arXiv subjects

Zongwei Zhou

Publications and source records attributed to Zongwei Zhou.

At least 19 recordsLinked to original sources

arXiv2026

LeCor: Learning to Be Corrected by Meta-Learned Test-Time Training for Interactive 3D Lung-Tumour Segmentation

Delineating lung tumours on computed tomography (CT) takes a considerable share of the time spent on radiotherapy planning, and a contour proposed by a model can be refined interactively by the clinician. Promptable foundation models such as SAM 3 support this workflow by writing each correction into a session memory that conditions the remaining slices, while the model weights stay fixed. On 690 test cases from five public CT cohorts, fine-tuning SAM 3 on lung tumours raises the Dice obtained from a single point prompt from 0.298 to 0.757, and seven rounds of corrections raise it further to 0.765, but under memory conditioning alone the accuracy on slices the annotator has not touched stops improving after six rounds. We therefore treat each correction as a training signal and propose LeCor, which performs test-time training on a small set of case adapters that are reset for every case and meta-learned such that a single gradient step driven by a click improves the slices that were not clicked. On the 133 test cases that span at least eight slices, LeCor raises the Dice reached after seven correction rounds from 0.787 with the fine-tuned model to 0.827, reduces the number of cases that never reach a Dice of 0.80 from 47 to 27, and reaches in three correction rounds the accuracy that the fine-tuned model attains in seven.

cs.CV↗

arXiv2026

Report Supervision

Segmentation models can surpass radiologists, classification models, and vision-language models in tumor detection. Importantly, segmentation models outline tumors, allowing radiologists to better verify and trust the AI output. Their main limitation is the scarcity of tumor masks: creating one 3D tumor mask takes up to 30 minutes, so most public CT datasets contain only a few hundred masks, and even the largest private datasets contain only a couple of thousand. Tumor masks are not produced in clinical routine, but radiology reports are. Public datasets contain tens of thousands of CT-Report pairs, and hospitals contain hundreds of thousands. These reports describe tumors in detail, providing large-scale, informative training data. Here, we introduce Report Supervision (R-Super), a training framework that uses reports to directly supervise and improve tumor segmentation. R-Super introduces new loss functions that teach segmentation models to segment tumors that match report descriptions of tumor count, sizes, and locations. Reports are only used for training. We evaluated R-Super on kidney and pancreatic tumor segmentation, exploring diverse training data sizes, up to 41,418 CT-Report plus 3,488 pancreatic tumor CT-Mask pairs. On external validation, R-Super increased tumor detection F1-Score and segmentation DSC by up to +15% with respect to mask-only training. It also surpassed alternative methods such as CLIP and multi-task learning. Leveraging numerous readily available reports to supplement scarce masks, R-Super strongly improves AI performance when very few training masks are available (e.g., 50), and when many masks are available (e.g., 3,488), unlocking scale in tumor segmentation.

cs.CV↗

arXiv2026

Gaussian Meta-Space Augmentation for Stacking Ensembles in Multimodal IPMN Risk Stratification

Pancreatic cancer is among the most lethal malignancies; risk stratification of intraductal papillary mucinous neoplasms (IPMNs) offers a crucial opportunity for early intervention but typically requires invasive tissue biopsy. Dominant vision-based approaches, including radiomics and deep learning, provide promising but initially separate discrimination opportunities. Similarly, multisequence MRI (T1W/T2W) and anatomically decomposed (head, body and tail) analysis of the pancreas provide additional and potentially complementary signals. Effective fusion of this information is crucial in ordinal IPMN dysplasia risk prediction and can be accomplished via a meticulously regularized and calibrated ensemble stacking combiner. We present cUPMI, a class-conditional Gaussian augmentation of a combiner's log-probability meta-features, and test it on various prediction paradigms. In our multi-center analysis, we find cUPMI adds limited value to properly regularized L2-logistic binary classification stacks, but consistently regularizes higher-capacity tree combiners in the binary and radiomics-only setting (RF +0.015 and XGBoost +0.024 binary AUC, positive in all seeds). Its cleanest ordinal benefit appears for XGBoost on an 8-stream radiomics task (3-class no < low < high, +0.022 QWK in all seeds). Separately, fold-locked fusion of radiomics and 2.5D CNN streams yields the strongest overall model, an RF stack reaching QWK 0.595 (95% CI [0.54, 0.64]) and binary AUC 0.839, surpassing radiomics, 2.5D ResNet, and 3D DenseNet-121 baselines.

