Article
Computer Science, Artificial Intelligence
Zhiquan He, Yupeng He, Wenming Cao
Summary: Deformable medical image registration is a crucial task in both theoretical research and clinical applications. This paper proposes a novel method that combines traditional registration techniques with deep learning approaches. By using attention-guided fusion of multi-scale deformation fields, the proposed method achieves significant improvement in registration accuracy and efficiency.
APPLIED INTELLIGENCE
(2023)
Article
Engineering, Biomedical
Xiao Liang, Howard Morgan, Ti Bai, Michael Dohopolski, Dan Nguyen, Steve Jiang
Summary: CBCT-based online adaptive radiotherapy requires accurate auto-segmentation, but DL-based direct segmentation of CBCT images is challenging due to poor quality and lack of well-labelled datasets. This study proposes a method using DIR and pseudo labels derived from deformed pCT contours for initial training, influencer volumes for defining the region of interest, and fine-tuning with a smaller set of true labels. Evaluation on nine patients shows that DL-based direct segmentation with influencer volumes improves performance to reach the level of DIR-based segmentation.
PHYSICS IN MEDICINE AND BIOLOGY
(2023)
Article
Oncology
Molly M. McCulloch, Guillaume Cazoulat, Stina Svensson, Sergii Gryshkevych, Bastien Rigaud, Brian M. Anderson, Ezgi Kirimli, Brian De, Ryan T. Mathew, Mohamed Zaid, Dalia Elganainy, Christine B. Peterson, Peter Balter, Eugene J. Koay, Kristy K. Brock
Summary: This study developed an automatic workflow for dose accumulation during liver cancer radiation therapy and evaluated the accuracy of different deformable image registration (DIR) algorithms. The use of contour-driven DIR methods showed better capability in estimating complex deformations of GI organs compared to intensity only-based DIR methods.
FRONTIERS IN ONCOLOGY
(2022)
Article
Radiology, Nuclear Medicine & Medical Imaging
Lin Ma, Weicheng Chi, Howard E. Morgan, Mu-Han Lin, Mingli Chen, David Sher, Dominic Moon, Dat T. Vo, Vladimir Avkshtol, Weiguo Lu, Xuejun Gu
Summary: In this study, a registration-guided DL segmentation framework for online cone beam computed tomography (CBCT) image was proposed. By integrating image registration algorithms and DL segmentation models, this framework overcame the issues of low image quality and limited training data, resulting in more accurate segmentation.
Article
Engineering, Biomedical
Huiqiao Xie, Yang Lei, Yabo Fu, Tonghe Wang, Justin Roper, Jeffrey D. Bradley, Pretesh Patel, Tian Liu, Xiaofeng Yang
Summary: An unsupervised deep learning-based CBCT-CBCT image registration method is proposed in this study to enable quantitative anatomic variation analysis. The method utilizes a spatial transformation-based network (STN) with a global and local GAN to predict the coarse- and fine-scale motions of CBCT images, and combines them with a global generated DVF to obtain the final deformation vector field. The experiments demonstrate that this method can achieve fast and accurate CBCT image registration, and effectively analyze and predict the anatomic changes of patients during radiotherapy.
PHYSICS IN MEDICINE AND BIOLOGY
(2023)
Article
Automation & Control Systems
Liang Qiu, Hongliang Ren
Summary: This article presents a joint learning framework named RSegNet for concurrent deformable registration and segmentation, achieving improved accuracy of both tasks. By minimizing an integrated loss function, utilizing data augmentation and dual-consistency supervision, the method demonstrates better anatomical consistency and deformation regularity, resulting in increased segmentation and registration accuracy.
IEEE TRANSACTIONS ON AUTOMATION SCIENCE AND ENGINEERING
(2022)
Article
Geochemistry & Geophysics
Haipeng Zhang, Chengcai Leng, Xiao Yan, Guorong Cai, Zhao Pei, Naigong Yu, Anup Basu
Summary: This letter introduces a novel method for remote sensing image registration based on local affine constraint. The method utilizes SIFT to extract feature points, performs initial matching using NNDR and FSC algorithm, and establishes fine registration with the local affine transformation circular region search algorithm. Experimental results demonstrate the subpixel accuracy and efficiency of the proposed method.
