Papers for

medical image analysis teams

Papers whose findings have a practical use for this group, as judged from the abstract. Open a paper to read what it means in practice.

Efficient classification of kidney disease genes from 3D MRI scans

Efficient Genotype Classification from 3D MRI in ADPKD via Density-Aware Point Sampling and E(3)-invariance

Abstract: Early detection of genetic kidney diseases, especially autosomal dominant polycystic kidney disease (ADPKD), is crucial because certain genotypes such as truncating PKD1 mutations cause rapid kidney growth and early progression to end-stage kidney disease. MRI-based genotype inference is a potential low-cost complement to genetic testing. However, although many deep learning methods have been developed for classifying various MR images, applying them to ADPKD remains challenging: (1) abdominal MR images are generally too large to be handled by deep learning models designed for smaller images like brain MRI, and (2) the amount of labeled data for ADPKD genetic mutations is insufficient to train deep learning models. In order to address these challenges, we propose E3PN, an efficient 3D MRI-based ADPKD genotype classification network that relies on point sampling and E(3)-invariance. We utilize a point representation of 3D MRI, which consumes less GPU memory than a voxel representation. For further memory efficiency, 3D points are sampled from a mixture distribution over the two kidneys in each image. Since E(3)-invariance has been shown to improve data efficiency, we also incorporate it into our model through coordinate canonicalization. On our ADPKD dataset, E3PN trains 163 times faster and uses 103x less GPU memory than a 3D ResNet, while also improving F1 from 0.165 to 0.226 over both baselines. These results establish a first benchmark for five-way ADPKD genotype classification from 3D MRI and show that point sampling and coordinate canonicalization yield two-orders-of-magnitude efficiency gains with a modest but consistent improvement in F1.

Fri 13 NovComputer Vision and Pattern RecognitionMachine Learning
The gist
Certain inherited kidney diseases can cause kidneys to grow quickly and fail early, so spotting them early is important. The authors created a method to identify specific gene types causing this disease using 3D MRI images, which are large and hard for normal computer models to handle. They represent the MRI data as points focused on kidneys and use a technique that treats the images as the same even if rotated or moved, making the computer work more efficiently. Their method uses much less computer memory and trains much faster than traditional approaches, while also improving accuracy a bit. This sets a new standard for identifying these gene types from MRI scans.
Open → 2611.17758v1

Anchored slice propagation improves 3D CT scan organ segmentation accuracy

Anchored Slice Propagation with Boundary Transport for 3D CT Segmentation from Three Orthogonal Annotated Slices

Abstract: Sli2Vol propagates a binary mask through affinities between adjacent slices, but an error can persist because each prediction becomes the input to the next step. Three orthogonal annotations reduce dependence on any one chain, yet they offer more than three starting masks: where a chain crosses another annotated plane, the correct label is already known. Those labels could correct drift during propagation and indicate how much to trust each chain at fusion. Between annotated planes, the information carried by the chain may matter as well. A binary mask records foreground membership but discards distance to the boundary, a cue that could help follow gradual shape changes between slices.We address these two sources of error with a fixed pretrained affinity backbone. We anchor each chain by correcting its prediction where it crosses another annotated plane, then use its agreement with those labels to weight fusion. We compare this anchored mask with two ways of transporting boundary geometry: a truncated signed-distance field (SDF) propagated by dense affinity averaging, and a contour whose vertices follow motion cues from the same affinity. The contour tracker starts from a hand-crafted rule and learns a motion correction from supervoxel pseudo-objects in 506 CT volumes, without manual organ labels. On 2,091 held-out structure instances from approximately 403 CT scans spanning 17 abdominal structures, anchoring raises mean Dice from 0.737 with orthogonal mask propagation to 0.769 and improves every structure. Beyond the anchored mask, learned contour tracking improves Dice by 5.7\% for the esophagus and 5.9\% for the postcava, while SDF transport improves right-adrenal Dice by 2.8\%. The anchored methods have similar mean Dice (0.769--0.772); contour tracking performs worse on the kidneys, stomach, and hepatic vessel tree.

Tue 3 NovComputer Vision and Pattern Recognition
The gist
Segmenting organs in 3D CT scans is hard because errors in one slice can affect the next when filling in slices between just a few annotated ones. The authors show that using labels from three orthogonal annotated slices as anchors helps correct mistakes as the segmentation spreads through the volume. They also use ways to carry shape boundary details between slices instead of just black-and-white masks. Their approach improves accuracy on several abdominal organs compared to previous mask propagation methods.
Open → 2611.05693v1

Fusion of images and clinical data improves inflammatory bowel disease predictions

Multimodal Fusion Methods for Whole-Slide Images and Clinical Data in Inflammatory Bowel Disease

Abstract: Multimodal AI fusion of histology and clinical data is an active research area in computational pathology, but fusion methods are rarely compared under identical conditions and across different clinical endpoints. We present a modular framework in which slide-level embeddings, obtained by training an attention-based aggregator over patch embeddings from a frozen foundation model, and per-visit clinical embeddings are fed into 15 interchangeable fusion methods with a shared output dimension and prediction head (except decision-level methods). Using a prospective cohort of newly diagnosed inflammatory bowel disease patients, we predicted STRIDE-II-aligned remission outcomes from baseline or follow-up visits. The best fusion method reached macro-F1 of 0.57-0.75, outperforming both single-modality models in four of six scenarios, and differed by outcome and input. Image reliance, the drop in balanced accuracy when the slide embedding is removed, tended to follow the remission type - highest for histological and lowest for endoscopic remission; methods with similar accuracy relied on the slide to very different degrees.