cs.CV↗

arXiv2026

MRI2Rep: Autoregressive Structured Report Generation for 3D Liver MRI

Manual reporting of 3D MRI studies is time-consuming, yet end-to-end structured report generation for 3D liver MRI remains underexplored due to volumetric complexity and scarce paired data. We propose MRI2Rep, an autoregressive framework for liver MRI report generation. From 3,929 real-world MRI-report pairs acquired over a 10-year single-institution cohort, a Report-to-Label Canonicalization (RLC) module converts free-text reports into structured, closed-vocabulary diagnostic sequences without lesion-level annotations. On a held-out test set, MRI2Rep achieves 76.0% case-level sensitivity, 29.4% lesion-level F1, compared with no more than 8.3% for adapted medical vision-language baselines, and 82.4% liver-level accuracy. In a blinded reader study, two radiologists rated 75% and 70% of AI-generated reports as clinically acceptable, compared with 95% and 100% for original reports. Our automated LLM-based judge, LLM-Eval, rated 61.8% of AI-generated reports as acceptable, applying a stricter standard and supporting its use as a conservative proxy. To our knowledge, this is the first end-to-end LI-RADS-structured reporting system for 3D liver MRI.

cs.CV↗

arXiv2026

BenchX: Benchmarking AI Models for Cancer Detection and Localization with Demographic and Protocol Biases

Artificial intelligence (AI) has achieved remarkable success in medical imaging, but it is widely recognized that these models often perform inconsistently across real-world clinical settings. Such inconsistencies occur when patient demographics and imaging protocols vary, for example, in detecting small tumors, analyzing scans from different contrast phases, or evaluating patients of different ages or sexes. To quantify these inconsistencies, we develop a large-scale, open benchmark of 85,355 CT scans that systematically evaluates 12 tumor-detection AI models across tumor size, location, patient subgroup, and imaging protocol. We leverage large language models (LLMs) to extract and organize subgroup information from clinical data, which makes the analysis both scalable and reproducible. Our benchmark reveals that current state-of-the-art AI models, optimized for average accuracy, perform poorly in rare or underrepresented subgroups, such as young, female African Americans. However, collecting sufficient annotated data for these rare cases is often impractical. The benchmark provides a foundation for building more reliable and robust AI models for tumor detection and highlighting the need for rigorous, subgroup-level evaluation in medical imaging and computer vision. Datasets, code

cs.CV↗

arXiv2026

Foundation VAEs for 3D CT Reconstruction, Augmentation, and Generation

Variational autoencoders (VAEs) compress high resolution CT volumes into compact latents while preserving clinically relevant structure. However, training CT-specific VAEs from scratch or heavily fine-tuning them incurs substantial computational and engineering cost, and often degrades under heterogeneous scanners, protocols, and diseases. This paper makes a progressive stride toward training-free medical VAEs by leveraging a critical observation: a single Foundation VAE, pretrained at scale on natural images and videos, can serve as a unified interface for CT Reconstruction, Augmentation, and Generation. With both encoder and decoder frozen, the Foundation VAE reconstructs CT volumes with preserved anatomy while suppressing acquisition noise; training segmentation models on these reconstructions improves surface accuracy by 3.9% NSD on average for pancreatic tumor and lung tumor. Within the same Foundation VAE latent space, a conditional latent diffusion model achieves 3.9% lower average FVD with 36.2% higher CT CLIP score, and improves multi-disease generation faithfulness across 18 types by 2.76% AUC. These results demonstrate Foundation VAEs as a practical interface for scalable CT representation reuse and faithful CT generation. Our code and demo are available at https://github.com/qic999/Foundation-VAE.

