Current Organizations / Institutes

The organizations/institutes from the scholar's latest article


Qiong Liu has a total of 22 co-authors. Here are the top co-authors by number of shared publications:


Latest Publications

  • Open Access

    Edge Feature Extraction Medical Image Segmentation Network Based on Frequency Domain Data Enhancement

    by Chaofan Li, Qiong Liu, Jinnan Teng

    Accurate boundary segmentation of anatomically complex structures such as organs and tumors remains a major challenge in medical image analysis. Traditional segmentation methods often overlook the extraction of high-frequency information such as texture and boundary details, despite their critical importance in guiding clinical surgical planning. To address this issue, we proposed a hierarchical frequency-domain enhancement strategy in the encoder, which performs effective separation and reconstruction of high- and low-frequency features via Fourier transform combined with a low-pass masking operation. This design enhances the model’s sensitivity to high-frequency texture and fine-grained boundary signals. Furthermore, in the decoder, we introduce a boundary refinement mechanism that dynamically combines Scharr operators with Laplacian-based sharpening to enhance structural clarity. This effectively alleviates segmentation discontinuities and boundary fragmentation issues commonly observed in regions with blurred edges using conventional methods. We conduct extensive experiments on both the BraTS 2020 MRI dataset and the Synapse CT dataset. Our method achieves an average Dice score of 86.79% on BraTS 2020 and 84.23% on Synapse, representing relative improvements of 0.99% and 0.90%, respectively, over strong state-of-the-art baselines. In terms of boundary precision, our approach reduces the 95% Hausdorff Distance (HD) by 13.2% on BraTS 2020 and 30.1% on Synapse, demonstrating superior boundary segmentation capability. Ablation studies further validate the effectiveness of each proposed component.


  • Open Access

    Boundary-enhanced sparse transformer for generalizable and accurate medical image segmentation

    by Chaofan Li, Qiong Liu, Jianxiang Song

    Medical image segmentation is a fundamental task in computer-aided diagnosis, playing a crucial role in organ structure analysis, lesion delineation, and treatment planning. However, current Transformer-based segmentation networks still face two major challenges. First, the global self-attention in the encoder often introduces redundant connections, leading to high computational cost and potential interference from irrelevant tokens. Second, the decoder shows limited capability in reconstructing fine-grained boundary structures, resulting in blurred segmentation contours. To address these issues, we proposed an efficient and accurate framework for general medical image segmentation. Specifically, in the encoder, we introduce a frequency-domain similarity measure and construct a Key-Semantic Dictionary (KSD) via amplitude spectrum cosine similarity. This enables stage-wise sparse attention matrices that reduce redundancy and enhance semantic relevance. In the decoder, we design a learnable gradient-based operator that injects boundary-aware logits bias into the attention mechanism, thereby improving structural detail recovery along object boundaries. On ACDC, the framework delivers a 0.55% gain in average Dice and a 14.6% reduction in HD over the second-best baseline. On ISIC 2018, it achieves increases of 1.01% in Dice and 0.21% in ACC over the second-best baseline, while using 88.8% fewer parameters than typical Transformer-based models. On Synapse, it surpasses the strongest prior approach by 1.03% in Dice and 6.35% in HD, yielding up to 8.36% Dice improvement and 52.46% HD reduction compared with widely adopted Transformer baselines. Comprehensive results confirm that the proposed frequency-domain sparse attention and learnable edge-guided decoding effectively balance segmentation accuracy, boundary fidelity, and computational cost. This framework not only suppresses redundant global correlations and enhances structural detail reconstruction, but is also robust to different medical imaging modalities, providing a lightweight and clinically applicable solution for high-precision medical image segmentation.


  • Open Access

    An Efficient Lightweight Transformer Framework for Precise Cardiac Segmentation via Frequency-Domain Decomposition

    by Chaofan Li, Qiong Liu, Jianxiang Song

    Cardiac MRI segmentation remains hindered by two persistent obstacles. First, boundary ambiguity arises from poor contrast between the myocardium and blood pool, the presence of papillary muscles and trabeculations within the cavity, partial-volume artifacts, and locally thin myocardial walls, all of which lead to contour discontinuities and erosion. Second, computational burden becomes prohibitive as conventional self-attention scales quadratically with the spatial and slice dimensions of high-resolution CMR volumes.To address these challenges, we introduce an efficient lightweight Transformer specifically designed for cardiac MRI. Our framework comprises two key modules: (1) a Laplacian pyramid decomposition, which separates features into high-, mid-, and low-frequency sub-bands, where the proposed lightweight attention compresses key–value representations to reduce complexity while retaining high-frequency boundary cues; and (2) a multi-scale efficient fusion bridge integrated into skip connections, which aligns cross-scale semantics and enhances residual details with minimal overhead. Experiments on ACDC and MS-CMRSeg 2019 show 0.5% relative improvements in Dice and 5.4% reductions in Hausdorff Distance (HD) over state-of-the-art models, achieving superior boundary accuracy and reduced computational complexity on high-resolution, multi-slice cardiac MRI.