Tue 3 NovComputer Vision and Pattern RecognitionArtificial IntelligenceMachine Learning
The gist
Predicting how inflammatory bowel disease patients will respond to treatment is challenging. The authors combined detailed images of tissue samples with patient clinical information using 15 different ways to merge these data types. Their best method predicted disease remission better than using images or clinical data alone, but the importance of the images varied depending on the remission type. This work helps understand how blending different medical data can improve disease outcome predictions.
Open → 2611.05579v1

OpenBrain release rates whole brain MRI parcellation confidence

OpenBrain: An Auditable Generated-Label Release for Whole-Brain MRI Parcellation

Abstract: Automatic tools can now generate whole-brain MRI parcellations at cohort scale, but generated-label releases often provide little evidence about which cases are reliable enough to reuse. We introduce OpenBrain, a confidence-aware release of 35,838 provenance-verified OpenNeuro T1w cases from 607 source datasets. OpenBrain is organized as an auditable case-record release rather than a directory of standalone segmentation files: each record links a primary parcellation, two committee parcellations, source/license provenance, raw QC-7 measurements, percentile risks, a combined risk score R, and a confidence tier. QC-7 is a reference-free reliability panel over label geometry, image-label consistency, and committee disagreement, so users can filter, weight, audit, or re-threshold generated labels without rerunning inference. We evaluate this released evidence under a frozen protocol. Across external cohorts, lower QC-7 risk corresponds to higher generated-label quality; in fixed-budget training, lower-risk selections improve paired macro Dice over random selection by +3.2 to +3.7 percentage points. OpenBrain also exposes raw axes, committee outputs, provenance, and datasheet documentation, so confidence claims remain inspectable after download. It does not turn generated labels into manual annotations or clinical-grade segmentations. Its contribution is a large open generated-parcellation substrate, a reusable case-level risk-evidence layer, and an external validation protocol for release-time QC. Dataset: https://huggingface.co/datasets/lanqz7766/OpenBrain. Code: https://github.com/lanqz7766/openbrain.

Tue 3 NovComputer Vision and Pattern Recognition
The gist
Whole-brain MRI images can be automatically divided into regions, but it's hard to know how reliable these automatic labels are. The authors created OpenBrain, a large release of over 35,000 brain images with detailed confidence scores and quality checks for each case. This lets users pick or trust brain segmentations based on clear evidence without redoing the labeling. Tests show that focusing on more reliable cases improves accuracy in training brain analysis models. OpenBrain also provides full data records so users can verify and understand confidence levels after downloading.
Open → 2611.05348v1

Tissue-aware deep learning improves nuclei segmentation in cancer images

TCINet: Tissue-Conditioned Adaptive Cross-Branch Interaction Network for Nuclei Instance Segmentation

Abstract: Separating touching nuclei requires local boundary detail, while cell type classification requires broader morphological context. These requirements also vary with the surrounding tissue environment. We propose the Tissue-Conditioned Adaptive Cross-Branch Interaction Network (TCINet), a single-stage model with Task-Specific Feature Adapters (TSFA) and Tissue-Conditioned Task-Aware Cross-Branch Interaction (TC-TACBI). TSFA produces branch-specific features at skip connections, and TC-TACBI controls three asymmetric paths among nucleus-presence (NP), horizontal-vertical (HV), and nucleus-type (NT) branches using predicted tissue probabilities. On PanNuke under standard three-fold cross-validation, TCINet achieves multiclass and binary panoptic quality scores of 0.519 and 0.692, respectively. Zero-shot tests on MoNuSeg yield a detection F1 score of 0.876. On PanNuke fold 0, adding TC-TACBI increases mPQ from 0.505 to 0.525. Code is available at https://github.com/celestial1321/TCINet.