cs.CV↗

arXiv2026

Beyond Masks: The Case for Medical Image Parsing

Medical imaging research has spent a decade getting very good at one thing: producing per-voxel masks. Masks tell us size, volume, and location, and a decade of clinical infrastructure rests on those outputs. Yet the report a radiologist writes contains almost nothing a mask can express. We argue that medical imaging research should adopt medical image parsing as its central output: a structured representation in which entities, attributes, and relationships are emitted together and mutually consistent. Entities are the named structures and findings, present or absent. Attributes describe those entities, capturing things like margin regularity, enhancement pattern, or severity grade. Relationships connect them, naming where one structure sits relative to another, what abuts what, and what has changed since the prior scan. A good parse satisfies three properties, in order: (1) decision (the parse names the right things in the current image), (2) reconstruction (its content is rich enough to regenerate that image), and (3) prediction (its content is rich enough to forecast how the patient state will evolve). Quantitative measurements are derived from this content; they are not predicted alongside it. To test how close the field is to producing such an output, we audit eleven representative systems against the three parsing primitives plus closure. None emits a well-formed parse. Entities are largely solved. Attributes, relationships, and closure remain near-empty. The path forward is not a new architecture. It is a commitment to a richer output, and to training signals that reward it. Segmentation taught models to measure. Parsing asks them to explain.

cs.CV↗

arXiv2026

RadThinking: A Dataset for Longitudinal Clinical Reasoning in Radiology

Cancer screening is a reasoning task. A radiologist observes findings, compares them to prior scans, integrates clinical context, and reaches a diagnostic conclusion confirmed by pathology. We present RadThinking, a Visual Question Answering (VQA) dataset that makes this reasoning explicit and trainable. RadThinking releases VQA pairs at three difficulty tiers. Foundation VQAs are atomic perception questions. Single-step reasoning VQAs apply one clinical rule. Compositional VQAs require multi-step chain-of-thought to reach a guideline category such as LI-RADS-5. For every compositional VQA, we release the chain of foundation VQAs that solves it. The chain follows the rules of the governing clinical reporting standard. The dataset spans 20,362 CT scans from 9,131 patients across 43 cancer groups, plus 2,077 verified healthy controls with >1-year follow-up. To our knowledge, RadThinking is the first cancer-screening VQA corpus that stratifies questions by reasoning depth and grounds compositions in clinical reporting standards. The foundation tier supplies atomic perception supervision. The compositional tier supplies chain-of-thought data and verifiable rewards for reinforcement-learning recipes such as DeepSeek-R1 and OpenAI o1. RadThinking enables systematic training and evaluation of whether AI systems can reason about cancer, not merely detect it.

cs.CV↗

arXiv2026

DeepTumorVQA: A Hierarchical 3D CT Benchmark for Stage-Wise Evaluation of Medical VLMs and Tool-Augmented Agents

Medical vision-language models (VLMs) and AI agents have made significant progress in learning to analyze and reason about clinical images. However, existing medical visual question answering (VQA) benchmarks collapse model capabilities into a single accuracy score, obscuring where and why models fail. We propose DeepTumorVQA, a hierarchical benchmark that follows the multi-stage evidence chain in tumor diagnosis and decomposes 3D CT reasoning into four stages: recognition, measurement, visual reasoning, and medical reasoning. Higher-level questions remain independently scorable, while their ground-truth evidence chains are defined over lower-level primitives. The benchmark contains 476K questions across 42 clinical subtypes on 9,262 3D CT volumes. In addition to a direct reasoning mode for VLMs, DeepTumorVQA provides tool-interaction environments for agent evaluation, where a model can call external tools, including segmentation models, measurement programs, and medical knowledge modules, before answering the question. Evaluating over 30 model configurations, we find that reliable quantitative measurement is the primary bottleneck, making later-stage visual and medical reasoning harder for VLMs, while tool augmentation substantially mitigates this issue. When tools are available, leveraging medical knowledge and tools to reason about medical images becomes a new challenge. We further show that ground-truth step-by-step tool-use traces from DeepTumorVQA can supervise agents and reduce tool-use and reasoning failures. This stage-wise progression from recognition to measurement to visual and medical reasoning provides a concrete roadmap for future medical VLM and AI agent studies. All data and code are released at https://github.com/Schuture/DeepTumorVQA.

cs.CV↗

arXiv2026

CrossPan: A Comprehensive Benchmark for Cross-Sequence Pancreas MRI Segmentation and Generalization

