Generalized Biomedicine Discovery

Luyao Tang; Yingkai Yang; Hanqi Chen; Jiewei Zheng; Chaoqi Chen; Cheng Chen

cs.LGSep 9, 2026

Generalized Biomedicine Discovery

Authors: Luyao Tang, Yingkai Yang, Hanqi Chen, Jiewei Zheng, Chaoqi Chen, Cheng Chen

Organizations: The University of Hong Kong, Hong Kong SAR, China · Shenzhen University, Shenzhen, China · Xiamen University, Xiamen, China

Abstract

In real-world clinical practice, medical images face open-world shifts: (i) long-tailed rare diseases, (ii) subtle lesions dominated by normal anatomy, and (iii) hierarchical taxonomies. Yet most open-world paradigms assume flat, balanced label spaces, leaving these biomedical demands unresolved. We introduce Generalized Biomedicine Discovery (GBD) and a unified benchmark spanning long-tail, anomaly, and taxonomy-aware discovery. Our key insight is that dominant known patterns form a visual manifold that masks subtle novelty. Inspired by expert diagnosis, we propose SCAN (Surprise-evoked Complementary AccommodatioN), which follows a cognition-inspired perceptual progression: it applies predictive suppression to filter expected norms, triggers surprise-evoked salience to highlight unexpected deviations, and performs complementary accommodation to integrate these shifts into global representations. Extensive experiments show that SCAN improves novel concept discovery while generally preserving established clinical knowledge, and it plugs into existing architectures to better navigate the known-unknown trade-off in medical imaging. Code is available at https://github.com/lytang63/generalized-biomedicine-discovery.

Figures & tables

Figure 1: GBD in practice. Real biomedical data shift beyond the closed world: standard classification fails on unseen diseases, OSR merges heterogeneous novelties into one “unknown”, and conventional discovery often assumes unlabeled data contains only new categories. GBD formalizes three realistic discovery regimes: (I) long-tail rare disease discovery among common cases, (II) normal-to-abnormal discovery of subtle lesions dominated by normal anatomy, and (III) within-taxonomy discovery that uncovers new sibling subtypes to refine diagnostic hierarchies.

Figure 2: Method overview of SCAN (Surprise-evoked Complementary AccommodatioN) for GBD. SCAN refines representations in three steps: (I) it learns a notion of dominant evidence and suppresses predictable visual patterns to expose residual signals; (II) it scores the residuals by “surprise” to highlight sparse, high-energy deviations and aggregates them into salient evidence; (III) it converts the salient evidence into a complementary update and injects it through a gated fusion, limiting disruption to known concepts while strengthening novel structure.

Method	Gastrovision			HistoSet			MLL23			Derm12345			Average
Method	All	Old	New	All	Old	New	All	Old	New	All	Old	New	All	Old	New
SEAL †	56.8	51.2	60.7	81.2	87.1	77.9	65.4	67.6	54.1	56.5	61.0	10.6	65.0	66.7	50.8
SelEx	50.1	52.0	47.4	76.4	97.5	70.5	66.4	68.3	61.3	28.9	29.1	27.9	55.5	61.7	51.8
+ SCAN	71.2	83.4	53.0	76.3	99.9	69.7	80.7	89.3	58.6	35.6	35.6	35.6	66.0	77.1	54.2
$\Delta$	21.1	31.4	5.6	-0.1	2.4	-0.8	14.3	20.9	-2.7	6.7	6.5	7.7	10.5	15.3	2.4
SimGCD	37.8	40.6	31.8	74.4	83.8	71.8	56.9	52.9	67.0	25.4	27.9	13.1	48.6	51.3	45.9

Table 1: Comparison on Setting I (Long-tail Rare Disease Discovery) across endoscopy , pathology , microscopy , and dermatology images. Results are accuracy on All, Old, and New subsets, with the best results highlighted in bold.

Method	Gastrovision			HistoSet			MLL23			Derm12345			Average
Method	All	Old	New	All	Old	New	All	Old	New	All	Old	New	All	Old	New
SEAL †	62.6	64.2	61.0	73.5	98.7	48.2	69.2	70.9	60.1	59.0	61.7	5.5	66.1	73.8	43.7
SelEx	50.3	51.9	41.7	64.3	97.7	47.5	58.5	58.0	59.8	29.3	28.2	40.7	50.6	58.9	47.4
+ SCAN	70.1	86.0	46.3	68.6	97.2	60.7	87.1	89.1	81.7	33.6	33.3	36.4	64.8	76.4	56.3
$\Delta$	19.8	34.1	4.6	4.4	-0.5	13.1	28.6	31.1	21.9	4.3	5.1	-4.3	14.2	17.5	8.8
SimGCD	39.7	42.9	25.6	61.1	94.1	44.5	57.9	55.9	63.4	26.6	27.2	19.9	46.3	55.0	38.3

Table 2: Comparison on Setting II (Normal-to-Abnormal Discovery) across endoscopy , pathology , microscopy , and dermatology images. Results are accuracy on All, Old, and New subsets, with the best results highlighted in bold.