Tue 3 NovComputer Vision and Pattern Recognition
The gist
Distinguishing touching cell nuclei and identifying their types in microscope images is tricky because it needs both detailed edges and broader cell shape information, which also change depending on the tissue. The authors created a new AI model called TCINet that adapts its tasks based on the tissue type and handles these challenges simultaneously. Their approach improves accuracy in segmenting and classifying nuclei on a large dataset and even works well when tested on images it was not trained on. This model uses special components to share and adapt information between different prediction tasks based on tissue conditions.
Open → 2611.04235v1

Multi-agent system generates realistic tumors with fine control on medical images

MATS: A Multi-Agent Framework for Controllable and Reflective Tumor Synthesis

Abstract: Synthesizing realistic tumors on medical images helps address data scarcity for clinical AI. Existing diffusion-based methods, however, lack fine-grained attribute-level controllability, closed-loop quality verification, and awareness of which morphological patterns are underrepresented in the target distribution. We present MATS, a multi-agent framework for text-controllable, cross-organ tumor synthesis. Given a target disease, MATS first profiles its morphological distribution over a fine-grained attribute space covering tumor geometry and appearance, and constructs a generation plan that amplifies underrepresented subtypes while keeping common ones dominant. It then localizes the target organ, proposes anatomically valid placement, and executes two-stage diffusion to produce the tumor mask and render texture, each conditioned on the corresponding attributes. A closed-loop reflective mechanism verifies attribute alignment at three checkpoints throughout synthesis and provides targeted feedback to guide re-generation when misalignment is detected. Trained exclusively on a single-source pancreatic neuroendocrine tumor dataset, MATS consistently improves downstream segmentation both in-domain and across LiTS, KiTS, and MSD-Pancreas benchmarks, whereas most prior generative augmentation methods fall below real-only training.

Fri 30 OctComputer Vision and Pattern Recognition
The gist
Medical AI needs more data to recognize different tumors well, but real patient data is limited. The authors created a system called MATS that can generate realistic tumor images with detailed control over tumor shape and appearance. MATS uses a smart plan to create more examples of rare tumor types while keeping common ones frequent. It also checks its own work during generation to ensure the tumors match the desired features. This helps improve tumor detection AI even when trained on limited real data.
Open → 2610.41159v1

Semantic spatial residual banks improve multi mask medical image segmentation

Single-Annotation, Multi-Hypothesis Segmentation with Semantic-Spatial Residual Banks

Abstract: Multi-rater segmentation datasets show that a single image can admit several valid masks, yet most training sets expose only one annotation per image. We study single-annotation multi-hypothesis segmentation: each training image exposes one fixed mask, while the model must return a finite bank that covers held-out human annotations at test time. Our experiments reveal a representation tradeoff. Banks learned directly from images recover variations in local boundaries and extent but lose structural hypotheses, whereas frozen semantic features preserve structural grouping but underrepresent spatial variation. We introduce the Semantic-Spatial Residual Bank (SSRB), which freezes a semantic bank trained with balanced assignments and learns a residual from the image for each hypothesis. Across five medical endpoints, SSRB improves coverage of held-out masks at $K=8$ over a Probabilistic U-Net under the same one-mask constraint. On LIDC and LIDC-hard, it also produces substantially more distinct useful masks. On BSDS, it preserves the semantic bank's performance in grouping regions. Its advantage grows as more masks are returned, showing that residuals learned for corresponding heads improve coverage of local annotation variation while preserving how regions are grouped.

Thu 29 OctComputer Vision and Pattern RecognitionMachine Learning
The gist
Some images can be segmented in different valid ways, but training data usually only shows one example. The authors study how to train models that, given one example mask per image, can produce multiple masks that cover other possible human annotations. They find that combining fixed semantic features with learned image residuals helps models capture both overall structure and local details in segmentations. Their new method, Semantic-Spatial Residual Bank, performs better on medical images and natural images compared to previous approaches when multiple masks are needed.
Open → 2610.41069v1

Lychee speeds up high-accuracy vector search on large datasets

Lychee: Certified Screening of Inverted Lists for High-Recall Vector Search

Abstract: Nearest-neighbor search over embedding vectors matches faces against watchlists, transactions against known fraud patterns, and medical images against confirmed cases, where a missed neighbor is a missed threat, fraud, or diagnosis. Because an approximate index cannot report which queries lost a true neighbor, these applications require recall of 0.999 to 0.9999, at a cost far below that of exact search over billions of vectors. Inverted-file indexes reach high recall by probing more lists, but they read and score every probed vector at full precision, so that throughput falls in inverse proportion to the number of probed lists. Graph indexes saturate below these targets, whereas quantized lists must rerank their candidates with the full vectors. Our insight is to decide for almost every probed vector that it cannot enter the top k of the probed lists without reading the vector, through the certified lower bound of exact search by certified exclusion. This paper turns this insight into Lychee, a GPU inverted-file search that screens the probed lists on tensor cores under a threshold from anchors of the same lists and computes exact distances only for the few survivors. We prove that Lychee returns exactly the k nearest vectors of the probed lists, the answer of an inverted-file search at full precision, while it reads a few hundred full vectors per query regardless of the number of probed lists. On eleven datasets from one million to one billion vectors and against ten GPU and CPU baselines, Lychee is 9.7 to 117 times faster than IVF-Flat at equal recall with 200 probed lists on datasets of up to 100 million vectors. It also reaches recall levels that CAGRA and HNSW do not reach and outruns exact search by certified exclusion by 5 to 18 times above ten million vectors. On four GPUs, Lychee searches one billion vectors at recall 0.9999, 44.5 times faster than IVF-Flat on BIGANN-1B.