Automatic pancreas segmentation is fundamental to abdominal MRI analysis, yet deep learning models trained on one MRI sequence often fail catastrophically when applied to another-a challenge that has received little systematic investigation. We introduce CrossPan, a multi-institutional benchmark comprising 1,386 3D scans across three routinely acquired sequences (T1-weighted, T2-weighted, and Out-of-Phase) from eight centers. Our experiments reveal three key findings. First, cross-sequence domain shifts are far more severe than cross-center variability: models achieving Dice scores above 0.85 in-domain collapse to near-zero (<0.02) when transferred across sequences. Second, state-of-the-art domain generalization methods provide negligible benefit under these physics-driven contrast inversions, whereas foundation models like MedSAM2 maintain moderate zero-shot performance through contrast-invariant shape priors. Third, semi-supervised learning offers gains only under stable intensity distributions and becomes unstable on sequences with high intra-organ variability. These results establish cross-sequence generalization-not model architecture or center diversity-as the primary barrier to clinically deployable pancreas MRI segmentation. Dataset and code are available at https://crosspan.netlify.app/.

cs.CV↗

arXiv2026

Distilling Photon-Counting CT into Routine Chest CT through Clinically Validated Degradation Modeling

Photon-counting CT (PCCT) provides superior image quality with higher spatial resolution and lower noise compared to conventional energy-integrating CT (EICT), but its limited clinical availability restricts large-scale research and clinical deployment. To bridge this gap, we propose SUMI, a simulated degradation-to-enhancement method that learns to reverse realistic acquisition artifacts in low-quality EICT by leveraging high-quality PCCT as reference. Our central insight is to explicitly model realistic acquisition degradations, transforming PCCT into clinically plausible lower-quality counterparts and learning to invert this process. The simulated degradations were validated for clinical realism by board-certified radiologists, enabling faithful supervision without requiring paired acquisitions at scale. As outcomes of this technical contribution, we: (1) train a latent diffusion model on 1,046 PCCTs, using an autoencoder first pre-trained on both these PCCTs and 405,379 EICTs from 145 hospitals to extract general CT latent features that we release for reuse in other generative medical imaging tasks; (2) construct a large-scale dataset of over 17,316 publicly available EICTs enhanced to PCCT-like quality, with radiologist-validated voxel-wise annotations of airway trees, arteries, veins, lungs, and lobes; and (3) demonstrate substantial improvements: across external data, SUMI outperforms state-of-the-art image translation methods by 15% in SSIM and 20% in PSNR, improves radiologist-rated clinical utility in reader studies, and enhances downstream top-ranking lesion detection performance, increasing sensitivity by up to 15% and F1 score by up to 10%. Our results suggest that emerging imaging advances can be systematically distilled into routine EICT using limited high-quality scans as reference.

cs.CV↗

arXiv2026

Project Imaging-X: A Survey of 1000+ Open-Access Medical Imaging Datasets for Foundation Model Development

Foundation models have demonstrated remarkable success across diverse domains and tasks, primarily due to the thrive of large-scale, diverse, and high-quality datasets. However, in the field of medical imaging, the curation and assembling of such medical datasets are highly challenging due to the reliance on clinical expertise and strict ethical and privacy constraints, resulting in a scarcity of large-scale unified medical datasets and hindering the development of powerful medical foundation models. In this work, we present the largest survey to date of medical image datasets, covering over 1,000 open-access datasets with a systematic catalog of their modalities, tasks, anatomies, annotations, limitations, and potential for integration. Our analysis exposes a landscape that is modest in scale, fragmented across narrowly scoped tasks, and unevenly distributed across organs and modalities, which in turn limits the utility of existing medical image datasets for developing versatile and robust medical foundation models. To turn fragmentation into scale, we propose a metadata-driven fusion paradigm (MDFP) that integrates public datasets with shared modalities or tasks, thereby transforming multiple small data silos into larger, more coherent resources. Building on MDFP, we release an interactive discovery portal that enables end-to-end, automated medical image dataset integration, and compile all surveyed datasets into a unified, structured table that clearly summarizes their key characteristics and provides reference links, offering the community an accessible and comprehensive repository. By charting the current terrain and offering a principled path to dataset consolidation, our survey provides a practical roadmap for scaling medical imaging corpora, supporting faster data discovery, more principled dataset creation, and more capable medical foundation models.

