Proceedings of machine learning research最新文献

{"title":"Orientation-Aware Diffusion Super-Resolution for 3T-Like Fetal MRI from Routine 1.5T Scans.","authors":"Xinliu Zhong, Ruiying Liu, Guohao Lin, Chuan Huang, Adam Ezra Goldman-Yassen, Amy Robben Mehollin-Ray, Yun Wang","doi":"","DOIUrl":"","url":null,"abstract":"<p><p>Fetal MRI plays a central role in assessing early brain development. While 3T scanners offer higher SNR and improved cortical detail, their increased sensitivity to motion, susceptibility artifacts, and <math> <msub><mrow><mi>B</mi></mrow> <mrow><mn>1</mn></mrow> </msub> </math> inhomogeneity limits wide adoption for routine fetal imaging. Consequently, most clinical examinations are performed at 1.5T, where greater motion tolerance comes at the cost of lower SNR, reduced gray-white matter contrast, and partial-volume blurring - factors that undermine downstream morphometric analysis. Bridging this quality gap without sacrificing motion robustness of 1.5T would enable 3T-like morphometric reliability in routine clinical acquisitions. We propose an orientation-aware diffusion super-resolution framework that synthesizes 3T-like fetal brain contrast from routine 1.5T scans. The model combines a Swin-UNet backbone with gated FiLM-based orientation embeddings and a residual error-shifting diffusion mechanism. Training leverages the FaBiAN phantom to generate controllable high-/low-resolution pairs with monotonic intensity remapping, geometric perturbations, and simulated signal voids, thereby ensuring generalization to clinical data. Our model produces markedly sharper gyri and mitigates partial-volume effects in both synthesized and clinical data. When evaluated using Fetal-SynthSeg following NeSVoR reconstruction, the framework consistently improves tissue segmentation accuracy over state-of-the-art restoration baselines, yielding more reliable morphometric estimates for fetal brain analysis.</p>","PeriodicalId":74504,"journal":{"name":"Proceedings of machine learning research","volume":"315 ","pages":"3827-3845"},"PeriodicalIF":0.0,"publicationDate":"2026-07-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13148435/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"147847204","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"NeuroLangSeg: Language-Guided Subcortical Segmentation with Pseudo-Supervision and Anatomical-Linguistic Validation.","authors":"Ruiying Liu, Jialu Liu, Xuzhe Zhang, Chuan Huang, Yun Wang","doi":"","DOIUrl":"","url":null,"abstract":"<p><p>Recent advances in vision-language models and LLMs have introduced contextual anatomical reasoning into brain MRI segmentation. However, the field still suffers from a fundamental limitation: the absence of a unified anatomical definition of the structures being segmented. Existing datasets rely on labels produced by heterogeneous manual workflows, often lacking explicit anatomical criteria or consistent annotation standards. As a result, models learn and evaluate within isolated labeling systems, limiting cross-model comparison and valid anatomical measurements. To address these challenges, we introduce <b>NeuroLangSeg</b>, a language-guided framework that enforces a consistent anatomical protocol for subcortical segmentation. A key component of the framework is an anatomical-linguistic evaluator that acts as a training discriminator, encouraging the model to produce outputs by assessing shape characteristics, protocol-defined spatial relationships, and age- and sex-adjusted volumetric norms. Building upon this constraint, NeuroLangSeg integrates a pretrained image encoder with protocol-aligned anatomical prompts and a masked pseudo-labeling strategy, enabling data-efficient and interpretable learning under limited supervision. Together, these components yield anatomically consistent segmentations and support subject-level reporting grounded in a unified anatomical standard. Evaluation across diverse MRI datasets-including comparisons with state-of-the-art models-shows that NeuroLangSeg achieves +4.1 DSC / +8.0 NSD in in-site settings and +3.6 DSC / +14.5 NSD in cross-site generalization over the average baseline, enabled by its LLM-visual integration, while delivering anatomically verifiable predictions suitable for both research and clinical use. GitHub: https://github.com/jlliu2001/SAT_MPL.</p>","PeriodicalId":74504,"journal":{"name":"Proceedings of machine learning research","volume":"315 ","pages":"1565-1597"},"PeriodicalIF":0.0,"publicationDate":"2026-07-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13155643/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"147870685","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Endo-SemiS: Towards Robust Semi-Supervised Image Segmentation for Endoscopic Video.","authors":"Hao Li, Daiwei Lu, Xing Yao, Nicholas Kavoussi, Ipek Oguz","doi":"","DOIUrl":"","url":null,"abstract":"<p><p>In this paper, we present <b>Endo-SemiS</b>, a semi-supervised segmentation framework for providing reliable segmentation of endoscopic video frames with limited annotation. Endo-SemiS uses 4 strategies to improve performance by effectively utilizing all available data, particularly unlabeled data: (1) Cross-supervision between two individual networks