Multi-Pass, Multi-View Blended Learning for High-Fidelity Volumetric CT Synthesis from Chest X-Rays
本文通过多通道多视角混合学习框架解决从单张2D胸透重建3D CT的问题，提高结构完整性和解剖细节。
0 citationsRead paper
Selected work
Representative Papers
A Two-Stage Deep Learning Framework for Segmentation of Ten Gastrointestinal Organs from Coronal MR Enterography
This study addresses the significant challenges in automatic segmentation of gastrointestinal organs from coronal MR enterography, which arise due to anatomical variability, low tissue contrast, and extreme class imbalance. To overcome these issues, the authors propose a coarse-to-fine, organ-aware two-stage deep learning framework. In the first stage, a DenseNet201-UNet++ architecture generates a coarse region-of-interest mask. The second stage employs a DenseNet121-SelfONN-UNet model trained on organ-specific image patches, augmented with a class-weighting strategy to mitigate imbalance. The proposed method achieves substantial performance gains, yielding an overall mean Dice similarity coefficient (mDSC) of 88.99% and mean intersection-over-union (mIoU) of 84.76%. Notably, segmentation accuracy for the appendix improves dramatically—from 6.76% to 85.76% in DSC—significantly outperforming existing baselines.
0 citationsRead paper
Tracing 3D Anatomy in 2D Strokes: A Multi-Stage Projection Driven Approach to Cervical Spine Fracture Identification
This study addresses the high computational complexity of detecting cervical spine fractures in 3D CT scans by proposing an efficient end-to-end framework based on multi-view 2D projections. The method reconstructs 3D anatomical structures using variance and energy projections, integrating YOLOv8 for localization, DenseNet121-Unet for segmentation, and a 2.5D temporal model for vertebra-level fracture identification. Interpretability is validated through saliency maps compared against expert annotations. The approach achieves strong performance with minimal computational overhead, yielding a 3D localization mIoU of 94.45%, a vertebra segmentation Dice score of 87.86%, and vertebra-level and patient-level F1 scores of 68.15% and 82.26%, respectively. Corresponding ROC-AUC values are 91.62% and 83.04%, demonstrating diagnostic accuracy comparable to that of radiology experts.
0 citationsRead paper
An Anatomy Aware Hybrid Deep Learning Framework for Lung Cancer Tumor Stage Classification
Lung cancer TNM staging relies on tumor size and spatial relationships with adjacent anatomical structures; however, existing end-to-end deep learning models lack interpretability and neglect critical anatomical context. To address this, we propose a hybrid deep learning framework integrating explicit anatomical priors: (1) a dedicated encoder-decoder network precisely segments the lungs, mediastinum, and tumor; (2) quantitative features—including maximum tumor diameter and shortest distances to key anatomical boundaries—are extracted from segmentation masks; and (3) clinical guideline rules are applied for T-stage classification. This is the first approach to explicitly embed clinically grounded anatomical context into the deep learning pipeline, overcoming the “black-box” limitation while ensuring both interpretability and clinical compliance. Evaluated on the Lung-PET-CT-Dx dataset, our method achieves 91.36% overall accuracy, with F1-scores of 0.93, 0.89, 0.96, and 0.90 for T1–T4 stages—significantly outperforming end-to-end baseline models.
0 citationsRead paper
CASR-Net: An Image Processing-focused Deep Learning-based Coronary Artery Segmentation and Refinement Network for X-ray Coronary Angiogram
This study addresses low segmentation accuracy, fragmented small vascular branches, and high false-positive rates in X-ray coronary angiography images. We propose a three-stage deep learning framework: (1) multi-channel preprocessing integrating CLAHE with an improved Ben Graham method to enhance contrast and suppress noise; (2) a backbone segmentation network leveraging a DenseNet121 encoder and a Self-ONN decoder to strengthen feature representation; and (3) a contour refinement module embedded to improve vascular boundary continuity and topological consistency. Evaluated via five-fold cross-validation on two public datasets, our method achieves IoU = 61.43%, Dice Similarity Coefficient (DSC) = 76.10%, and clDice = 79.36%, outperforming state-of-the-art models. The framework delivers robust, clinically relevant vessel segmentation, thereby supporting early diagnosis and precise treatment planning for coronary artery disease.
