Pos4CHD: A Positive-Free Mask-Guided Counterfactual Synthesis Framework for Fetal Congenital Heart Defect Analysis
Fetal congenital heart defect diagnosis is severely bottlenecked by the scarcity of positive cases, especially for rare subtypes that can occur in fewer than 1 in 10,000 births. This paper proposes Pos4CHD, a framework that synthesizes pseudo-positive echocardiography samples using only normal images and semantic masks, completely bypassing the need for real pathological training data. By combining clinically guided mask manipulation with a specialized generator called FE-Gen, the authors demonstrate that synthetically augmented models can approach or exceed real-data baselines on classification, detection, and segmentation tasks.
Pos4CHD addresses a genuinely hard clinical problem-positive-free fetal CHD synthesis-and delivers a well-engineered pipeline comprising rule-based mask manipulation and the custom FE-Gen generator. Quantitative results on in-house datasets show strong generation fidelity and downstream gains, including zero-shot classification ($93.25\%$ ACC on EbA) and segmentation ($+10.49\%$ DSC). Nevertheless, the approach hinges on manually designed transformation rules per subtype and very small real-positive validation sets, which temper the breadth of the claimed generalizability across the more than 27 CHD subtypes.
The clinical motivation is compelling and the technical contributions are well-targeted. FE-Gen introduces two specific, principled improvements for ultrasound adaptation—histogram regularization and learnable semantic queries—which tangibly improve over naive ControlNet adaptation. The downstream evaluation is unusually broad for a synthesis paper, spanning zero-shot rare-CHD classification, lesion detection, and anatomy segmentation, and the authors transparently report failure modes such as subtle VSD structural misses.
Generalization claims are constrained by a reliance on expert-crafted, subtype-specific mask transformation rules ($R(\cdot, c, \psi_c)$) whose hyperparameters are not fully specified, and by very small validation cohorts: the zero-shot EbA and TAPVC classifiers are tested on only 42 and 40 positive cases respectively, and the VSD detection training set contains merely 28 real positives. Additionally, the in-house datasets are not publicly available, making independent reproduction difficult.
The user study-while showing a $46.4\%$ real-vs-synthetic discrimination rate-used only 20 images per category, which is underpowered for strong perceptual fidelity claims, and the failure cases in Fig. 14 confirm that subtle pathological structures are not yet reliably synthesized.
The evidence supports FE-Gen's superiority over generic text-to-image baselines such as DreamBooth, SD-LoRA, T2I-Adapter, and standard ControlNet in fine-grained mask alignment and histogram fidelity. However, the paper fails to quantitatively compare against the most closely related fetal ultrasound synthesis methods it discusses-namely FetalFlex and FVEDM-in either generation quality or downstream task performance, weakening the claim of SOTA applicability. Moreover, the downstream baselines are trained without synthesized data rather than competing counterfactual synthesis pipelines, so the gains reflect data-augmentation utility rather than methodological dominance over alternative CHD synthesis frameworks.
Although the authors state that code and models are publicly available, the core datasets-WHU-4CH-8k, WHU-4CH-Seg517, and the CHD evaluation sets-are in-house and not publicly released, which blocks independent reproduction of the training and evaluation pipeline. The paper does report training hyperparameters such as learning rates, LoRA rank, and regularization weight ($\omega_h=0.1$), but the clinical mask-manipulation rules and filtering thresholds are described only at a high level, leaving critical implementation details underspecified.
The scarcity of positive cases remains a major bottleneck in developing automated diagnostic systems for fetal Congenital Heart Defects~(CHDs), particularly for rare subtypes. Although normal cases are accessible with effort, positive cases are hard to collect due to low incidence rates and privacy concerns, causing severe class imbalance or absence of positive training samples. Existing works resort to synthesizing positives with conditional generative models, but they are task-specific and require positive training samples, hindering application for fetal CHDs. To this end, we propose a positive-free mask-guided counterfactual generation framework for fetal echocardiography, namely Pos4CHD, which enables realistic, high-quality synthesis of pseudo-positive samples from semantic masks, with only normal samples at hand. Technically, we first develop FE-Gen, a mask-guided generative model tailored for fetal heart ultrasound synthesis that mitigates histogram drift and enhances image-mask spatial alignment. Building on FE-Gen, Pos4CHD employs a mask-to-image generation paradigm for CHD synthesis. It first manipulates semantic masks to simulate fetal-heart structural anomalies and then renders anatomically consistent images. Extensive experiments on in-house datasets, together with expert evaluations, demonstrate the visual fidelity~($-12.97$ FID) and anatomical consistency~($+3.96\%$ DSC) of synthesized images. Further validations across diverse real-world CHD diagnosis tasks exhibit promising results and consistent increments, including zero-shot classification~(achieving $93.25\%$ ACC on EbA and $72.92\%$ ACC on TAPVC), lesion detection~($+6.95\%$ mAP50 on VSD), and anatomy segmentation~($+10.49\%$ DSC on fetal heart) compared with baselines. Codes and models are publicly available on \href{https://github.com/DREAMXFAR/Pos4CHD}{GitHub}.
Pick a starting point or write your own. Challenges run in the background, so you can keep reading while the AI investigates.
No challenges yet. Disagree with the review? Ask the AI to revisit a specific claim.