Thu 22 OctInformation Retrieval
The gist
Searching for the closest items in large collections of data is used in things like recognizing faces, spotting fraud, or diagnosing diseases. The authors developed Lychee, a method that quickly filters out unlikely matches before doing detailed checks, which saves a lot of time. This works even when searching billions of items and still finds nearly all the closest matches. Lychee uses graphics processors to do this filtering efficiently and maintains very high accuracy.
Open → 2610.30697v1

Thyroid ultrasound classification with interpretable measurement based model

Learning to Measure and Diagnose: A Clinical-Dictionary Radiomic Bottleneck Model for Thyroid Nodule Classification

Abstract: Thyroid ultrasound supports the assessment of nodule malignancy, but classification predictions often lack evidence linked to familiar clinical descriptions. Radiomics provides explicitly defined measurements that can support such explanations through a clinical dictionary. However, using these measurements as interpretable evidence requires reliable estimation from ultrasound images and preservation of measurement fidelity when optimizing classification performance. We propose a clinical-dictionary radiomic bottleneck model for thyroid nodule classification. Within this model, Family-Aware Radiomic Learning (FARL) estimates measurements with balanced supervision across feature families. Fidelity-Guided Bottleneck Modeling (FGBM) constructs a fixed bottleneck, followed by joint optimization of measurement estimation and classification under fidelity guidance. Decision evidence decomposition then traces feature contributions through a linear classifier and organizes them by clinical dictionary categories. Experiments on ThyroidXL and TN3K show competitive classification performance, while case studies illustrate feature contributions for benign and malignant nodules. These findings indicate interpretable classification with inspectable quantitative evidence informed by clinical knowledge. Code is available online.

Wed 21 OctComputer Vision and Pattern Recognition
The gist
Thyroid ultrasound images help doctors figure out if nodules are cancerous, but current computer methods often don’t explain their decisions clearly. The authors created a model that estimates specific measurable features from images that doctors know about, and uses these features to decide if a nodule is benign or malignant. Their model also shows which features influenced the decision, making it easier for doctors to trust and understand the result. Tests showed the model performs well and provides clear evidence linked to medical knowledge.
Open → 2610.29738v1

MRI super-resolution model reconstructs varied image resolutions quickly

Arbitrary-Scale Anisotropic 3D MRI Super-Resolution with Gaussian Prototypes

Abstract: Reducing MRI acquisition time improves patient comfort, reduces motion artefacts, and increases scanner throughput. Clinical protocols often shorten acquisitions by sacrificing spatial resolution, particularly along the slice axis. Super-resolution can recover higher-resolution images, but existing methods are typically specialised to one scale or acquisition geometry, limiting their use across heterogeneous clinical protocols. We present GSMR3D, a continuous volumetric model that reconstructs arbitrary isotropic and anisotropic resolutions with a single checkpoint per MRI contrast. It combines a learned dictionary of anisotropic 3D Gaussian prototypes with scale-conditioned frequency encoding and an exact measurement-consistency projection, keeping reconstructions consistent with the acquired low-resolution data. We train separate models on 3,731 contrast-enhanced T1-weighted and 30,254 T1-weighted brain MRI volumes from clinical and population-based cohorts. We evaluate GSMR3D against geometry-specific specialists for isotropic, in-plane, and through-plane super-resolution using reconstruction quality and downstream tasks: brain tumour segmentation, brain-metastasis detection, and brain parcellation. GSMR3D outperforms specialists at isotropic $2\times$ and achieves better LPIPS, DISTS, and FID across all geometries, although specialists retain higher PSNR in anisotropic settings. It more faithfully preserves glioma segmentations across all geometries, achieves the highest lesion sensitivity on an independent brain-metastasis cohort, and the highest brain-parcellation agreement for isotropic and in-plane reconstruction. Reconstruction takes about 6s per volume, adding little overhead relative to acquisition-time savings. GSMR3D offers a flexible approach to MRI scan-time reduction with one model per contrast across geometries while preserving information relevant to downstream tasks.

