acceptodds
Withdrawn from ICLR 2027

HyperGen: Controllable Pathological Anomaly Synthesis via Hyperbolic Semantic Manifold Learning

Withdrawn from ICLR 2027, so left out of the paper lists and the map.

Abstract

Medical anomaly synthesis has emerged as a promising strategy for alleviating the scarcity of annotated pathological data in clinical image analysis. Recent diffusion-based methods can generate anatomically realistic lesions, yet their latent representations remain largely unstructured, making it difficult to systematically control pathological characteristics and generate diverse abnormalities. To address this challenge, we propose **HyperGen**, a novel framework that reformulates anomaly synthesis as geometry-guided semantic navigation on a learned hyperbolic manifold. HyperGen first learns a lesion-centric hyperbolic embedding, where the radial coordinate is aligned with Hounsfield Unit (HU) density while the angular coordinates capture complementary morphological and textural variations. By manipulating the manifold in polar coordinates and projecting the navigated embedding back to the Euclidean space, HyperGen provides semantic conditioning for a diffusion model to generate diverse and anatomically consistent pathological images. Extensive experiments on lung CT demonstrate that HyperGen produces more realistic and structurally coherent lesions than strong diffusion-based baselines. Furthermore, the synthesized anomalies consistently improve downstream lesion segmentation, validating the effectiveness of geometry-guided semantic manifold learning for controllable medical anomaly generation.

Primary area:
learning on graphs and other geometries & topologies
Keywords:
Medical Anomaly Synthesis,Controllable Image Generation,Hyperbolic Representation Learning

Reviews

Among:
Shape:
1234this paper, Bottom 2%
Figure 1. Mean review rating of the reviewed papers at ICLR 2027. This paper averages 1.33 over 3 reviews, above 2% of the others.
Reviewer_QaDbOpenReview ↗
Rating
1
Confidence
3
Informativeness
2
Presentation
2
Soundness
1
Read review
Paper summary

This paper proposes HyperGen, a framework for controllable pathological anomaly synthesis based on hyperbolic semantic manifold learning. The method first extracts lesion-specific representations through mask-guided feature aggregation and maps them into a Poincaré ball, where the radial coordinate is explicitly aligned with lesion attenuation (HU values), while angular directions are intended to capture morphological and textural variations. The learned representations are then mapped back to the tangent space and injected into a 3D diffusion model through FiLM conditioning, combined with step-wise background stitching to preserve surrounding anatomy. By manipulating the radial and angular coordinates, the authors aim to achieve interpretable control over synthesized lesions.

The method is evaluated on LIDC-IDRI and EMIDEC. The authors report improvements in generation quality measured by FID, SSIM, and PSNR, as well as enhanced downstream nnU-Net segmentation performance using synthetic data augmentation. Qualitative visualizations are further provided to illustrate semantic controllability and pathological condition preservation.

Critical strengths

1. Clinically meaningful semantic conditioning.

The paper addresses an important problem in controllable medical image synthesis by incorporating clinically interpretable attributes into the generation process. In particular, aligning the latent radial coordinate with lesion attenuation (HU values) provides an intuitive way to control pathological appearance. This is a potentially useful direction for generating diverse lesions with meaningful radiological characteristics, although the necessity of hyperbolic geometry remains unclear.

2. Evaluation of downstream clinical utility.

Beyond conventional image quality metrics, the paper evaluates the usefulness of synthesized lesions for downstream segmentation. The reported improvements in nnU-Net performance (e.g., Dice from 78.26% to 83.02% in the normal-case augmentation setting) suggest that the generated data may provide practical benefits for medical image analysis. This is a positive aspect of the evaluation, although stronger baseline comparisons and controlled ablations are still needed to establish the source of these gains.

Critical weaknesses

1. The necessity and contribution of hyperbolic geometry are not convincingly established.

The central methodological claim is insufficiently justified. According to Eqs. (4) and (7), the hyperbolic embedding is mapped back to the tangent space through the inverse logarithmic map, yielding the original Euclidean feature vector in the absence of navigation. Moreover, the proposed radial navigation (Eqs. (10)–(11)) is mathematically equivalent to scaling the Euclidean feature vector by a scalar factor. The HU-based radial alignment can also be formulated as a constraint on the Euclidean feature norm. Although hyperbolic geometry may influence the representation learning stage, the paper does not demonstrate any advantage over an equivalent Euclidean formulation. Without a carefully matched Euclidean baseline, it is difficult to determine whether the reported improvements arise from hyperbolic geometry or simply from HU supervision and feature scaling. This substantially weakens the paper's central novelty claim.

2. Insufficient experimental evidence to support the claimed controllability and performance advantages.

The experimental evaluation lacks several critical controls. First, no systematic ablation isolates the contributions of hyperbolic geometry, HU-based supervision, and manifold navigation. Second, the claimed disentanglement between lesion density and morphology is primarily supported by qualitative visualizations, without quantitative measurements of control accuracy, monotonicity, or attribute independence. Third, the improvements over strong lesion-focused diffusion baselines are relatively modest in some settings (e.g., 83.02 vs. 82.66 Dice), yet no variance estimates or statistical significance tests are reported. The baseline comparisons and cross-modality evaluation protocols also require further clarification. Consequently, the current experiments do not provide sufficient evidence to support the claimed geometric advantages or the generality of the proposed approach.

Questions for authors

Q1. Necessity of Hyperbolic Geometry (Clarification and Essential Validation)

According to Eqs. (4) and (7), the composition of the exponential and logarithmic maps recovers the original Euclidean tangent vector. Furthermore, the proposed radial navigation appears mathematically equivalent to scaling the Euclidean feature vector by a factor of α\alpha. Could the authors clarify which component specifically benefits from non-Euclidean geometry? More importantly, can the authors provide a controlled comparison against an equivalent Euclidean formulation with the same HU supervision, architecture, and radial scaling? This is essential for assessing the paper's central methodological contribution.