IEEE GEOSCIENCE AND REMOTE SENSING LETTERS
(2022)
Article
Environmental Sciences
Siliang Du, Yilin Xiao, Jingwei Huang, Mingwei Sun, Mingzhong Liu
Summary: GLFNet is a method proposed for the detection and matching of local features in remote-sensing images, leveraging existing sparse feature points as guided points. It addresses the challenge of establishing correct matches among images with significant differences in lighting or perspectives by searching for regions in the target image with features similar to the guided points in the source image. GLFNet consists of a coarse-level match network and a fine-level regression network to efficiently search for accurate matches.
Article
Chemistry, Multidisciplinary
Han Xie, Wenqi Zheng, Hyunchul Shin
Summary: A novel pedestrian detector using deformable attention-guided network (DAGN) was developed in this research, featuring a deformable convolution with an attention module (DCAM) and optimized loss function. Extensive evaluations on multiple datasets showed promising detection performance compared to state-of-the-art methods.
APPLIED SCIENCES-BASEL
(2021)
Article
Engineering, Biomedical
Shuwei Shao, Zhongcai Pei, Weihai Chen, Wentao Zhu, Xingming Wu, Baochang Zhang
Summary: The study introduces an efficient unsupervised deformable image registration framework incorporating the HAT module and AFS mechanism, demonstrating superior registration accuracy in experiments.
INTERNATIONAL JOURNAL OF COMPUTER ASSISTED RADIOLOGY AND SURGERY
(2022)
Article
Environmental Sciences
Yameng Hong, Chengcai Leng, Xinyue Zhang, Zhao Pei, Irene Cheng, Anup Basu
Summary: This paper introduces a new method called HOLBP for image registration, which improves traditional methods by redefining gradient and angle calculation and adding gradient direction information to form a 138-dimension descriptor vector. The experimental results demonstrate the stability and efficiency of this method for different test images.
Article
Multidisciplinary Sciences
Songwei Wang, Yuhang Wang, Ke Niu, Qian Li, Xiaoping Rao, Hui Zhao, Liwei Chen, Li Shi
Summary: This study introduces a novel method based on the JEMI network to solve the issue of regional localization in brain science research. By converting the brain slice and ARA into a segmentation map with unified modality, the method effectively overcomes the influence of non-unified modal images and achieves accurate and rapid localization of the brain slice.
Article
Computer Science, Interdisciplinary Applications
Dongming Wei, Sahar Ahmad, Yuyu Guo, Liyun Chen, Yunzhi Huang, Lei Ma, Zhengwang Wu, Gang Li, Li Wang, Weili Lin, Pew-Thian Yap, Dinggang Shen, Qian Wang
Summary: In this paper, a recurrently usable deep neural network is proposed for the registration of infant brain MR images. By using brain tissue segmentation maps for registration and training a single registration network that is recurrently applied in inference, the proposed method overcomes the challenge of fast brain development in infants. Experimental results show that the method achieves the highest registration accuracy while preserving the smoothness of the deformation field.
IEEE TRANSACTIONS ON MEDICAL IMAGING
(2022)
Article
Computer Science, Artificial Intelligence
Hee Guan Khor, Guochen Ning, Yihua Sun, Xu Lu, Xinran Zhang, Hongen Liao
Summary: The study proposes a Multi-Task Learning framework that combines registration and segmentation tasks to enhance image realism. By fusing high-level features, the registration network focuses on deformable parts with the help of initial anatomical segmentation. The proposed methodology outperforms previous approaches in brain and uterus MRI registration tasks, achieving state-of-the-art registration quality scores with 0.8% and 0.5% increases in DSC, respectively.
MEDICAL IMAGE ANALYSIS
(2023)
Article
Engineering, Biomedical
Chulong Zhang, Wenfeng He, Lin Liu, Jingjing Dai, Isah Salim Ahmad, Yaoqin Xie, Xiaokun Liang
Summary: In this paper, a medical image registration method based on volumetric feature points integration and bio-structure guidance is proposed. By using surface-registered point pairs and voxel feature point pairs to guide the training process, higher registration accuracy is achieved. The method has been validated on paired CT-CBCT datasets and shows a 6% improvement in precision compared to other deep learning methods, reaching a state-of-the-art status.
PHYSICS IN MEDICINE AND BIOLOGY
(2023)
Article
Computer Science, Interdisciplinary Applications
Rui Yan, Liangqiong Qu, Qingyue Wei, Shih-Cheng Huang, Liyue Shen, Daniel L. Rubin, Lei Xing, Yuyin Zhou
Summary: In this paper, a robust and label-efficient self-supervised federated learning (FL) framework for medical image analysis is proposed. The method introduces a novel Transformer-based self-supervised pre-training paradigm, which enhances the robustness of models against heterogeneous data and improves knowledge transfer to downstream models. Extensive experiments on simulated and real-world medical imaging non-IID federated datasets demonstrate the effectiveness of the proposed method.