Sat 17 OctComputer Vision and Pattern Recognition
The gist
MRI scans often take a long time, so doctors reduce image detail to speed things up. The authors introduce a model called GSMR3D that can improve MRI images to higher resolution, no matter how the original scan was done. This model works well across different scanning styles and helps keep important medical details visible. It does this quickly, potentially allowing shorter scans without losing critical information.
Open → 2610.24858v1

Invertible adapters improve latent space alignment for medical image generation

Semantic Latent Alignment for Flow Matching with Invertible Bi-Lipschitz Adapters

Abstract: Pretrained latent spaces trade compression, reconstruction, and the learnability of the coordinates a downstream generator has to work in, which complicates adapting these spaces to different domains or tasks: continued training is expensive, and without a new alignment objective it often fails to improve downstream generation. We instead leave the encoder and decoder fixed and change only the coordinates between them, using an invertible bi-Lipschitz adapter aligned offline to a semantic teacher and frozen during class-conditional flow matching. Invertibility preserves compatibility with the pretrained autoencoder, while the bi-Lipschitz constraint limits how far alignment can distort its latent geometry. This provides a simple way to improve semantic organization while retaining the compression and reconstruction properties of the original latent space. On two medical imaging datasets, constrained latent adaptation improves feature metrics, seed stability and counterfactual edit strength, while remaining competitive on downstream classification. In contrast, unconstrained adaptation can become highly anisotropic and sensitive despite remaining invertible, highlighting the importance of controlling geometric distortion. Our results suggest that carefully constrained post-hoc adaptation offers a parameter-efficient and tuning-free alternative to jointly modifying the representation during generative training.

Thu 15 OctComputer Vision and Pattern RecognitionMachine Learning
The gist
It can be hard to adjust a computer’s understanding of compressed images to do new tasks without retraining everything, which is costly and tricky. The researchers found a way to keep the image encoder and decoder fixed but change the coordinates between them using a special invertible adapter that preserves structure and semantic meaning. This approach improves image features and stability in medical images while avoiding distortions that can cause problems. Their method is a simpler, more efficient way to adapt pretrained models for new image generation tasks without retraining the whole system.
Open → 2610.21332v1

Cardiac MRI sequences generated from single 3D image without ECG

One Frame, Full Heartbeat: ECG-Free 4D Cardiac Cine MRI Synthesis via Radial-Decomposed Flow Matching

Abstract: Cine cardiovascular magnetic resonance (CMR) captures the cardiac cycle as a four-dimensional (4D) sequence, but standard acquisition requires electrocardiogram (ECG) gating and repeated breath holds. Visual realism alone does not establish accurate patient-specific ejection fraction (EF) or ventricular volumes. We present PhaseFlow3D, a generative framework that synthesizes a complete 4D cine sequence from a single end-diastolic (ED) three-dimensional (3D) volume without ECG. To capture asymmetric systolic and diastolic dynamics, it represents the cardiac cycle as a piecewise linear phase anchored at ED and end-systolic (ES) time points. At inference, a population-level canonical template supplies this phase without patient-specific temporal information. A phase-conditioned rectified flow model generates a cardiac motion trajectory in latent space. Radial Contraction Decomposition converts each latent state into a 3D displacement field, combining a physics-informed radial component for centripetal myocardial contraction with an image-conditioned residual for rotation and out-of-plane motion. Each frame is generated by directly warping the ED volume, bypassing variational autoencoder decoding. On the combined ACDC and M&Ms benchmark, PhaseFlow3D achieves the lowest EF mean absolute error, the only positive left-ventricular volume-curve $R^2$, and the best distributional quality among compared methods. Ablations confirm each component's contribution. Downstream evaluations demonstrate the utility of the synthesized sequences and displacement fields for segmentation, pathology classification, label propagation, and myocardial strain analysis.

Tue 6 OctComputer Vision and Pattern Recognition
The gist
Traditional heart MRI scans need an ECG and multiple breath holds to capture the whole beating cycle. The authors developed a method called PhaseFlow3D that can create a full 4D heart motion sequence from just one 3D scan taken at one moment, without using an ECG. Their technique models the heart's motion by combining a known motion pattern with patient-specific adjustments, producing realistic images that accurately measure heart function. Tests show this method improves the accuracy of heart volume and function estimates compared to previous approaches.
Open → 2610.09185v1

Model predicts cancer survival by matching images and gene data

Anchor-driven Multi-modal Multi-scale Expert Selection for Survival Prediction

Abstract: The integrative analysis of histopathological Whole-Slide Images (WSIs) and transcriptomic profiles holds significant promise for cancer survival prediction. However, existing methods typically project multi-modal features directly into a shared latent space without explicit alignment, leading to the entanglement of mismatched morphological cues and molecular signals. Furthermore, current fusion strategies often treat the extreme spatial heterogeneity of WSIs uniformly, lacking mechanisms to adaptively prioritize clinically relevant tissue scales for individual patients. To address these limitations, we propose an Anchor-driven Multi-modal Multi-scale Expert Selection (AM$^2$ES) framework for survival prediction. Specifically, we present an Anchor-driven Multi-modal Fusion (AMF) module, which introduces learnable semantic anchors as cross-modal mediators to bridge the semantic gap by enforcing a structurally regularized alignment between transcriptomic features and multi-scale pathology representations. Built upon this aligned semantic space, we further design a Hierarchical Mixture-of-Experts (H-MoE) selection module to decouple the hierarchical prognostic selection process. Mimicking the pathologist's diagnostic workflow, H-MoE performs (i) Intra-scale Expert Filtering to discriminatively identify salient tumor regions within each magnification, and (ii) Inter-scale Hierarchy Routing to dynamically weight and select the most informative resolution levels. Extensive experiments on multiple TCGA cancer cohorts demonstrate that our AM$^2$ES achieves state-of-the-art performance while offering fine-grained interpretability by visualizing how specific molecular pathways drive the expert routing decisions across tissue scales. The code will be released at https://github.com/taozh2017/AM2ES.