cs.CV↗

arXiv2026

Meissa: Multi-modal Medical Agentic Intelligence

Multi-modal large language models (MM-LLMs) have shown strong performance in medical image understanding and clinical reasoning. Recent medical agent systems extend them with tool use and multi-agent collaboration, enabling complex decision-making. However, these systems rely almost entirely on frontier models (e.g., GPT), whose API-based deployment incurs high cost, high latency, and privacy risks that conflict with on-premise clinical requirements. We present Meissa, a lightweight 4B-parameter medical MM-LLM that brings agentic capability offline. Instead of imitating static answers, Meissa learns both when to engage external interaction (strategy selection) and how to execute multi-step interaction (strategy execution) by distilling structured trajectories from frontier models. Specifically, we propose: (1) Unified trajectory modeling: trajectories (reasoning and action traces) are represented within a single state-action-observation formalism, allowing one model to generalize across heterogeneous medical environments. (2) Three-tier stratified supervision: the model's own errors trigger progressive escalation from direct reasoning to tool-augmented and multi-agent interaction, explicitly learning difficulty-aware strategy selection. (3) Prospective-retrospective supervision: pairing exploratory forward traces with hindsight-rationalized execution traces enables stable learning of effective interaction policies. Trained on 40K curated trajectories, Meissa matches or exceeds proprietary frontier agents in 10 of 16 evaluation settings across 13 medical benchmarks spanning radiology, pathology, and clinical reasoning. Using over 25x fewer parameters than typical frontier models like Gemini-3, Meissa operates fully offline with 22x lower end-to-end latency compared to API-based deployment. Data, models, and environments are released at https://github.com/Schuture/Meissa.

cs.AI↗

arXiv2026

RayD3D: Distilling Depth Knowledge Along the Ray for Robust Multi-View 3D Object Detection

Multi-view 3D detection with bird's eye view (BEV) is crucial for autonomous driving and robotics, but its robustness in real-world is limited as it struggles to predict accurate depth values. A mainstream solution, cross-modal distillation, transfers depth information from LiDAR to camera models but also unintentionally transfers depth-irrelevant information (e.g. LiDAR density). To mitigate this issue, we propose RayD3D, which transfers crucial depth knowledge along the ray: a line projecting from the camera to true location of an object. It is based on the fundamental imaging principle that predicted location of this object can only vary along this ray, which is finally determined by predicted depth value. Therefore, distilling along the ray enables more effective depth information transfer. More specifically, we design two ray-based distillation modules. Ray-based Contrastive Distillation (RCD) incorporates contrastive learning into distillation by sampling along the ray to learn how LiDAR accurately locates objects. Ray-based Weighted Distillation (RWD) adaptively adjusts distillation weight based on the ray to minimize the interference of depth-irrelevant information in LiDAR. For validation, we widely apply RayD3D into three representative types of BEV-based models, including BEVDet, BEVDepth4D, and BEVFormer. Our method is trained on clean NuScenes, and tested on both clean NuScenes and RoboBEV with a variety types of data corruptions. Our method significantly improves the robustness of all the three base models in all scenarios without increasing inference costs, and achieves the best when compared to recently released multi-view and distillation models.

cs.CV↗

arXiv2026

Early and Prediagnostic Detection of Pancreatic Cancer from Computed Tomography

Pancreatic ductal adenocarcinoma (PDAC), one of the deadliest solid malignancies, is often detected at a late and inoperable stage. Retrospective reviews of prediagnostic CT scans, when conducted by expert radiologists aware that the patient later developed PDAC, frequently reveal lesions that were previously overlooked. To help detecting these lesions earlier, we developed an automated system named ePAI (early Pancreatic cancer detection with Artificial Intelligence). It was trained on data from 1,598 patients from a single medical center. In the internal test involving 1,009 patients, ePAI achieved an area under the receiver operating characteristic curve (AUC) of 0.939-0.999, a sensitivity of 95.3%, and a specificity of 98.7% for detecting small PDAC less than 2 cm in diameter, precisely localizing PDAC as small as 2 mm. In an external test involving 7,158 patients across 6 centers, ePAI achieved an AUC of 0.918-0.945, a sensitivity of 91.5%, and a specificity of 88.0%, precisely localizing PDAC as small as 5 mm. Importantly, ePAI detected PDACs on prediagnostic CT scans obtained 3 to 36 months before clinical diagnosis that had originally been overlooked by radiologists. It successfully detected and localized PDACs in 75 of 159 patients, with a median lead time of 347 days before clinical diagnosis. Our multi-reader study showed that ePAI significantly outperformed 30 board-certified radiologists by 50.3% (P < 0.05) in sensitivity while maintaining a comparable specificity of 95.4% in detecting PDACs early and prediagnostic. These findings suggest its potential of ePAI as an assistive tool to improve early detection of pancreatic cancer.