IEEE TRANSACTIONS ON MEDICAL IMAGING
(2023)
Article
Radiology, Nuclear Medicine & Medical Imaging
Lianli Liu, Liyue Shen, Yong Yang, Emil Schueler, Wei Zhao, Gordon Wetzstein, Lei Xing
Summary: In this study, a NeRP beam data modeling technique was developed to predict beam characteristics from sparse measurements and detect data collection errors. The model showed promising results in predicting beam data and verifying its accuracy, which can simplify the commissioning and QA process without compromising the quality of medical physics service.
Article
Radiology, Nuclear Medicine & Medical Imaging
Charles Huang, Yusuke Nomura, Yong Yang, Lei Xing
Summary: In this study, the MetaPlanner Boosted VMAT (MPBV) approach is proposed to generate boosted VMAT plans through a fully automated framework. The MPBV approach maintains or improves dosimetric performance compared to traditional approaches with multiple arcs while substantially reducing delivery time. The efficacy of MPBV is evaluated on prostate and head and neck cases using clinically relevant plan quality metrics.
Article
Engineering, Biomedical
M. R. Ashraf, C. Gibson, L. Skinner, X. Gu, L. Xing, L. Wang
Summary: In this study, a high-resolution 3D printed radioluminescence-based imaging phantom was developed for quality assurance of a robotic arm linear accelerator. The phantom consisted of a scintillating sheet, fiducial markers, a low-cost camera, and a 3D printed enclosure. Various quality assurance tests were performed and compared with film data, showing excellent agreement and high spatial and temporal resolution.
PHYSICS IN MEDICINE AND BIOLOGY
(2023)
Article
Engineering, Biomedical
Oscar Pastor-Serrano, Steven Habraken, Mischa Hoogeman, Danny Lathouwers, Dennis Schaart, Yusuke Nomura, Lei Xing, Zoltan Perko
Summary: In radiotherapy, the internal movement of organs between treatment sessions causes errors in radiation dose delivery. Traditional motion models based on PCA are either patient-specific or population-based, but a hybrid approach is proposed in this study to predict patient-specific variations using population data. A deep learning probabilistic framework called DAM is used to generate deformation vector fields based on planning CT scans, and the model is trained using dataset of 312 CT pairs. Results show that DAM matches and improves upon previous PCA-based models, and accurately predicts deformations in treatment sessions.
PHYSICS IN MEDICINE AND BIOLOGY
(2023)
Article
Engineering, Biomedical
Charles Huang, Varun Vasudevan, Oscar Pastor-Serrano, Md Tauhidul Islam, Yusuke Nomura, Piotr Dubrowski, Jen-Yeu Wang, Joseph B. Schulz, Yong Yang, Lei Xing
Summary: In this work, we propose a content-based image retrieval method for retrieving dose distributions based on anatomical similarity. Our method trains a representation model to produce latent space embeddings of patients' anatomical information and achieves excellent retrieval performance. Various CBIR methods were evaluated, with the multitask Siamese network showing the best performance. The results demonstrate the potential of integrating CBIR into automated treatment planning workflows.
PHYSICS IN MEDICINE AND BIOLOGY
(2023)
Article
Multidisciplinary Sciences
Md Tauhidul Islam, Lei Xing
Summary: The authors develop a cartography strategy based on gene-gene interactions to transform high-dimensional gene expression data into a spatially configured genomap, enabling accurate deep pattern discovery. This approach presents significant challenges and opportunities in the field of single cell genomics for biomedical research. The unique cartography method captures gene interactions in the spatial configuration of genomaps, allowing for the extraction of deep genomic interaction features and the discovery of discriminative patterns in the data.
NATURE COMMUNICATIONS
(2023)
Article
Engineering, Biomedical
Siqi Ye, Liyue Shen, Md Tauhidul Islam, Lei Xing
Summary: In this study, a reference-free statistical implicit neural representation (INR) framework is proposed for image super-resolution in biomedical imaging. The framework utilizes a small number of observed low-resolution images for training to generate high-quality super-resolved images. The efficacy of the framework is validated at different magnification scales.