Tue 6 OctComputer Vision and Pattern Recognition
The gist
Predicting how long cancer patients might survive can be hard because tumor images and gene tests are very different types of data. The authors created a method that helps these two kinds of information 'talk' to each other better by using special guide points called anchors. Their system also mimics how doctors look at tumors at different zoom levels, picking the most important parts and scales for each patient. This approach works better than past methods and shows which genes influence the decisions at different tissue details.
Open → 2610.07694v1

Signal noise factorization improves image recognition under distortion

Signal-Noise Factorization Isolates Nuisance Variation into Removable Subspaces

Abstract: Recent theoretical work identified fundamental properties of representation geometry that shape inference ability of deep neural networks. These include signal-noise factorization (SNF), the ability to segregate signal from noise, and signal-signal factorization (SSF), the ability to segregate task-specific and task-irrelevant signals. Here, we built regularizers that reinforce these two properties during training. We compared networks trained with these regularizers to $L_2$-regularized baseline networks on the CIFAR-100 classification task to understand how our regularizers shape representation geometry and impact performance on a well-known computer vision baseline. Enhancing SNF via regularization improved model performance but enhancing SSF did not. Motivated by biomedical applications, we investigated how our regularizers affected performance on the BloodMNIST dataset treated with MedMNIST-C corruptions at five severity levels, and found even larger performance gains using the SNF regularizer. To understand the mechanism by which SNF-regularization produces improved performance, we analyzed the nuisance subspaces across regularization regimes, finding that the SNF-regularized models represent noise in distinct subspaces, separate from class-relevant signal. Because this geometry is explicit, the dominant corruption-induced directions can be estimated on held-out data and projected out of the representations. This manipulation led to a substantial gain in accuracy. These results show that regularizers that enforce signal-noise factorization can produce substantial improvements on computer vision tasks that contain out-of-distribution image distortions at inference time. They also highlight how shaping representations affects model performance: isolating nuisance variables from categorical ones is more important than maintaining factorized representations of categorical variables.

Wed 30 SeptMachine LearningComputer Vision and Pattern Recognition
The gist
Deep neural networks can sometimes mix up important information (signal) with distracting details (noise), which hurts their performance. The authors created training techniques to help networks separate these clearly. Their approach helped models do better, especially on images with various distortions and noise. By isolating the distracting noise into separate parts, they could even remove those parts to get more accurate results. This shows that focusing on separating noise from signal is more important than separating different types of important information.
Open → 2610.00751v1

New method improves error control in predictive models for varied inputs

ReCIRC: Rectified Conformal Risk Control

Abstract: Many applications of black-box predictive models require controlling task-relevant error rates, such as missed lesion pixels in segmentation or missed labels in multilabel classification. Conformal risk control (CRC; Angelopoulos et al., arXiv:2208.02814) gives distribution-free guarantees for such losses, but it calibrates a single threshold shared by all inputs. Because conditional risk varies with the input, this marginal guarantee often overprotects easy cases and underprotects hard ones. We propose ReCIRC (Rectified Conformal Risk Control), which inverts each input's estimated local risk curve to reparameterize the calibrated threshold as a risk budget $a$ representing a common target conditional risk, and then applies CRC unchanged to the resulting family. ReCIRC retains CRC's finite-sample marginal guarantee regardless of the accuracy of the estimated curves, while accurate curves yield approximate conditional risk control and, under additional conditions, asymptotically exact conditional risk control; they also support a risk-calibration diagnostic. Across three synthetic and five real-data settings spanning segmentation, multilabel and multiclass classification, and regression, ReCIRC attained the lowest average worst-group risk and mean positive group excess in every setting, while maintaining marginal risk close to the target, whereas changes in prediction size were application-dependent.

