DeNuC: Decoupling Nuclei Detection and Classification in Histopathology

Zijiang Yang; Chen Kuang; Dongmei Fu

cs.CVMar 4, 2026

DeNuC: Decoupling Nuclei Detection and Classification in Histopathology

Authors: Zijiang Yang, Chen Kuang, Dongmei Fu

Organizations: School of Automation and Electrical Engineering, University of Science and Technology Beijing, Beijing 100083, China · Beijing Engineering Research Center of Industrial Spectrum Imaging

Abstract

Pathology Foundation Models (FMs) have shown strong performance across a wide range of pathology image representation and diagnostic tasks. However, FMs do not exhibit the expected performance advantage over traditional specialized models in Nuclei Detection and Classification (NDC). In this work, we reveal that jointly optimizing nuclei detection and classification leads to severe representation degradation in FMs. Moreover, we identify that the substantial intrinsic disparity in task difficulty between nuclei detection and nuclei classification renders joint NDC optimization unnecessarily computationally burdensome for the detection stage. To address these challenges, we propose DeNuC, a simple yet effective method designed to break through existing bottlenecks by Decoupling Nuclei detection and Classification. DeNuC employs a lightweight model for accurate nuclei localization, subsequently leveraging a pathology FM to encode input images and query nucleus-specific features based on the detected coordinates for classification. Extensive experiments on three widely used benchmarks demonstrate that DeNuC effectively unlocks the representational potential of FMs for NDC and significantly outperforms state-of-the-art methods. Notably, DeNuC improves F1 scores by 4.2% and 3.6% (or higher) on the BRCAM2C and PUMA datasets, respectively, while using only 16% (or fewer) trainable parameters compared to other methods. Code is available at https://github.com/ZijiangY1116/DeNuC.

Figures & tables

Figure 1: Analysis of model performance during the NDC training. (a) We evaluate the representation capability of UNI2-H [ 2 ] during joint optimization for NDC on the OCELOT [ 13 ] dataset via linear probing. UNI2-H undergoes a severe representation degradation in the early training phase. Although the performance subsequently recovers, it fails to regain its initial optimal level and rapidly deteriorates into over-fitting. (b) We evaluate the detection and classification F1 scores of a ConvNeXt-S [ 8 ] model pre-trained solely on ImageNet-1K throughout the training process on PUMA [ 14 ] . Despite lacking domain-specific pathology pre-training, nuclei detection converges approximately 2.3

\times

faster than classification, highlighting the inherent difficulty disparity between the two tasks.

Figure 2: Comparison of DeNuC and SOTA methods. DeNuC achieves significantly superior performance across three benchmark datasets.

Figure 3: Illustration of (a) Joint optimization and (b) Decoupled optimization for NDC. Existing methods require the backbone to accommodate additional optimization objectives beyond nuclei representation, leading to representation degradation. In contrast, DeNuC employs an independent lightweight detection network

\mathcal{D}

for nuclei localization, thereby allowing the backbone to focus exclusively on representation learning.

Method	Training	BRCAM2C				OCELOT			PUMA
Method	Params.	$F^{Lym.}$	$F^{Tum.}$	$F^{Oth.}$	$F^{Avg.}$	$F^{Tum.}$	$F^{Oth.}$	$F^{Avg.}$	$F^{Lym.}$	$F^{Tum.}$	$F^{Oth.}$	$F^{Avg.}$
CGT [ 9 ]	37M	56.42	75.98	50.44	60.95	68.77	61.30	65.03	76.55	79.66	54.20	70.14
SENC [ 10 ]	27M	57.94	76.50	49.42	61.29	70.02	62.08	66.05	77.38	81.51	54.38	71.09
CellViT [ 6 ]	143M	67.20	78.20	51.81	65.73	67.36	60.22	63.79	79.07	81.16	57.96	72.73
Hover-Net [ 4 ]	34M	62.31	75.25	49.58	62.38	69.27	61.03	65.15	75.72	78.36	49.53	67.87
DPA-P2PNet [ 17 ]	32M	59.65	77.26	55.26	64.06	70.07	59.92	64.99	76.80	81.87	54.04	70.90

Table 1: Comparison of nucleus detection and classification in F1-score % (

\uparrow

F^{Tum.}

F^{Lym.}

F^{Oth.}

, and

F^{Avg.}

denote the F1 score of tumor nucleus, lymphocytes, other nucleus, and average, respectively. The best results are highlighted in bold , and the second-best results are in underlined . DeNuC significantly outperforms other methods.

Backbone	Params.	BRCAM2C		OCELOT		PUMA
Backbone	Params.	$F^{Det.}$	$F^{Avg.}$	$F^{Det.}$	$F^{Avg.}$	$F^{Det.}$	$F^{Avg.}$
SN (0.5 $\times$ )	0.3M	86.61	71.43	79.89	68.85	92.23	75.45
SN (1.0 $\times$ )	1.0M	86.98	71.58	80.97	69.74	93.13	75.98
SN (1.5 $\times$ )	2.7M	87.00	71.90	81.07	69.76	93.28	76.19
SN (2.0 $\times$ )	4.3M	87.52	71.97	81.33	69.94	93.57	76.36
ResNet-50	26M	87.29	71.90	81.48	70.03	93.22	76.07

Table 2: Ablation study of

\mathcal{D}

in F1-score % (

\uparrow

). (Left) Comparison of different backbones. (Right) Ablation of cross-dataset training strategy.

N_{nu}

denotes the number of nuclei in the training set.

F^{Det.}

denotes the F1 score of detection. "SN" denotes ShuffleNetV2. "Separated" and "Joint" indicate that models are trained individually on each dataset and jointly across all datasets, respectively.