PHYSICS IN MEDICINE AND BIOLOGY
(2023)
Article
Engineering, Biomedical
Yusuke Nomura, M. Ramish Ashraf, Mengying Shi, Lei Xing
Summary: This study developed a novel deep learning-based model to correct the fluorescence images contaminated by Cherenkov light, scattered light or background noise for accurate dosimetric application.
PHYSICS IN MEDICINE AND BIOLOGY
(2023)
Article
Engineering, Biomedical
Xiao Jia, Eugene Shkolyar, Mark A. Laurie, Okyaz Eminaga, Joseph C. Liao, Lei Xing
Summary: This study developed a deep learning algorithm for accurate detection of bladder tumors in white-light cystoscopy. The algorithm, CystoNet-T, uses a transformer-augmented pyramidal CNN architecture to improve tumor detection performance. The algorithm achieved high F1 and AP scores on a test set of patient data.
PHYSICS IN MEDICINE AND BIOLOGY
(2023)
Article
Multidisciplinary Sciences
Yuming Jiang, Zhicheng Zhang, Wei Wang, Weicai Huang, Chuanli Chen, Sujuan Xi, M. Usman Ahmad, Yulan Ren, Shengtian Sang, Jingjing Xie, Jen-Yeu Wang, Wenjun Xiong, Tuanjie Li, Zhen Han, Qingyu Yuan, Yikai Xu, Lei Xing, George A. Poultsides, Guoxin Li, Ruijiang Li
Summary: Significant progress has been made in using deep learning for cancer detection and diagnosis, but predicting treatment response and outcomes remains challenging. This study presents a biology-guided deep learning approach that simultaneously predicts tumor microenvironment status and treatment outcomes from medical images. The model was validated for gastric cancer prognosis and response to adjuvant chemotherapy, and it complements clinically approved biomarkers, providing information for patient selection.
NATURE COMMUNICATIONS
(2023)
Proceedings Paper
Optics
Xiao Jia, Eugene Shkolyar, Okyaz Eminaga, Mark A. Laurie, Zixia Zhou, Timothy Lee, Md Tauhidul Islam, Max Q. -H. Meng, Joseph C. Liao, Lei Xing
Summary: In this study, a deep-learning algorithm called CystoNet-F was developed to enhance the detection of flat lesions in white-light cystoscopy. By incorporating domain translation, transfer learning, and region of interest detection, the algorithm improves the ability to detect lesions in standard white-light cystoscopy.
ADVANCED PHOTONICS IN UROLOGY 2023
(2023)
Proceedings Paper
Optics
Mark Laurie, Okyaz Eminaga, Eugene Shkolyar, Xiao Jia, Timothy Lee, Jin Long, Md Tauhidul Islam, Hubert Lau, Lei Xing, Joseph C. Liao
Summary: In this study, four advanced sequential models (SlowFast, Multiscale Vision Transformers, X3D, and CNN-LSTM) were used to train and classify bladder tumors. The best performing model was X3D with a sequence length of 8, achieving 100% sensitivity, 94.7% sensitivity, and 80.0% specificity. Sequential modeling has the potential to accurately classify a wide variety of bladder tumors in a real-time setting.
ADVANCED PHOTONICS IN UROLOGY 2023
(2023)
Proceedings Paper
Computer Science, Artificial Intelligence
Bowen Song, Liyue Shen, Lei Xing
Summary: Recently, deep learning has been used to address medical image reconstruction problems in CT scans with limited views. However, these models have difficulty generalizing to new domains and the privacy concerns limit access to training data. To overcome these challenges, a source-free black-box test-time adaptation method called PINER is introduced. By leveraging implicit neural representation learning, the method can adapt to unknown noise levels and outperforms state-of-the-art algorithms.
2023 IEEE/CVF WINTER CONFERENCE ON APPLICATIONS OF COMPUTER VISION (WACV)
(2023)
Article
Computer Science, Artificial Intelligence
Shengtian Sang, Yuyin Zhou, Md Tauhidul Islam, Lei Xing
Summary: This article introduces a novel attention-based method called AFMA, which addresses the issue of information loss in small object segmentation. By utilizing different feature levels of the original image, AFMA quantifies the inner relationship between large and small objects of the same category, compensating for the loss of high-level feature information and improving segmentation accuracy. Extensive experiments demonstrate the substantial and consistent improvement of small object segmentation achieved by our method.
IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE
(2023)