Tue 29 SeptMachine Learning
The gist
Controlling errors in AI predictions can be hard because some cases are easier and some are harder. The authors propose ReCIRC, a way to adjust error limits based on how risky each case is, rather than a single rule for all. This helps avoid being too strict for easy cases or too loose for difficult ones. ReCIRC keeps strong overall error guarantees while better balancing errors across different inputs.
Open → 2609.38112v1

Ai segments tumors using images and reports without masks

RT-Super: Learning Tumor Segmentation from Longitudinal Images and Reports

Abstract: Multi-tumor segmentation is important for early cancer detection and allows radiologists to visualize, verify, and understand AI predictions. However, tumor segmentation masks are expensive, time-consuming, and unavailable for many tumor types in public data. Instead, hospitals have vast, readily available data that can guide segmentation: radiology reports, longitudinal images, and multi-phase images. We use this readily available data to substitute for tumor masks in training AI for tumor segmentation. To this end, we propose a new architecture, RT-Super. It has a teacher network, which analyzes the patient's longitudinal images and reports to create high-quality tumor masks. These masks train a student network, which sees a single image and no report. At inference, when longitudinal images and reports are unavailable, we use the student. RT-Super uses a new CNN-Transformer architecture and novel Consistency Losses that exploit tumor location consistency across longitudinal images. We train RT-Super to segment esophagus, uterus and spleen tumors, which have few or no public masks. Even without training masks, RT-Super can segment these tumors and surpass public AI models. Overall, we demonstrate that learning from longitudinal images, multi-phase images, and reports can overcome mask scarcity and advance multi-cancer detection and segmentation. Code: https://github.com/MrGiovanni/RT-Super

Mon 28 SeptComputer Vision and Pattern Recognition
The gist
Tumor outlines in scans are hard to get because drawing them takes a lot of work. The authors designed an AI system that learns to find tumors using reports and multiple images taken over time, instead of relying on these outlines. The AI first uses reports and scans from the same patient over time to create tumor masks. Then, it trains another AI to find tumors from a single scan alone. This method works well even on tumors without public data masks and can help detect multiple cancers.
Open → 2609.35637v1

Axon synthesis improves 3d segmentation in brain microscopy images

AxonSynth: Domain-Randomized Synthetic Data for Zero-Shot 3D Axon Segmentation in Light-Sheet Microscopy

Abstract: Accurate segmentation of axons in 3D microscopy data is important for analyzing white-matter organization, but dense ground truth labels are expensive to obtain. Existing supervised axon segmentation methods rely on target-domain annotations and can be brittle when tissue type, species, modality, or acquisition conditions change. We present AxonSynth, a domain-randomized synthetic-data framework for training 3D axon segmentation models without manually annotated real training volumes. AxonSynth generates dense synthetic axon labels with orientation priors that reflect realistic fiber configurations and renders them with randomized density, contrast, bias fields, blur, and noise. A three-class 3D U-Net is trained to predict background, axon sheath and intra-axonal space. We evaluate zero-shot transfer on 10 held-out light-sheet microscopy (LSM) patches from macaque and human brain samples labeled with one of three axonal markers, comparing against calibrated thresholding and Frangi filtering using overlap, corrected detection, false-positive, and topology metrics. On macaque samples, AxonSynth achieved the best corrected Dice and corrected precision (0.826 and 0.851), compared with 0.765 and 0.754 for thresholding and 0.685 and 0.762 for Frangi. On human samples, corrected Dice was comparable to thresholding (0.857 vs. 0.868), while component-count error decreased from 22,504 to 3,377. Across all held-out patches, AxonSynth reduced component-count error in 10/10 patches and Euler-characteristic error in 8/10. These results show that synthetic-label domain randomization can reduce dependence on manual axon annotation while supporting synthetic-to-real 3D segmentation.

Fri 25 SeptComputer Vision and Pattern Recognition
The gist
Identifying tiny nerve fibers called axons in 3D brain images helps scientists understand how the brain is connected, but labeling these fibers manually is very hard and time-consuming. The authors developed AxonSynth, a method that creates realistic fake 3D axon images with varied features to train computer models without needing manual labels. Their approach works well on real images from different species and stains, matching or beating traditional methods while reducing errors in detecting individual axons. This shows synthetic data can help computers learn to identify axons without expensive manual work.
Open → 2609.31431v1

NnFoundation models improve 3D radiology AI across many tasks

nnFoundation: 3D Foundation Models for Radiology

Abstract: Radiological artificial intelligence has advanced rapidly, yet most systems remain narrowly task-specific, data-intensive, and fragile under domain shift. Foundation models promise more transferable and data-efficient solutions, but existing approaches are limited in scale, evaluated narrowly, and often assume that a single pretrained model can support diverse downstream tasks. Here we present nnFoundation, complementary convolutional and transformer-based 3D radiological foundation models. Developed within the Human Radiome Project (THRP), nnFoundation is trained on 2.1 million CT, MRI, and PET image volumes from 125 institutional and public datasets. We evaluate them across 108 tasks spanning segmentation, detection, classification, report generation, and image retrieval, including evaluations under domain shift, by external partners and in low-data and low-compute regimes. Across all task types, our convolution- and transformer-based nnFoundation models consistently outperform both prior 3D foundation models and training from scratch, establishing state-of-the-art performance for radiological imaging. However, performance follows a consistent task-dependent structure: the convolutional nnFoundation model dominates spatially localized tasks, whereas the transformer-based nnFoundation model excels in tasks requiring global semantic reasoning and in frozen-feature settings. Dynamically aligning the foundation model topology with the dataset characteristics post-hoc further improves transfer across heterogeneous 3D settings. These results show that transferable 3D radiological performance is governed not by a single universal model, but by the interplay of scalable pretraining, complementary architectures, and dataset-aware adaptation. We release nnFoundation models integrated into nnU-Net and nnDetection, enabling immediate application across established radiology workflows.