0 citationsRead paper
Recent publications
Latest Papers
Multi-Pass, Multi-View Blended Learning for High-Fidelity Volumetric CT Synthesis from Chest X-Rays
本文通过多通道多视角混合学习框架解决从单张2D胸透重建3D CT的问题，提高结构完整性和解剖细节。
0 citationsRead paper
A Two-Stage Deep Learning Framework for Segmentation of Ten Gastrointestinal Organs from Coronal MR Enterography
This study addresses the significant challenges in automatic segmentation of gastrointestinal organs from coronal MR enterography, which arise due to anatomical variability, low tissue contrast, and extreme class imbalance. To overcome these issues, the authors propose a coarse-to-fine, organ-aware two-stage deep learning framework. In the first stage, a DenseNet201-UNet++ architecture generates a coarse region-of-interest mask. The second stage employs a DenseNet121-SelfONN-UNet model trained on organ-specific image patches, augmented with a class-weighting strategy to mitigate imbalance. The proposed method achieves substantial performance gains, yielding an overall mean Dice similarity coefficient (mDSC) of 88.99% and mean intersection-over-union (mIoU) of 84.76%. Notably, segmentation accuracy for the appendix improves dramatically—from 6.76% to 85.76% in DSC—significantly outperforming existing baselines.
0 citationsRead paper
Tracing 3D Anatomy in 2D Strokes: A Multi-Stage Projection Driven Approach to Cervical Spine Fracture Identification
This study addresses the high computational complexity of detecting cervical spine fractures in 3D CT scans by proposing an efficient end-to-end framework based on multi-view 2D projections. The method reconstructs 3D anatomical structures using variance and energy projections, integrating YOLOv8 for localization, DenseNet121-Unet for segmentation, and a 2.5D temporal model for vertebra-level fracture identification. Interpretability is validated through saliency maps compared against expert annotations. The approach achieves strong performance with minimal computational overhead, yielding a 3D localization mIoU of 94.45%, a vertebra segmentation Dice score of 87.86%, and vertebra-level and patient-level F1 scores of 68.15% and 82.26%, respectively. Corresponding ROC-AUC values are 91.62% and 83.04%, demonstrating diagnostic accuracy comparable to that of radiology experts.
0 citationsRead paper
An Anatomy Aware Hybrid Deep Learning Framework for Lung Cancer Tumor Stage Classification
Lung cancer TNM staging relies on tumor size and spatial relationships with adjacent anatomical structures; however, existing end-to-end deep learning models lack interpretability and neglect critical anatomical context. To address this, we propose a hybrid deep learning framework integrating explicit anatomical priors: (1) a dedicated encoder-decoder network precisely segments the lungs, mediastinum, and tumor; (2) quantitative features—including maximum tumor diameter and shortest distances to key anatomical boundaries—are extracted from segmentation masks; and (3) clinical guideline rules are applied for T-stage classification. This is the first approach to explicitly embed clinically grounded anatomical context into the deep learning pipeline, overcoming the “black-box” limitation while ensuring both interpretability and clinical compliance. Evaluated on the Lung-PET-CT-Dx dataset, our method achieves 91.36% overall accuracy, with F1-scores of 0.93, 0.89, 0.96, and 0.90 for T1–T4 stages—significantly outperforming end-to-end baseline models.
0 citationsRead paper
CASR-Net: An Image Processing-focused Deep Learning-based Coronary Artery Segmentation and Refinement Network for X-ray Coronary Angiogram
This study addresses low segmentation accuracy, fragmented small vascular branches, and high false-positive rates in X-ray coronary angiography images. We propose a three-stage deep learning framework: (1) multi-channel preprocessing integrating CLAHE with an improved Ben Graham method to enhance contrast and suppress noise; (2) a backbone segmentation network leveraging a DenseNet121 encoder and a Self-ONN decoder to strengthen feature representation; and (3) a contour refinement module embedded to improve vascular boundary continuity and topological consistency. Evaluated via five-fold cross-validation on two public datasets, our method achieves IoU = 61.43%, Dice Similarity Coefficient (DSC) = 76.10%, and clDice = 79.36%, outperforming state-of-the-art models. The framework delivers robust, clinically relevant vessel segmentation, thereby supporting early diagnosis and precise treatment planning for coronary artery disease.
0 citationsRead paper