New Linear Baseline Excels in Cardiac CT Shape Completion.

Matej Gazda, Jakub Gazda, Juraj Gazda, Peter Drotar· August 21, 2026 View original

Key takeaways

  • A new open eleven-structure statistical shape model for whole-heart CT is released.
  • A simple linear conditional-Gaussian estimator significantly outperforms deep learning for shape completion.
  • The model provides a robust baseline for cardiac image analysis and cohort unification.
  • Accurate reconstruction of missing cardiac structures is crucial for comprehensive research.

Who benefits

HealthcareMedical DevicesPharmaceuticalsResearch & Development

Summary

This paper introduces a robust linear baseline for whole-heart cardiac shape completion on CT scans and releases an open eleven-structure statistical shape model. The closed-form conditional-Gaussian estimator significantly outperforms deep learning and nearest-neighbor methods in reconstructing missing cardiac structures.

Public cardiac imaging datasets often lack consistent annotations across different heart structures, making it difficult to combine them for comprehensive analysis. This research addresses this by releasing an open eleven-structure statistical shape model of the heart, built from 383 automatically labeled CT cases. Crucially, it also presents a strong linear baseline for cardiac shape completion. The study demonstrates that a simple closed-form conditional-Gaussian estimator achieved superior performance in reconstructing missing non-chamber structures, with a mean per-vertex error of 3.717 mm. This significantly outperformed a mask-conditioned graph variational autoencoder (5.248 mm) and nearest-neighbor retrieval (8.931 mm). The findings were consistent across internal and external validation sets, highlighting the effectiveness of this linear approach and providing a valuable new resource for cardiac imaging research.

Why it matters

Medical imaging professionals and AI developers can leverage this open model and robust baseline to improve the accuracy and consistency of cardiac analysis, potentially leading to better diagnostic tools and research capabilities.

How to implement this in your domain

  1. 1Integrate the released eleven-structure cardiac statistical shape model into existing medical image analysis pipelines.
  2. 2Benchmark current deep learning models for cardiac shape completion against this strong linear baseline.
  3. 3Utilize the conditional-Gaussian estimator for robust reconstruction of missing cardiac structures in CT data.
  4. 4Explore how this model can facilitate cohort unification research by providing shared correspondence across diverse datasets.

Original post by Matej Gazda, Jakub Gazda, Juraj Gazda, Peter Drotar

"arXiv:2608.19932v1 Announce Type: new Abstract: Public cardiac cohorts annotate different subsets of the heart, so shapes from separate sources cannot be pooled without shared correspondence. Among released cardiac shape resources, none we identified carries the atrial appendage,…"

View on X

Originally posted by Matej Gazda, Jakub Gazda, Juraj Gazda, Peter Drotar on X · view source

Want to go deeper?

Turn these trends into skills with Learnijoy's hands-on AI & tech courses.

Explore courses

More in AI Research

AI ResearchAI Engineering & DevTools

Decoding Silent Reading from Non-Invasive EEG

This research demonstrates that open-vocabulary word-level and semantic information can be reliably decoded from non-invasive EEG during silent reading. Using a contrastive decoder and a large dataset from a single participant, the study shows decoding scales log-linearly with training data and extends to rare words.

Ingo Marquardt, Anthilia Alchanat, Priyanka JainAug 21, 2026
AI ResearchAI Engineering & DevTools

Exact Learning Coefficients for Singular Models

This paper presents the first deterministic algorithm for exactly computing local learning coefficients (Real Log Canonical Thresholds) for two-dimensional singular models. This breakthrough provides ground truth for calibrating sampling-based estimators and reveals algebraic structure in learning coefficients, outperforming sampling in shallow regimes.

Gr\'egoire Sergeant-Perthuis (CQSB, Sorbonne Universit\'e), Elias Tsigaridas (Ouragan Team, INRIA), Jules Tsukahara (Ouragan Team, INRIA)Aug 21, 2026
AI Engineering & DevToolsAI Research

Standardized ML Evaluation for Power System Protection

This paper proposes a standardized framework for evaluating machine learning applications in power system protection, addressing inconsistencies in current research. It defines seven critical study dimensions and instantiates the framework with a case study on fault classification and localization using a public benchmark.

Julian Oelhaf, Georg Kordowich, Paula Andrea P\'erez-Toro, Christian Bergler, Johann J\"ager, Andreas Maier, Siming BayerAug 21, 2026