  • Open Access

    Medical image segmentation based on frequency domain decomposition SVD linear attention

    by Liu Qiong, Li Chaofan, Teng Jinnan, Chen Liping, Song Jianxiang

    Convolutional Neural Networks (CNNs) have achieved remarkable segmentation accuracy in medical image segmentation tasks. However, the Vision Transformer (ViT) model, with its capability of extracting global information, offers a significant advantage in contextual information compared to the limited receptive field of convolutional kernels in CNNs. Despite this, ViT models struggle to fully detect and extract high-frequency signals, such as textures and boundaries, in medical images. These high-frequency features are essential in medical imaging, as targets like tumors and pathological organs exhibit significant differences in texture and boundaries across different stages. Additionally, the high resolution of medical images leads to computational complexity in the self-attention mechanism of ViTs. To address these limitations, we propose a medical image segmentation network framework based on frequency domain decomposition using a Laplacian pyramid. This approach selectively computes attention features for high-frequency signals in the original image to enhance spatial structural information effectively. During attention feature computation, we introduce Singular Value Decomposition (SVD) to extract an effective representation matrix from the original image, which is then applied in the attention computation process for linear projection. This method reduces computational complexity while preserving essential features. We demonstrated the segmentation validity and superiority of our model on the Abdominal Multi-Organ Segmentation dataset and the Dermatological Disease dataset, and on the Synapse dataset our model achieved a score of 82.68 on the Dice metrics and 17.23 mm on the HD metrics. Experimental results indicate that our model consistently exhibits segmentation effectiveness and improved accuracy across various datasets.


  • Open Access

    Optimization of the Algorithm for the Implementation of Point Spread Function in the 3D-OSEM Reconstruction Algorithm Based on the List-Mode Micro PET Data

    by Jie Zhao, Yunfeng Song, Qiong Liu, Shijie Chen, Jyh-Cheng Chen

    Positron emission tomography (PET) is a popular research topic. People are becoming more interested in PET images as they become more widely available. However, the partial volume effect (PVE) in PET images remains one of the most influential factors causing the resolution of PET images to degrade. It is possible to reduce this PVE and achieve better image quality by measuring and modeling the point spread function (PSF) and then accounting for it inside the reconstruction algorithm. In this work, we examined the response characteristics of the MetisTM PET/CT system by acquiring 22Na point source at different locations in the field of view (FOV) of the scanner and reconstructing with small pixel size for images to obtain their radial, tangential, and axial full-width half maximum (FWHM). An image-based model of the PSF model was then obtained by fitting asymmetric two-dimensional Gaussians on the 22Na images. This PSF model determined by FWHM in three directions was integrated into a three-dimensional ordered subsets expectation maximization (3D-OSEM) algorithm based on a list-mode format to form a new PSF-OSEM algorithm. We used both algorithms to reconstruct point source, Derenzo phantom, and mouse PET images and performed qualitative and quantitative analyses. In the point source study, the PSF-OSEM algorithm reduced the FWHM of the point source PET image in three directions to about 0.67 mm, and in the phantom study, the PET image reconstructed by the PSF-OSEM algorithm had better visual effects. At the same time, the quantitative analysis results of the Derenzo phantom were better than the original 3D-OSEM algorithm. In the mouse experiment, the results of qualitative and quantitative analyses showed that the imaging quality of PSF-OSEM algorithm was better than that of 3D-OSEM algorithm. Our results show that adding the PSF model to the 3D-OSEM algorithm in the MetisTM PET/CT system helps to improve the resolution of the image and satisfy the qualitative and quantitative analysis criteria.