Method	$P_{det}$	Training	BRCAM2C				OCELOT			PUMA
Method	$P_{det}$	Params.	$F^{Lym.}$	$F^{Tum.}$	$F^{Oth.}$	$F^{Avg.}$	$F^{Tum.}$	$F^{Oth.}$	$F^{Avg.}$	$F^{Lym.}$	$F^{Tum.}$	$F^{Oth.}$	$F^{Avg.}$
Linear	✓	4.6K	69.73	85.10	61.08	71.97	73.83	66.04	69.94	81.00	85.25	62.85	76.37
Lora	✓	2.4M	71.79	85.33	61.40	72.84	74.19	66.93	70.56	81.23	85.70	63.38	76.77
Fully	✓	681M	71.03	85.50	61.11	72.55	73.87	65.52	69.69	81.45	85.37	63.13	76.65
End-to-end	✗	686M	68.21	79.42	56.67	68.10	67.37	58.15	62.76	79.50	83.19	62.20	74.97

Table 3: Ablation study on the optimization strategy of the classification network in F1-score % (

\uparrow

). Providing nuclei coordinates highly improves performance.

Explore similar work

Mar 7, 2026

NuNext: Reframing Nucleus Detection as Next-Point Detection

Nucleus detection in histopathology is pivotal for a wide range of clinical applications. Existing approaches either regress nuclear proxy maps that require complex post-processing, or employ dense anchors or queries that introduce severe foreground-background imbalance. In this work, we reformulate nucleus detection as next-point prediction, wherein a multimodal large language model is developed to directly output foreground nucleus centroids from the input image. The model is trained in two stages. In the supervised learning stage, we propose spatial-aware soft supervision to relax strict centroid matching and a chain-of-visual-thought strategy to incorporate visual priors that facilitate coordinate prediction. In the reinforcement fine-tuning stage, we design distribution matching reward, low-variance group filtering, and fine-grained advantage shaping to further improve the model's detection quality. Extensive experiments on nine widely used benchmarks demonstrate the superiority of our method. Code will be released soon.

Zhongyi Shui, Honglin Li, Xiaozhong Ji +7

Zhejiang University · Westlake University · Shanghai Artificial Intelligence Laboratory +3

Computational Pathology Object Detection RL Fine-Tuning

Ask PDF

May 31, 2026

AMN: An Adaptive Multi-Scale Fusion Network with Boundary and Uncertainty Modeling for Nuclei Segmentation

Accurate classification of nuclei subtypes in histopathology images is critical for downstream tasks including tumor grading, immune infiltrate quantification, and prognosis prediction. Existing approaches rely on either convolutional or transformer-based encoders in isolation, limiting their ability to simultaneously capture fine-grained local texture and long-range spatial context. We present AMN (Adaptive Multi-Scale Nuclei Network), a dual-encoder segmentation framework that jointly leverages a Swin Transformer and a ResNet-50 feature pyramid, fused via a learned per-channel gating mechanism that dynamically weighs each encoder's contribution at every scale. AMN is trained with a multi-objective loss combining class-weighted focal loss, boundary-aware loss with positive-pixel emphasis, and a novel uncertainty-modulated classification term that suppresses overconfident erroneous predictions. Evaluated on the CoNIC benchmark across seven nuclei classes, AMN achieves a mean Dice of 0.82 and mean F1 of 0.68, with an F1 of 0.67 on the diagnostically challenging lymphocyte class. AMN outperforms eight baseline models spanning pure-CNN, pure-transformer, and recent hybrid architectures: U-Net, ResU-Net, DeepLabV3+, SegNet, ViT-Small, HmsU-Net, ConvFormer-UNet, and BEFUnet. Cross-dataset evaluation on MoNuSeg demonstrates strong generalization without retraining and validating the domain robustness of the learned representations.

Spoorthi M, Suja Palaniswamy

Department of Computer Science & Engineering, Amrita School of Computing, Bengaluru, Amrita Vishwa Vidyapeetham, India

Histopathology Image Segmentation Hybrid CNN-Transformer Architectures Multi-Scale Feature Fusion Medical Image Segmentation

Ask PDF

Jul 30, 2026

Beyond Classification: Pathology Foundation Models as Detection Encoders for Mitotic Figures

Pathology foundation models (FMs) are models trained on vast amounts of typically unlabeled data and have been shown to yield regularized latent spaces that can be used effectively in downstream classification tasks. This is also true for the classification of mitotic figures vs. other cells. However, it is so far unclear if the latent space of current FMs provides features that are discriminant and spatially suitably resolved to also serve as a backbone for dense object detection paradigms. In this work, we investigate this question for common current pathology FMs (UNI, UNI2-h, Virchow, Virchow2, H-optimus-0, H-optimus-1) and compare their performance against a fully end-to-end trained baseline based on a ResNet50 architecture. We combine FM backbones with representatives of single stage, dual stage and self-attention-based detectors (RetinaNet, Faster R-CNN, Deformable DETR respectively) on the multi-domain MIDOG++ dataset, and on the TUPAC16 dataset as an out-of-domain case. We show that the H-optimus-0 and Virchow models yielded competitive performance, indicating that the latent spaces of current FMs, all trained on image-level self-supervision, are suitable for direct mitotic figure detection and may be slightly more robust on our out-of-domain test case. All code is made available publicly at https://github.com/DeepMicroscopy/FM4MFdet.

Sweta Banerjee, Alireza Teimoury, Nils Porsche +11

Flensburg University of Applied Sciences, Flensburg, Germany · Julius-Maximilians-Universität Würzburg, Würzburg, Germany · Friedrich-Alexander-Universität Erlangen-Nürnberg, Germany +4

Computational Pathology Pathology Foundation Models Object Detection Whole-Slide Image Analysis

Ask PDF