresented disease subtypes. While data augmentation offers a potential solution, existing methods fail to generate sufficiently realistic lesion morphologies that preserve the complex spatial relationships and cellular architectures characteristic of histopathological tissues. Here we present PathoGen, a diffusion-based generative model that enables controllable, high-fidelity inpainting of lesions into benign histopathology images. Unlike conventional augmentation techniques, PathoGen leverages the iterative refinement process of diffusion models to synthesize lesions with natural tissue boundaries, preserved cellular structures, and authentic staining characteristics. We validate PathoGen across four diverse datasets representing distinct diagnostic challenges: kidney, skin, breast, and prostate pathology. Quantitative assessment confirms that PathoGen outperforms state-of-the-art generative baselines, including conditional GAN and Stable Diffusion, in image fidelity and distributional similarity. Crucially, we show that augmenting training sets with PathoGen-synthesized lesions enhances downstream segmentation performance compared to traditional geometric augmentations, particularly in data-scarce regimes. Besides, by simultaneously generating realistic morphology and pixel-level ground truth, PathoGen effectively overcomes the manual annotation bottleneck. This approach offers a scalable pathway for developing generalizable medical AI systems despite limited expert-labeled data.

يمثل تقديم PathoGen، وهو نموذج توليدي يعتمد على الانتشار، تقدمًا كبيرًا في تركيب الآفات الواقعية في صور علم الأمراض النسيجية، حيث يعالج النقص الحاد في بيانات الآفات المعلّمة من قبل الخبراء، خاصةً بالنسبة للأمراض النادرة. يعزز هذا النموذج عملية إدخال الآفات في الصور الحميدة مع الحفاظ على الحدود الطبيعية للأنسجة والهياكل الخلوية.

La introducción de PathoGen, un modelo generativo basado en difusión, marca un avance significativo en la síntesis de lesiones realistas en imágenes de histopatología, abordando la escasez crítica de datos de lesiones anotados por expertos, especialmente para patologías raras. Este modelo mejora la inpainting de lesiones en imágenes benignas mientras preserva los límites naturales de los tejidos y las estructuras celulares.

L'introduction de PathoGen, un modèle génératif basé sur la diffusion, représente une avancée significative dans la synthèse de lésions réalistes dans les images d'histopathologie, répondant à la pénurie critique de données de lésions annotées par des experts, en particulier pour les pathologies rares. Ce modèle améliore l'inpainting de lésions dans des images bénignes tout en préservant les frontières naturelles des tissus et les structures cellulaires.

The introduction of PathoGen, a diffusion-based generative model, marks a significant advancement in the synthesis of realistic lesions in histopathology images, addressing the critical shortage of expert-annotated lesion data, especially for rare pathologies. This model enhances the inpainting of lesions into benign images while preserving natural tissue boundaries and cellular structures.

Robust Atypical Mitosis Classification with DenseNet121: Stain-Aware Augmentation and Hybrid Loss for Domain Generalization

Was this article worth reading? Share it

LucidQuery AI

WasItAI

stone identification

Attentive AI

GPTHumanizer

Supametas.AI

Ready to build your own newsroom?