  • Open Access

    Optimization of Spatial Resolution and Image Reconstruction Parameters for the Small-Animal Metis™ PET/CT System

    by Jie Zhao, Qiong Liu, Chaofan Li, Yunfeng Song, Ying Zhang, Jyh-Cheng Chen

    Purpose: The aim of this study was to investigate the optimization of the spatial resolution and image reconstruction parameters related to image quality in an iterative reconstruction algorithm for the small-animal Metis™ PET/CT system. Methods: We used a homemade Derenzo phantom to evaluate the image quality using visual assessment, the signal-to-noise ratio, the contrast, the coefficient of variation, and the contrast-to-noise ratio of the 0.8 mm hot rods of eight slices in the center of the phantom PET images. A healthy mouse study was performed to analyze the influence of the optimal reconstruction parameters and the Gaussian post-filter FWHM. Results: In the phantom study, the image quality was the best when the phantom was placed at the end, keeping the central axis parallel to the X-axis of the system, and selecting between 30 and 40 iterations, a 0.314 mm reconstructed voxel size, and a 1.57 mm Gaussian post-filter FWHM. The optimization of the spatial resolution could reach 0.6 mm. In the animal study, it was suitable to choose a voxel size of 0.472 mm, between 30 and 40 iterations, and a 2.36 mm Gaussian post-filter FWHM. Conclusions: Our results indicate that the optimal imaging conditions and reconstruction parameters are very necessary to obtain high-resolution images and quantitative accuracy, especially for the high-precision recognition of tiny lesions.


  • Open Access

    DCCL: Dual-channel hybrid neural network combined with self-attention for text classification

    by Chaofan Li, Qiong Liu, Kai Ma

    Text classification is a fundamental task in natural language processing. The Chinese text classification task suffers from sparse text features, ambiguity in word segmentation, and poor performance of classification models. A text classification model is proposed based on the self-attention mechanism combined with CNN and LSTM. The proposed model uses word vectors as input to a dual-channel neural network structure, using multiple CNNs to extract the N-Gram information of different word windows and enrich the local feature representation through the concatenation operation, the BiLSTM is used to extract the semantic association information of the context to obtain the high-level feature representation at the sentence level. The output of BiLSTM is feature weighted with self-attention to reduce the influence of noisy features. The outputs of the dual channels are concatenated and fed into the softmax layer for classification. The results of the multiple comparison experiments showed that the DCCL model obtained 90.07% and 96.26% F1-score on the Sougou and THUNews datasets, respectively. Compared to the baseline model, the improvement was 3.24% and 2.19%, respectively. The proposed DCCL model can alleviate the problem of CNN losing word order information and the gradient of BiLSTM when processing text sequences, effectively integrate local and global text features, and highlight key information. The classification performance of the DCCL model is excellent and suitable for text classification tasks.


  • Technical Note: Performance evaluation of a small‐animal PET/CT system based on NEMA NU 4–2008 standards

    by Qiong Liu, Chaofan Li, Jiguo Liu, Kishore Krish, Xinlei Fu, Jie Zhao, Jyh‐Cheng Chen

    Purpose: The MetisTM PET/CT is a self‐developed, silicon photomultiplier (SiPM) detector‐based, rodent PET/CT system. The objective of this study was to evaluate the performance of the system using the National Electrical Manufacturers Association (NEMA) NU 4–2008 standard protocol.Methods: Energy resolution, spatial resolution, sensitivity, scatter fraction (SF), noise‐equivalent count rate (NECR), and image quality (IQ) characteristics were measured. A micro Derenzo phantom experiment was performed to evaluate the spatial resolution using three‐dimensional ordered‐subsets expectation maximization (3D‐OSEM) and maximum likelihood expectation maximization (MLEM) reconstructed images. In addition, the CT imaging agent Ioverol 350 was mixed with fluorine‐18 (18F)‐fluorodeoxyglucose (FDG) and then injected into the micro Derenzo phantom to evaluate the PET/CT imaging. In vivo PET/CT imaging studies were also conducted in a healthy mouse and rat using 18F‐FDG.Results: The mean energy resolution of the system was 15.3%. The tangential resolution was 0.82 mm full‐width half‐maximum (FWHM) at the center of the field of the view (FOV), and the radial and axial resolution were generally lower than 2.0 mm FWHM. The spatial resolution was significantly improved when using 3D‐OSEM, especially the axial FWHM could be improved by up to about 57%. The system absolute sensitivity was 7.7% and 6.8% for an energy window of 200–750 and 350–750 keV respectively. The scatter fraction was 8.2% and 12.1% for the mouse‐ and rat‐like phantom respectively. The peak NECR was 1343.72 kcps at 69 MBq and 640.32 kcps at 53 MBq for the mouse‐ and rat‐like phantom respectively. The 1‐mm fillable rod in the IQ phantom can be clearly observed. We can identify the 0.6‐mm aperture of the micro Derenzo phantom image clearly using 3D‐OSEM (10 subsets, 5 iterations). We also performed the fusion of the PET and CT images of the mouse and the brain imaging of the rat.Conclusions: The results show that the system has the characteristics of high‐resolution, high‐sensitivity, and excellent IQ and is suitable for rodent imaging‐based research.