Selected work

Representative Papers

MedPlex: Deep Vision-Language Co-Adaptation for Clinically Grounded Medical Segmentation

Aug 13, 2026

This study addresses the underutilization of textual knowledge and the limitations of late-stage language guidance in medical image segmentation by proposing an end-to-end vision-language framework. Innovatively transforming text from a posterior condition into continuous structured supervision, the method employs bidirectional fusion and class/region-level multi-granularity concept alignment to integrate clinical text throughout the encoding process. This enables joint evolution of visual-linguistic representations for precise segmentation guidance. The proposed approach achieves state-of-the-art performance across CT/MR multi-organ, cardiac, and tumor segmentation benchmarks while effectively supporting real-world free-text supervision. Consequently, this work significantly enhances both clinical applicability and model generalizability by establishing text as a persistent supervisory signal rather than a mere post-hoc constraint.

0 citationsRead paper

A Neighborhood Attention Transformer Network for Enhanced 3D Segmentation of the Left Anterior Descending Artery

Aug 12, 2026

This study addresses the challenging segmentation of the left anterior descending (LAD) artery in non-contrast, free-breathing 3D CT scans, where its small size, low soft-tissue contrast, and high anatomical variability hinder accurate delineation. To tackle this, the authors propose NA-UNETR, a 3D Transformer-based architecture that integrates neighborhood attention and dilated neighborhood attention modules to jointly capture local details and global context. The model employs LoRA for parameter-efficient transfer learning and introduces a novel homoscedastic uncertainty-weighted composite loss combining Dice-Focal and Hausdorff distance to dynamically refine boundary accuracy. Evaluated under extremely limited annotation, NA-UNETR achieves a Dice score of 45.64% on in-house data and significantly improves to 79.49% on the ImageCAS dataset, outperforming nnU-Net and Swin UNETR in boundary metrics (HD95 = 38.16 mm, ASD = 10.01 mm).

0 citationsRead paper

Model Confidence Under Answer-Preserving Attacks: An Informativeness-Manipulability Frontier

Aug 06, 2026

Current vision-language systems rely on confidence scores for answer filtering, yet their robustness under adversarial attacks that strictly preserve the exact byte sequence of generated answers remains unclear. This work proposes a white-box image perturbation method that systematically manipulates model confidence while keeping the generated answer completely unchanged. We demonstrate for the first time that confidence can be significantly suppressed even under this stringent constraint, and we introduce a verifiable robustness criterion based on uniform-magnitude certificates along with an analysis of its failure mechanisms. Experiments show that the proposed attack is effective across all tested models and benchmarks, reducing confidence below the accuracy baseline; in gating simulations, it leads to an erroneous answer acceptance rate as high as 84.8%, indicating that confidence-based gating can actually degrade overall system performance.

0 citationsRead paper

FocusSDF: Boundary-Aware Learning for Medical Image Segmentation via Signed Distance Supervision

Nov 14, 2025

Medical image segmentation often suffers from blurred boundaries due to insufficient encoding of boundary information. To address this, we propose FocusSDF—a novel adaptive boundary-aware loss function based on the Signed Distance Function (SDF)—which dynamically weights pixels in the neighborhood of lesion/organ boundaries to explicitly guide the network to focus on boundary regions. Its key innovation lies in being the first to incorporate SDF directly into the supervision mechanism, enabling boundary-sensitive end-to-end optimization while maintaining compatibility with multimodal imaging and mainstream segmentation architectures. Evaluated on public datasets covering brain aneurysms, stroke, liver, and breast tumors, FocusSDF consistently outperforms five state-of-the-art distance-transform-based losses. Significant improvements are observed in boundary-specific metrics—including Hausdorff distance and average surface distance—demonstrating its effectiveness in fine-grained boundary modeling.

0 citationsRead paper

Recent publications

Latest Papers

MedPlex: Deep Vision-Language Co-Adaptation for Clinically Grounded Medical Segmentation

Aug 13, 2026

This study addresses the underutilization of textual knowledge and the limitations of late-stage language guidance in medical image segmentation by proposing an end-to-end vision-language framework. Innovatively transforming text from a posterior condition into continuous structured supervision, the method employs bidirectional fusion and class/region-level multi-granularity concept alignment to integrate clinical text throughout the encoding process. This enables joint evolution of visual-linguistic representations for precise segmentation guidance. The proposed approach achieves state-of-the-art performance across CT/MR multi-organ, cardiac, and tumor segmentation benchmarks while effectively supporting real-world free-text supervision. Consequently, this work significantly enhances both clinical applicability and model generalizability by establishing text as a persistent supervisory signal rather than a mere post-hoc constraint.

0 citationsRead paper

A Neighborhood Attention Transformer Network for Enhanced 3D Segmentation of the Left Anterior Descending Artery

Aug 12, 2026

This study addresses the challenging segmentation of the left anterior descending (LAD) artery in non-contrast, free-breathing 3D CT scans, where its small size, low soft-tissue contrast, and high anatomical variability hinder accurate delineation. To tackle this, the authors propose NA-UNETR, a 3D Transformer-based architecture that integrates neighborhood attention and dilated neighborhood attention modules to jointly capture local details and global context. The model employs LoRA for parameter-efficient transfer learning and introduces a novel homoscedastic uncertainty-weighted composite loss combining Dice-Focal and Hausdorff distance to dynamically refine boundary accuracy. Evaluated under extremely limited annotation, NA-UNETR achieves a Dice score of 45.64% on in-house data and significantly improves to 79.49% on the ImageCAS dataset, outperforming nnU-Net and Swin UNETR in boundary metrics (HD95 = 38.16 mm, ASD = 10.01 mm).

0 citationsRead paper

Model Confidence Under Answer-Preserving Attacks: An Informativeness-Manipulability Frontier

Aug 06, 2026

Current vision-language systems rely on confidence scores for answer filtering, yet their robustness under adversarial attacks that strictly preserve the exact byte sequence of generated answers remains unclear. This work proposes a white-box image perturbation method that systematically manipulates model confidence while keeping the generated answer completely unchanged. We demonstrate for the first time that confidence can be significantly suppressed even under this stringent constraint, and we introduce a verifiable robustness criterion based on uniform-magnitude certificates along with an analysis of its failure mechanisms. Experiments show that the proposed attack is effective across all tested models and benchmarks, reducing confidence below the accuracy baseline; in gating simulations, it leads to an erroneous answer acceptance rate as high as 84.8%, indicating that confidence-based gating can actually degrade overall system performance.

0 citationsRead paper

FocusSDF: Boundary-Aware Learning for Medical Image Segmentation via Signed Distance Supervision

Nov 14, 2025

Medical image segmentation often suffers from blurred boundaries due to insufficient encoding of boundary information. To address this, we propose FocusSDF—a novel adaptive boundary-aware loss function based on the Signed Distance Function (SDF)—which dynamically weights pixels in the neighborhood of lesion/organ boundaries to explicitly guide the network to focus on boundary regions. Its key innovation lies in being the first to incorporate SDF directly into the supervision mechanism, enabling boundary-sensitive end-to-end optimization while maintaining compatibility with multimodal imaging and mainstream segmentation architectures. Evaluated on public datasets covering brain aneurysms, stroke, liver, and breast tumors, FocusSDF consistently outperforms five state-of-the-art distance-transform-based losses. Significant improvements are observed in boundary-specific metrics—including Hausdorff distance and average surface distance—demonstrating its effectiveness in fine-grained boundary modeling.

0 citationsRead paper