Min Xu

Medical anomaly detection in zero- and few-shot scenarios remains challenging due to significant domain shifts across imaging modalities and anatomical regions, where models trained on one setting often fail to generalize to unseen domains. While recent approaches leverage vision-language alignment for cross-modal adaptation, they primarily focus on semantic consistency and overlook the intrinsic stability of spatial organization in normal medical images. In this work, we introduce a Hypergraph-based Spatial Organization Module that represents image patches as nodes and captures group-wise structural dependencies through high-order hyperedges encoding spatial locality and structural similarity. To enforce structural regularity, we design a spatial consistency loss that promotes hyperedge-level coherence for normal samples while remaining tolerant to abnormal variations, and further incorporate cross-modal alignment to enhance generalization to unseen modalities. During inference, anomaly scores are derived from hypergraph deviation energy at both image and pixel levels and can be seamlessly integrated with zero-shot text-based comparison and few-shot memory mechanisms. Extensive experiments on multiple benchmarks demonstrate improved cross-domain performance in both classification and segmentation, highlighting spatial organization consistency as a powerful inductive bias for medical anomaly detection.

The automated discovery of structural patterns in macromolecular complexes remains a central challenge in cryo-electron tomography, particularly in highly heterogeneous datasets. Although fully unsupervised clustering methods have shown promise in grouping subtomograms by structural similarity, they often ignore a crucial source of information: the partial ground truth routinely available to structural biologists from prior studies or manual annotations. In this work, we propose a semi-supervised structural discovery framework that utilizes partial supervision to guide clustering without compromising the ability to uncover previously unknown structures. At the core of our method is a label-anchored probabilistic clustering mechanism that seeds the latent space using a small subset of labeled examples and refines it through a multi-resolution consensus strategy based on PCA-space voting. This is complemented by an entropy-based confidence scoring scheme that attenuates the influence of ambiguous samples, as well as a feature propagation procedure that extends structural labels to low-confidence regions using local similarity in feature space. Together, these components create a stable and adaptive pipeline capable of discovering both known and novel structures. Our approach is efficient, requires as little as 1% of labeled data per class, and consistently produces clearer, more interpretable feature embeddings compared to fully unsupervised methods, with well-separated clusters from the very first iterations. Extensive experiments on simulated and realistic tomographic datasets demonstrate that this semi-supervised strategy significantly improves clustering performance, robustness, and biological relevance in cryo-electron tomography analysis. These methods are integrated as extensions to the existing Deep Iterative Subtomogram Clustering Approach pipeline, enhancing its capability for guided structural discovery.

In this work, we propose a disentangled latent optimization-based method for parameterizing grouped deforming 3D objects into shape and deformation factors in an unsupervised manner. Our approach involves the joint optimization of a generator network along with the shape and deformation factors, supported by specific regularization techniques. For efficient amortized inference of disentangled shape and deformation codes, we train two order-invariant PoinNet-based encoder networks in the second stage of our method. We demonstrate several significant downstream applications of our method, including unsupervised deformation transfer, deformation classification, and explainability analysis. Extensive experiments conducted on 3D human, animal, and facial expression datasets demonstrate that our simple approach is highly effective in these downstream tasks, comparable or superior to existing methods with much higher complexity.

Objective. Colorectal liver metastases (CRLM) represent a major clinical challenge because outcomes after hepatic resection vary widely between patients. Preoperative risk stratification remains limited, and radiomics may provide non invasive imaging biomarkers to support prognosis. This study aimed to develop a CT based radiomics signature capable of generating risk scores for predicting overall survival (OS) in patients undergoing CRLM resection. Approach. Preoperative CT scans from 197 CRLM patients were retrospectively obtained from The Cancer Imaging Archive. A total of 851 radiomics features were extracted using 3D Slicer, along with 256 deep features from a 3D convolutional neural network. Data were randomly divided into training (70%) and testing (30%) sets. Feature selection included correlation-based filtering, univariate Cox regression, and multivariate Cox regression with least absolute shrinkage and selection operator (LASSO) regularization. A radiomics-based risk score (Rad score) was calculated to stratify patients into high- and low- risk groups using the median value. Model performance was compared with clinical variables using time-dependent receiver operating characteristic analysis. Main results. The eight feature signature was prognostic in both internal splits. In multivariable Cox models, the Rad score remained an independent predictor of OS in the training cohort (HR = 2.671, 95% CI 2.023–3.527, P -value < 0.001; Harrell C index = 0.738, 95% CI 0.678–0.793) and the testing cohort (HR = 3.036, 95% CI 1.421–6.486, P -value = 0.004; Harrell C index = 0.663, 95% CI 0.558–0.752). Kaplan Meier analysis showed shorter survival in the high risk group than the low risk group in training (median OS 56.5 versus 76.00 months; log rank P -value = 0.0000) and testing (61.45 versus 74.25 months; log rank P -value = 0.0103). In an external cohort of 105 patients, the Rad score also separated risk (Harrell C index = 0.614, 95% CI 0.534–0.697; median OS 15.7 versus 29.87 months; log rank P -value = 0.0015). Significance. A compact CT radiomics signature derived from preoperative imaging provided independent prognostic information for OS and enabled risk stratification in internal testing and external validation. Further validation in independent CRLM cohorts is required before clinical deployment.