The MYOSAIQ Challenge: Myocardial Segmentation with Automated Infarct Quantification
研究通过MYOSAIQ挑战赛,利用UNet等深度学习方法解决心肌梗死量化问题,使用439个CMR数据集,提高了左心室和心肌分割的准确性。
研究通过MYOSAIQ挑战赛,利用UNet等深度学习方法解决心肌梗死量化问题,使用439个CMR数据集,提高了左心室和心肌分割的准确性。
为解决视频异常检测中噪声注入导致的物理姿态不真实问题,提出STEP框架,利用PCA投影姿态序列,并结合置信度分数加权机制,提高长视频序列处理性能。
This work addresses the challenges of modeling complex temporal dynamics and patient heterogeneity in longitudinal clinical data by proposing a temporal graph-based contrastive graph neural network approach. The method constructs multivariate disease trajectories as temporal graphs and incorporates structure-aware random walks to guide contrastive learning, effectively preserving both temporal context and trajectory topology. By uniquely integrating structure-aware random walks with contrastive learning, the framework learns expressive embeddings of patient observation nodes, significantly enhancing robust clustering of patients with similar disease progression patterns and uncovering latent evolutionary structures inherent in longitudinal clinical data.
Standard CNNs for whole-heart multi-chamber CT segmentation often lack explicit anatomical constraints, compromising clinical reliability. This work proposes a lightweight approach that explicitly incorporates statistical shape priors through a shape-aware loss and a 3D U-Net variant guided by spatial label distribution heatmaps. The method is systematically evaluated on the MM-WHS CT and WHS++ datasets. Results reveal that, despite modern architectures implicitly learning substantial anatomical regularities from data, explicitly integrating handcrafted shape priors yields only marginal and inconsistent performance gains—and frequently leads to degradation. These findings underscore the limited added value of manually designed anatomical priors when deployed within highly data-driven deep learning models.
This work proposes TopKGraphs, a non-parametric and interpretable method for robust node similarity estimation in sparse, noisy, or heterogeneous networks. By anchoring random walks at source nodes and guiding transitions via Jaccard similarity, the approach treats walks as local neighborhood samplers. It generates partial node rankings and constructs an affinity matrix through robust rank aggregation, effectively integrating local structural cues with global walk-based information. Notably, TopKGraphs operates without relying on stationary distributions or learned embeddings. Extensive evaluations demonstrate its superior performance over established baselines—including Jaccard, Dice, personalized PageRank, and Node2Vec—across synthetic graphs, k-nearest neighbor graphs, and protein–protein interaction networks, highlighting its accuracy and robustness under diverse and challenging conditions.
研究通过MYOSAIQ挑战赛,利用UNet等深度学习方法解决心肌梗死量化问题,使用439个CMR数据集,提高了左心室和心肌分割的准确性。
为解决视频异常检测中噪声注入导致的物理姿态不真实问题,提出STEP框架,利用PCA投影姿态序列,并结合置信度分数加权机制,提高长视频序列处理性能。
This work addresses the challenges of modeling complex temporal dynamics and patient heterogeneity in longitudinal clinical data by proposing a temporal graph-based contrastive graph neural network approach. The method constructs multivariate disease trajectories as temporal graphs and incorporates structure-aware random walks to guide contrastive learning, effectively preserving both temporal context and trajectory topology. By uniquely integrating structure-aware random walks with contrastive learning, the framework learns expressive embeddings of patient observation nodes, significantly enhancing robust clustering of patients with similar disease progression patterns and uncovering latent evolutionary structures inherent in longitudinal clinical data.
Standard CNNs for whole-heart multi-chamber CT segmentation often lack explicit anatomical constraints, compromising clinical reliability. This work proposes a lightweight approach that explicitly incorporates statistical shape priors through a shape-aware loss and a 3D U-Net variant guided by spatial label distribution heatmaps. The method is systematically evaluated on the MM-WHS CT and WHS++ datasets. Results reveal that, despite modern architectures implicitly learning substantial anatomical regularities from data, explicitly integrating handcrafted shape priors yields only marginal and inconsistent performance gains—and frequently leads to degradation. These findings underscore the limited added value of manually designed anatomical priors when deployed within highly data-driven deep learning models.
This work proposes TopKGraphs, a non-parametric and interpretable method for robust node similarity estimation in sparse, noisy, or heterogeneous networks. By anchoring random walks at source nodes and guiding transitions via Jaccard similarity, the approach treats walks as local neighborhood samplers. It generates partial node rankings and constructs an affinity matrix through robust rank aggregation, effectively integrating local structural cues with global walk-based information. Notably, TopKGraphs operates without relying on stationary distributions or learned embeddings. Extensive evaluations demonstrate its superior performance over established baselines—including Jaccard, Dice, personalized PageRank, and Node2Vec—across synthetic graphs, k-nearest neighbor graphs, and protein–protein interaction networks, highlighting its accuracy and robustness under diverse and challenging conditions.