that supervise each other; (2) Uncertainty-guided pseudo-labels from unlabeled data, which are generated by selecting high-confidence regions to improve their quality; (3) Joint pseudo-label supervision, which aggregates reliable pixels from the pseudo-labels of both networks to provide accurate supervision for unlabeled data; and (4) Mutual learning, where both networks learn from each other at the feature and image levels, reducing variance and guiding them toward a consistent solution. Additionally, a separate corrective network that utilizes spatiotemporal information from endoscopy video to improve segmentation performance. Endo-SemiS is evaluated on two clinical applications: kidney stone laser lithotomy from ureteroscopy and polyp screening from colonoscopy. Compared to state-of-the-art segmentation methods, Endo-SemiS substantially achieves superior results on both datasets with limited labeled data. The code is publicly available at https://github.com/MedICL-VU/Endo-SemiS.</p>","PeriodicalId":74504,"journal":{"name":"Proceedings of machine learning research","volume":"315 ","pages":"1675-1696"},"PeriodicalIF":0.0,"publicationDate":"2026-07-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13294828/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148347636","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"VFMStitch: A Vision-Foundation-Model Empowered Framework for 3D Ultrasound Stitching via Geometric-Semantic Feature Fusion.","authors":"Xing Yao, Nick DiSanto, Runxuan Yu, Jiacheng Wang, Daiwei Lu, Gabriel Arenas, Baris Oguz, Alison Pouch, Nadav Schwartz, Brett C Byram, Ipek Oguz","doi":"","DOIUrl":"","url":null,"abstract":"<p><p>3D ultrasound (3DUS) stitching expands the field-of-view (FOV) by registering partially overlapping 3DUS volumes acquired from different probe positions. This task is intrinsically difficult due to large inter-volume translations and rotations, the impact of the sector-shaped FOV, as well as the heavy noise and artifacts inherent to ultrasound. With the rapid progress of Vision Foundation Models (VFMs) such as DINOv3, VFM-derived features have recently shown promise for downstream medical image registration tasks. However, existing VFM-based approaches primarily focus on deformable registration and are rarely evaluated for rigid alignment under large motions. Moreover, the feasibility of leveraging VFM-derived features for robust 3DUS stitching remains largely unexplored. In this study, we introduce <b>VFMStitch</b>, the first training-free, VFM-empowered 3DUS stitching framework that integrates point-cloud (PCD)-based geometric features with DINOv3-derived semantic descriptors. Extensive experiments demonstrate that VFMStitch substantially improves rigid registration accuracy compared to existing methods, validating the effectiveness of geometric-semantic fusion for challenging 3DUS stitching scenarios. The code is available at github.com/MedICL-VU/VFMStitch.</p>","PeriodicalId":74504,"journal":{"name":"Proceedings of machine learning research","volume":"315 ","pages":"2621-2639"},"PeriodicalIF":0.0,"publicationDate":"2026-07-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13331288/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148392738","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Quantitative Convergence Analysis of Projected Stochastic Gradient Descent for Non-Convex Losses via the Goldstein Subdifferential.","authors":"Yuping Zheng, Andrew Lamperski","doi":"","DOIUrl":"","url":null,"abstract":"<p><p>Stochastic gradient descent (SGD) is the main algorithm behind a large body of work in machine learning. In many cases, constraints are enforced via projections, leading to projected stochastic gradient algorithms. In recent years, a large body of work has examined the convergence properties of projected SGD for non-convex losses in asymptotic and non-asymptotic settings. Strong quantitative guarantees are available for convergence measured via Moreau envelopes. However, these results cannot be compared directly with work on unconstrained SGD, since the Moreau envelope construction changes the gradient. Other common measures based on gradient mappings have the limitation that convergence can only be guaranteed if variance reduction methods, such as mini-batching, are employed. This paper presents an analysis of projected SGD for non-convex losses over compact convex sets. Convergence is measured via the distance of the gradient to the Goldstein subdifferential generated by the constraints. Our proposed convergence criterion directly reduces to commonly used criteria in the unconstrained case, and we obtain convergence without requiring variance reduction. We obtain results for data that are independent, identically distributed (IID) or satisfy mixing conditions ( <math><mi>L</mi></math> -mixing). In these cases, we derive asymptotic convergence and <math><mi>O</mi> <mfenced> <mrow> <msup><mrow><mi>N</mi></mrow> <mrow><mo>-</mo> <mn>1</mn> <mo>/</mo> <mn>3</mn></mrow> </msup> </mrow> </mfenced> </math> non-asymptotic bounds in expectation, where <math><mi>N</mi></math> is the number of steps. In the case of IID sub-Gaussian data, we obtain almost-sure asymptotic convergence and high-probability non-asymptotic <math><mi>O</mi> <mfenced> <mrow> <msup><mrow><mi>N</mi></mrow> <mrow><mo>-</mo> <mn>1</mn> <mo>/</mo> <mn>5</mn></mrow> </msup> </mrow> </mfenced> </math> bounds. In particular, these are the first non-asymptotic high-probability bounds for projected SGD with non-convex losses.