Tue 22 SeptComputer Vision and Pattern Recognition
The gist
Radiology AI tools often work well only for specific tasks and struggle with new types of medical images. The authors created nnFoundation, a set of large-scale 3D models trained on millions of CT, MRI, and PET scans to work better across many radiology tasks like finding, segmenting, and describing images. They show that different model types excel at different tasks, and combining them improves accuracy and flexibility. Their models perform better than older methods, even when data or computing resources are limited.
Open → 2609.26924v1

Prototype purification improves multi disease detection in chest x rays

Purification and Regulation: Comorbidity-Aware Multi-Label Few-Shot Learning for Medical Image Classification

Abstract: Multi-label few-shot learning (MLFSL) remains a significant challenge in medical image analysis (MIA). Current metric-based meta-learning methods face two critical limitations in MIA. First, conventional prototype generation often entangles irrelevant disease information, leading to contaminated prototypes and degraded performance. Second, prior studies typically enforce inter-class separability in embedding space, largely neglecting the inherent correlations among diseases. To overcome these challenges, we propose Prototype Purification and Regulation (PPR), a novel MLFSL framework for MIA. PPR first performs prototype purification by leveraging sample-level comorbidity scores to emphasize disease-specific features, producing purified prototypes that better characterize each disease. Building upon these purified prototypes, PPR further addresses the underexplored problem of inter-class prototype distance in MIA by incorporating disease-level comorbidity statistics to adaptively regulate inter-class similarity, forming a comorbidity-aware embedding space. Overall, PPR sequentially enables the model to capture pure disease features and inter-class relationships for reliable MLFSL in MIA. Extensive experiments across four chest X-ray benchmark datasets, including cross-domain evaluation, show that PPR consistently outperforms state-of-the-art methods, significantly improving disease detection while demonstrating robust generalization and clinical applicability.

Fri 18 SeptComputer Vision and Pattern RecognitionMachine Learning
The gist
Identifying multiple diseases from medical images is hard, especially when only a few examples of each disease are available. The paper presents a new method called Prototype Purification and Regulation (PPR) that cleans up disease representations and respects how diseases often appear together. This helps the model recognize diseases more accurately and reliably. The authors tested PPR on chest X-ray tasks, showing it works better than existing methods even when applied to different datasets.
Open → 2609.21541v1

Self-supervised learning strategies evaluated for lung ultrasound imaging

Which Pretext Task Transfers? Self-Supervised Pretraining Objectives for Lung Ultrasound

Abstract: Self-supervised learning (SSL) can reduce the need for labelled medical images, but the choice of pretext objective remains unclear for lung ultrasound (LUS). Contrastive learning, masked reconstruction, and joint-embedding predictive architectures (JEPA) differ in the space in which their targets are defined, yet existing ultrasound studies compare them under different corpora, backbones, and evaluation protocols. We compare these three objective families using the same encoder backbone, pretraining corpus, optimisation schedule, and frozen-evaluation protocol. Encoders are pretrained on COVID-BLUeS LUS videos and evaluated with linear, $k$NN, and attentive probes at 5\%, 10\%, 50\%, and 100\% label budgets. Evaluation is performed on POCUS using patient-level five-fold cross-validation and on the independently acquired Mendeley-Uganda dataset, which is excluded from both pretraining and probe fitting. At the full label budget under linear probing, VideoMAE and V-JEPA achieve $66.5 \pm 13.1$ and $65.4 \pm 11.7$ balanced accuracy on POCUS, while MoCo achieves $42.1 \pm 1.2$. On Mendeley-Uganda, the ranking reverses: MoCo performs best at $62.7 \pm 1.0$, followed by VideoMAE at $53.8 \pm 2.8$, while V-JEPA falls near chance at $35.1 \pm 4.9$. These results show that POCUS probe accuracy alone does not identify the objective that transfers best across datasets. We also outline planned representation-level analyses to examine this reversal. Code is publicly available at https://github.com/moeinheidari7829/LUSVideoSSL.

Tue 15 SeptComputer Vision and Pattern Recognition
The gist
Getting labeled lung ultrasound images is hard, so the authors studied ways to teach computers to learn from videos without labels first, then use a few labeled images to recognize lung features. They tested three popular methods on the same datasets and found that which method works best depends on the specific set of lung ultrasound images being used. This suggests that success on one dataset doesn't guarantee the method will do well on others. The authors plan further analysis to understand these differences better and shared their code openly.
Open → 2609.16551v1