Q2. Baseline Comparisons and Experimental Provenance (Clarification)

Several baseline results in Tables 2 and 3 appear identical to those previously reported in LeFusion. Were these results independently reproduced or directly adopted from the original publication? If adopted, could the authors clarify the sources and explain the differences in baseline citation attribution? Additionally, why are stronger published LeFusion configurations, particularly LeFusion-J on EMIDEC, not included in the comparisons? A clarification of the experimental protocols and comparison fairness would materially affect my assessment of the claimed performance improvements.

Q3. Quantitative Validation of Semantic Controllability (Essential Validation)

The paper claims that radial navigation controls lesion attenuation while angular navigation controls morphology and texture. However, the evidence is primarily qualitative. Can the authors provide quantitative evidence showing (i) the relationship between radial scaling and the actual HU values of generated lesions, and (ii) whether angular manipulation changes morphological characteristics while preserving lesion attenuation? Without such evidence, it is difficult to assess whether the proposed representation genuinely supports disentangled semantic control.

Q4. Cross-Modality Evaluation on EMIDEC (Clarification)

The proposed radial alignment relies on CT Hounsfield Units, whereas EMIDEC contains cardiac MRI data. Could the authors clarify how the HU-based supervision is adapted to MRI, and whether the model is retrained on EMIDEC or directly transferred from LIDC-IDRI? This distinction is important for evaluating the claimed generalizability of the learned representation. Additionally, could the authors clarify how the average metrics in Table 3 were calculated, as some reported averages do not match the arithmetic means of the corresponding subsets?

Reviewer_Dgc4OpenReview ↗
Rating
2
Confidence
3
Informativeness
2
Presentation
2
Soundness
3
Read review
Paper summary

This paper proposes HyperGen, a framework for controllable pathological anomaly synthesis using hyperbolic semantic manifold learning. It extracts lesion-specific features through mask-guided aggregation and maps them into a Poincaré ball, where radial coordinates represent lesion density and angular directions capture morphological variations. The learned representations are used to condition a 3D diffusion model through FiLM, enabling controllable lesion generation while preserving surrounding anatomy. Experiments on LIDC-IDRI demonstrate improvements in FID, SSIM, and PSNR over existing diffusion-based methods. Downstream nnU-Net segmentation results further show the utility of synthetic lesions, while additional experiments on EMIDEC evaluate generalization and semantic controllability.

Critical strengths

The paper introduces a structured latent representation that explicitly associates lesion density with radial coordinates, providing an interpretable mechanism for manipulating pathological characteristics. This is valuable for generating diverse lesions with clinically relevant variations.

HyperGen achieves improvements in FID, SSIM, and PSNR over diffusion-based baselines on LIDC-IDRI. More importantly, synthetic data augmentation improves downstream nnU-Net segmentation performance, demonstrating practical value beyond visual fidelity. Additional experiments on EMIDEC further broaden the empirical evaluation.

Critical weaknesses

The exponential and logarithmic mappings effectively recover the original Euclidean features, while radial navigation is mathematically equivalent to feature scaling. Without a controlled Euclidean baseline, the necessity and actual contribution of hyperbolic geometry remain unclear.

The claimed disentanglement between lesion density and morphology lacks explicit supervision and quantitative evaluation. ablation studies are also missing.

Questions for authors

Given that the exponential and logarithmic maps are inverses and radial navigation effectively reduces to Euclidean feature scaling, what specific advantages does hyperbolic geometry provide? Could the authors present a controlled Euclidean baseline with identical architecture and supervision to isolate its contribution?

How are angular directions manipulated to control morphology independently of lesion density? Could the authors provide quantitative evidence of this disentanglement? Additionally, are the relatively small downstream improvements over the strongest baseline statistically significant across multiple runs?

Reviewer_2tnMOpenReview ↗
Rating
1
Confidence
4
Informativeness
2
Presentation
2
Soundness
1
Read review
Paper summary

The paper proposes HyperGen, a framework for generating medical anomalies using geometry-guided semantic navigation in a learned hyperbolic space. It learns geometric representations of lesions, capturing HU density, morphology, and texture, which are then used to guide a diffusion model in generating realistic pathological images. Experiments on lung CT show that HyperGen generates more realistic and structurally consistent lesions than existing diffusion-based methods, while improving downstream lesion segmentation performance.

Critical strengths
  • The paper introduces geometry-guided medical image synthesis, incorporating meaningful information such as lesion density and morphology to guide image generation.
  • The experiments demonstrate improved performance compared to existing diffusion-based baselines.
Critical weaknesses
  • While the main idea is easy to understand, the paper is difficult to follow due to unclear writing and several presentation issues. For example, the text in Fig. 2 is too small to read at normal size, and there are formatting problems around lines 80–81.
  • The paper introduces several modules, but their design choices and motivations are not sufficiently justified. For example, it is unclear how the lesion-focused aggregation features relate to the hyperbolic manifold embedding.
  • The descriptions in Fig. 2 and the main text are not sufficiently clear to understand how the proposed modules are implemented and connected.
  • The experiments mainly compare against diffusion-based baselines, with limited evidence that the generated images are useful for real-world medical applications.
Questions for authors

My main concerns are: 1. unclear writing and figures in the paper; 2. The "geometry-guidance" appears very weak, 3. The lack of real-world validation on the generated results.

From OpenReview, CC BY 4.0. Authors as signed there; shown unedited.

Related papers

Discussion (0)

Sign in to comment.