</p>","PeriodicalId":74504,"journal":{"name":"Proceedings of machine learning research","volume":"313 ","pages":"1-39"},"PeriodicalIF":0.0,"publicationDate":"2026-02-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13249411/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148221157","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Probabilistic Nested Homogeneous Spaces for Dimensionality Reduction.","authors":"Xiran Fan, Baba C Vemuri","doi":"","DOIUrl":"","url":null,"abstract":"<p><p>Dimensionality reduction is a key ingredient of many machine learning algorithms and is paramount to their success. For manifold-valued data, the nonlinear equivalent of the well-known principal component analysis (PCA), called, principal geodesic analysis (PGA) is used quite often. An alternative to PGA that is more general and flexible, called \"Nested Homogeneous Spaces (NHS)\" for dimensionality reduction of manifold-valued data was recently introduced. In this paper, we present a novel probabilistic version of the NHS model (PNHS) for dimensionality reduction of high dimensional manifold-valued data in Riemannian homogeneous spaces. The PNHS model has several advantages over its deterministic counterpart namely, the NHS model. In particular, the ability to, quantify uncertainty in parameter estimates and tackle missing data. We demonstrate these advantages via real and synthetic data examples.</p>","PeriodicalId":74504,"journal":{"name":"Proceedings of machine learning research","volume":"282 ","pages":"1033-1047"},"PeriodicalIF":0.0,"publicationDate":"2026-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13592668/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149019185","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Meta-iLaD: Identifiable Latent Dynamics via Meta-Learning of Dynamics Environments.","authors":"Yubo Ye, Sweekar Piya, Xiajun Jiang, Linwei Wang","doi":"","DOIUrl":"","url":null,"abstract":"<p><p>Learning <i>latent dynamics</i> is central to assessing the current states and forecasting the future trajectories of high-dimensional time series. For locally-stationary latent dynamics <math><mi>ℱ</mi> <mfenced> <mrow> <msub><mrow><mi>z</mi></mrow> <mrow><mo><</mo> <mi>t</mi></mrow> </msub> <mo>;</mo> <mi>c</mi></mrow> </mfenced> </math> with latent dynamics state <math> <msub><mrow><mi>z</mi></mrow> <mrow><mi>t</mi></mrow> </msub> </math> and environment variable <math><mi>c</mi></math> , prior identifiability results have largely focused on <math> <msub><mrow><mi>z</mi></mrow> <mrow><mi>t</mi></mrow> </msub> </math> when conditioned on predefined label <math><mi>u</mi></math> of dynamics environments. This leaves two limitations: reliance on predefined labels that hinder generalization to unseen environments, and a limited understanding of the identifiability of <math><mi>ℱ</mi></math> and <math><mi>c</mi></math> which-although offering important structural properties for the identifiability of <math> <msub><mrow><mi>z</mi></mrow> <mrow><mi>t</mi></mrow> </msub> </math> -are learned jointly with <math> <msub><mrow><mi>z</mi></mrow> <mrow><mi>t</mi></mrow> </msub> </math> . We address these limitations with Meta-iLaD, a novel identifiable latent dynamics framework attained by meta-learning across dynamics environments. Meta-iLaD introduces a novel conditional prior of <math><mi>c</mi></math> , modeled as a feedforward meta-learner to rapidly extract <math><mi>c</mi></math> from few-shot examples. Meta-iLaD further establishes identifiability for <math> <msub><mrow><mi>z</mi></mrow> <mrow><mi>t</mi></mrow> </msub> <mo>,</mo> <mi>c</mi></math> , and <math><mi>ℱ</mi></math> simultaneously, for a general formulation of <math><mi>ℱ</mi> <mfenced> <mrow> <msub><mrow><mi>z</mi></mrow> <mrow><mo><</mo> <mi>t</mi></mrow> </msub> <mo>;</mo> <mi>c</mi></mrow> </mfenced> </math> without restricting the dimension of <math><mi>c</mi></math> or how it modulates <math><mi>ℱ</mi></math> . On synthetic and real data, we provide strong empirical evidence that 1) conditioning on few-shot examples enables generalization to out-of-distribution environments, and 2) identifiability for <math><mi>c</mi></math> and <math><mi>ℱ</mi></math> is critical for accurate forecasting beyond reconstructing observed trajectories.