Method	Gastrovision			HistoSet			MLL23			Derm12345			Average
Method	All	Old	New	All	Old	New	All	Old	New	All	Old	New	All	Old	New
SEAL †	54.9	51.3	64.3	92.9	95.7	89.3	70.6	83.3	47.2	27.7	62.4	16.4	61.5	73.2	54.3
SelEx	50.1	45.7	53.8	77.5	99.4	71.4	62.5	69.9	55.6	24.8	65.1	18.2	53.7	70.0	49.8
+ SCAN	66.2	79.6	46.3	88.5	96.4	83.2	81.4	89.0	61.9	27.8	69.9	20.9	66.0	83.7	53.1
$\Delta$	16.1	33.8	-7.5	11.0	-3.0	11.7	18.9	19.0	6.3	3.0	4.8	2.7	12.3	13.7	3.3
SimGCD	42.9	43.0	42.5	89.4	91.7	87.9	53.2	57.4	49.4	19.3	40.8	15.9	51.2	58.2	48.9

Table 3: Comparison on Setting III (Within-Taxonomy Discovery) across endoscopy , pathology , microscopy , and dermatology images. Results are accuracy on All, Old, and New subsets, with the best results highlighted in bold.

Panel of Figure 3 — Figure 3: Training dynamics on Gastrovision (Setting I): Known/Novel accuracy and the evolution of the surprise signal ( $S$ ) and knownness ( $k$ ) for SelEx and SelEx+SCAN.

Panel of Figure 4 — Figure 4: Mechanism, qualitative, and efficiency analyses of SCAN.

Generalized Category Discovery

Generalized category discovery learns known classes from labeled examples while grouping unlabeled data into known and previously unseen categories.

See all 26 papers

MedXplore: Towards Reliable and Unbiased Generalized Category Discovery in Medical Imaging
Jul 30, 2026Jianwei He, Kailin Lyu, Junhao Dong +6Medical Image Representation Learning Generalized Category Discovery
Institute of Automation, Chinese Academy of Sciences Beijing, China · Nanyang Technological University Singapore, Singapore
Transcending the Annotation Bottleneck: AI-Powered Discovery in Biology and Medicine
Feb 23, 2026Soumick ChatterjeeUnsupervised Learning Self-Supervised Learning
Human Technopole, Milan, Italy · Faculty of Computer Science, Otto von Guericke University Magdeburg, Magdeburg, Germany
MMBU: A Massive Multi-modal Biomedical Understanding Benchmark to Probe the Perception Capabilities of Vision-Language Models
Jun 4, 2026Ryan D'Cunha, Alejandro Lozano, Xiaoxiao Sun +17Medical Image Analysis Medical VLMs
Stanford University, Stanford, CA, USA · Instituto Tecnológico de Monterrey, Mexico · Monash University, Melbourne, VIC, Australia +4

Jul 30, 2026J/K move · Enter open · S save

MedXplore: Towards Reliable and Unbiased Generalized Category Discovery in Medical Imaging

Jianwei He, Kailin Lyu, Junhao Dong, Long Xiao+5

Institute of Automation, Chinese Academy of Sciences Beijing, China · Nanyang Technological University Singapore, Singapore

Medical Image Representation Learning Generalized Category Discovery

Open paper AskPDF arXiv

Deep learning has shown strong potential in medical image analysis, but most existing methods rely on large-scale annotations and a closed-world assumption that rarely holds in clinical practice. Although Generalized Category Discovery (GCD) has advanced rapidly on natural images, it remains underexplored in medical imaging. To address this issue, we propose MedXplore, a unified framework for reliable and unbiased medical GCD, optimizing from both perceptual and decision levels. Specifically, at the perceptual level, taking a frequency domain perspective, Frequency-SNR Adaptive Attention and Consistency (FAAC) performs learnable full-spectrum filtering and global-local energy contrast activation to not only highlight local abnormal signals relative to the global context, but also provide reliable semantic anchors for patch consistency learning. At the decision level, Adaptive Cosine-Angular Margin (ACAM) adjusts angular margins using semantic difficulty and feature confidence to balance intra-class compactness and inter-class separability. Together, the two modules improve lesion-sensitive representation learning and mitigate old-class bias. Experiments on multiple benchmarks show an average 8.5% gain in All accuracy over the strongest competing methods. On Kvasir, MedXplore reduces false-old errors from 14.50% to 0.80%, demonstrating strong robustness under severe old-new ambiguity.

Abstract

Generalized Category Discovery

Explore similar work

MedXplore: Towards Reliable and Unbiased Generalized Category Discovery in Medical Imaging

Transcending the Annotation Bottleneck: AI-Powered Discovery in Biology and Medicine

MMBU: A Massive Multi-modal Biomedical Understanding Benchmark to Probe the Perception Capabilities of Vision-Language Models