cs.CV↗

arXiv2026

Large-Scale Label Quality Assessment for Medical Segmentation via a Vision-Language Judge and Synthetic Data

Large-scale medical segmentation datasets often combine manual and pseudo-labels of uneven quality, which can compromise training and evaluation. Low-quality labels may hamper performance and make the model training less robust. To address this issue, we propose SegAE (Segmentation Assessment Engine), a lightweight vision-language model (VLM) that automatically predicts label quality across 142 anatomical structures. Trained on over four million image-label pairs with quality scores, SegAE achieves a high correlation coefficient of 0.902 with ground-truth Dice similarity and evaluates a 3D mask in 0.06s. SegAE shows several practical benefits: (I) Our analysis reveals widespread low-quality labeling across public datasets; (II) SegAE improves data efficiency and training performance in active and semi-supervised learning, reducing dataset annotation cost by one-third and quality-checking time by 70% per label. This tool provides a simple and effective solution for quality control in large-scale medical segmentation datasets. The dataset, model weights, and codes are released at https://github.com/Schuture/SegAE.

cs.CV↗

arXiv2025

Auditing Significance, Metric Choice, and Demographic Fairness in Medical AI Challenges

Open challenges have become the de facto standard for comparative ranking of medical AI methods. Despite their importance, medical AI leaderboards exhibit three persistent limitations: (1) score gaps are rarely tested for statistical significance, so rank stability is unknown; (2) single averaged metrics are applied to every organ, hiding clinically important boundary errors; (3) performance across intersecting demographics is seldom reported, masking fairness and equity gaps. We introduce RankInsight, an open-source toolkit that seeks to address these limitations. RankInsight (1) computes pair-wise significance maps that show the nnU-Net family outperforms Vision-Language and MONAI submissions with high statistical certainty; (2) recomputes leaderboards with organ-appropriate metrics, reversing the order of the top four models when Dice is replaced by NSD for tubular structures; and (3) audits intersectional fairness, revealing that more than half of the MONAI-based entries have the largest gender-race discrepancy on our proprietary Johns Hopkins Hospital dataset. The RankInsight toolkit is publicly released and can be directly applied to past, ongoing, and future challenges. It enables organizers and participants to publish rankings that are statistically sound, clinically meaningful, and demographically fair.

cs.CV↗

arXiv2025

Expectation-Maximization as the Engine of Scalable Medical Intelligence

Large, high-quality, annotated datasets are the foundation of medical AI research, but constructing even a small, moderate-quality, annotated dataset can take years of effort from multidisciplinary teams. Although active learning can prioritize what to annotate, scaling up still requires extensive manual efforts to revise the noisy annotations. We formulate this as a missing-data problem and develop ScaleMAI, a framework that unifies data annotation and model development co-evolution through an Expectation-Maximization (EM) process. In this iterative process, the AI model automatically identifies and corrects the mistakes in annotations (Expectation), while the refined annotated data retrain the model to improve accuracy (Maximization). In addition to the classical EM algorithm, ScaleMAI brings human experts into the loop to review annotations that cannot be adequately addressed by either Expectation or Maximization step (<5%). As a result, ScaleMAI progressively creates an annotated dataset of 47,315 CT scans (4.8x larger than the largest public dataset, PanTS) including 4,163,720 per-voxel annotations for benign/malignant tumors and 88 anatomical structures. ScaleMAI iteratively trains a model that exceeds human expert performance in tumor diagnosis (+7%), and outperforms models developed from smaller, moderate-quality datasets, with statistically significant gains in tumor detection (+10%) and segmentation (+14%) on two prestigious benchmarks.

cs.CV↗