</p>","PeriodicalId":74504,"journal":{"name":"Proceedings of machine learning research","volume":"306 ","pages":"148560-148582"},"PeriodicalIF":0.0,"publicationDate":"2026-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13626392/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149119935","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"EndoStreamDepth: Temporally Consistent Monocular Depth Estimation for Endoscopic Video Streams.","authors":"Hao Li, Daiwei Lu, Jiacheng Wang, Robert J Webster, Ipek Oguz","doi":"","DOIUrl":"","url":null,"abstract":"<p><p>This work presents <b>EndoStreamDepth</b>, a monocular depth estimation framework for endoscopic video streams. It provides accurate depth maps with sharp anatomical boundaries for each frame, temporally consistent predictions across frames, and real-time throughput. Unlike prior work that uses batched inputs, EndoStreamDepth processes individual frames with a temporal module to propagate inter-frame information. The framework contains three main components: (1) a single-frame depth network with endoscopy-specific transformation to produce accurate depth maps, (2) multi-level Mamba temporal modules that leverage inter-frame information to improve accuracy and stabilize predictions, and (3) a hierarchical design with comprehensive multi-scale supervision, where complementary loss terms jointly improve local boundary sharpness and global geometric consistency. We conduct comprehensive evaluations on two publicly available colonoscopy depth estimation datasets, with quantitative results reported on phantom and simulated data that provide ground truth depth. Compared to state-of-the-art monocular depth estimation methods, EndoStreamDepth substantially improves performance, and it produces depth maps with sharp, anatomically aligned boundaries, which are essential to support downstream tasks such as automation for robotic surgery. The code is publicly available at https://github.com/MedICL-VU/EndoStreamDepth.</p>","PeriodicalId":74504,"journal":{"name":"Proceedings of machine learning research","volume":"315 ","pages":"1697-1721"},"PeriodicalIF":0.0,"publicationDate":"2026-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13441292/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148681222","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"LOTUS: Latent Outpainting Diffusion Model for Three-Dimensional Ultrasound Stitching.","authors":"Xing Yao, Runxuan Yu, Nick DiSanto, Ehsan K Aghdam, Kanyifeechukwu Oguine, Daiwei Lu, Ange Lou, Jiacheng Wang, Dewei Hu, Gabriel Arenas, Baris Oguz, Alison Pouch, Nadav Schwartz, Brett C Byram, Ipek Oguz","doi":"","DOIUrl":"","url":null,"abstract":"<p><p>3D ultrasound (3DUS) stitching can enlarge the field-of-view (FOV) by registering partially overlapping 3DUS images collected from different probe positions. However, standard registration algorithms frequently encounter difficulties with this task, primarily due to the sector-shaped FOV, which often leads to pronounced local minima, thereby obstructing optimization efforts. To address these limitations, we propose LOTUS, a novel Latent Diffusion Model (LDM) specifically designed for 3DUS FOV outpainting. LOTUS innovatively encodes the 3DUS data into a compact latent space and performs outpainting at test time, effectively extending the sector-shaped FOV into a standard rectangular shape. This transformation facilitates a more robust registration by mitigating the issues of local minima associated with the original FOV shape. Experimental results show that LOTUS significantly improves the accuracy of the registration as well as the efficiency of the outpainting process compared to existing models. The code is available at github.com/MedICL-VU/LOTUS.</p>","PeriodicalId":74504,"journal":{"name":"Proceedings of machine learning research","volume":"306 ","pages":"1740-1754"},"PeriodicalIF":0.0,"publicationDate":"2026-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13345505/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148414297","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Partial VOROS: A Cost-aware Performance Metric for Binary Classifiers with Precision and Capacity Constraints.","authors":"Christopher Ratigan, Kyle Heuton, Carissa Wang, Lenore Cowen, Michael C Hughes","doi":"","DOIUrl":"","url":null,"abstract":"<p><p>The ROC curve is widely used to assess binary classifiers. Yet for some applications, such as alert systems for monitoring hospitalized patients, conventional ROC analysis cannot meet two key deployment needs: enforcing a constraint on precision to avoid false alarm fatigue and imposing an upper bound on the number of predicted positives to represent the capacity of hospital staff. The usual area under the curve metric also does not reflect asymmetric costs for false positives and false negatives. In this paper we address all three of these issues. First, we show how the subset of classifiers that meet precision and capacity constraints occupy a feasible region in ROC space. We establish the polygonshaped geometry of this region. We then define the <i>partial area of lesser classifiers</i>, a performance metric that is monotonic with cost and only accounts for the feasible region. Averaging this area over a desired distribution for cost parameters results in the partial volume over the ROC surface, or <i>partial VOROS</i>. In experiments predicting mortality risk from vital sign history on several datasets, we show this cost-aware metric can outperform alternatives at ranking classifiers for in-hospital alerts.</p>","PeriodicalId":74504,"journal":{"name":"Proceedings of machine learning research","volume":"300 ","pages":"2080-2088"},"PeriodicalIF":0.0,"publicationDate":"2026-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13583593/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148993055","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}