Seungryong Cho
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Seungryong Cho has a total of 188 co-authors. Here are the top co-authors by number of shared publications:
Jonghwan Min
Department of Nuclear and Quantum Engineering, Korea Advanced Institute of Science and Technology(KAIST), 291 Daehak-ro, Yuseong-gu, Daejeon, 34141, Republic of Korea
8 shared publicationsTaewon Lee
Hoseo Univ, Dept Semicond Engn, Asan 31499, South Korea
7 shared publicationsRizza Pua
Department of Nuclear and Quantum Engineering, Korea Advanced Institute of Science and Technology(KAIST), 291 Daehak-ro, Yuseong-gu, Daejeon, 34141, Republic of Korea
6 shared publications- Hyeongseok Kim
Harvard-MIT Division of Health Sciences and Technology, Massachusetts Institute of Technology, 77 Massachusetts Ave, Cambridge, Massachusetts, 02139, UNITED STATES
5 shared publications Jin Sung Kim
Affiliations Department of Radiation Oncology, MPBEL, Yonsei Cancer Center, Yonsei University College of Medicine, Seoul, Republic of Korea, Department of Medicine, Yonsei University College of Medicine, Seoul, Republic of Korea
5 shared publications- Seoyoung Lee
KAIST
5 shared publications
Latest Publications
- Open Access
- 10 July 2026
- Published by IOP Publishing in Physics in Medicine & Biology
- Vol. 71 (13) , 135034
Four-dimensional on-beam computed tomography reconstruction using projection-difference images
Purpose. Accurate verification of delivered dose requires anatomical information during the treatment beam delivery. However, conventional cone-beam computed tomography (CBCT) is typically acquired prior to beam delivery and fails to account for intrafractional tumor shift/drift. This study aims to develop a 4D on-beam computed tomography (4D OBCT) framework to capture anatomical states during irradiation. Materials and methods. A thoracic phantom with rigid tumor motion was used to develop a conformal arc treatment plan. During delivery, the target was displaced by 1 cm superior to its planned position to simulate a baseline shift, and megavoltage transmission data were acquired using an electronic portal imaging device. Transmission data corresponding to the planned tumor position were obtained from the planning CT (pCT) via Monte Carlo simulation. Volumetric difference images were reconstructed from the differences between the measured and simulated transmission data for each phase. Subsequently, the images were superimposed onto the corresponding phases of the 4D pCT to obtain 4D OBCT. For evaluation, we compared 4D OBCT with 4D pCT (ground-truth) and 4D CBCT which were acquired after the target displacement. A dosimetric performance was evaluated by recalculating 4D dose distributions on the three images and comparing in dose-volume histograms and gamma indices. Results. 4D OBCT achieved lower root mean square error than 4D CBCT in both the lung (58.46 HU vs 74.37 HU) and chest wall regions (23.78 HU vs 66.84 HU). This quantitative accuracy was also reflected in the dosimetric comparison. Under 2%/2 mm and 1%/1 mm criteria, 4D OBCT achieved gamma pass rates of 99.76% and 99.08%, respectively, whereas 4D CBCT yielded 72.50% and 57.77%, respectively. Conclusions. The proposed 4D OBCT framework successfully reconstructed accurate phase-resolved images of the target during irradiation. By providing high-resolution volumetric images that reflect intrafractional motion, 4D OBCT could provide a robust solution for precise post-delivery dosimetric verification and adaptation.
- 10 July 2026
- Published by IOP Publishing in Physics in Medicine & Biology
- Vol. 71 (13) , 135034
Open AccessFour-dimensional on-beam computed tomography reconstruction using projection-difference images
Purpose. Accurate verification of delivered dose requires anatomical information during the treatment beam delivery. However, conventional cone-beam computed tomography (CBCT) is typically acquired prior to beam delivery and fails to account for intrafractional tumor shift/drift. This study aims to develop a 4D on-beam computed tomography (4D OBCT) framework to capture anatomical states during irradiation. Materials and methods. A thoracic phantom with rigid tumor motion was used to develop a conformal arc treatment plan. During delivery, the target was displaced by 1 cm superior to its planned position to simulate a baseline shift, and megavoltage transmission data were acquired using an electronic portal imaging device. Transmission data corresponding to the planned tumor position were obtained from the planning CT (pCT) via Monte Carlo simulation. Volumetric difference images were reconstructed from the differences between the measured and simulated transmission data for each phase. Subsequently, the images were superimposed onto the corresponding phases of the 4D pCT to obtain 4D OBCT. For evaluation, we compared 4D OBCT with 4D pCT (ground-truth) and 4D CBCT which were acquired after the target displacement. A dosimetric performance was evaluated by recalculating 4D dose distributions on the three images and comparing in dose-volume histograms and gamma indices. Results. 4D OBCT achieved lower root mean square error than 4D CBCT in both the lung (58.46 HU vs 74.37 HU) and chest wall regions (23.78 HU vs 66.84 HU). This quantitative accuracy was also reflected in the dosimetric comparison. Under 2%/2 mm and 1%/1 mm criteria, 4D OBCT achieved gamma pass rates of 99.76% and 99.08%, respectively, whereas 4D CBCT yielded 72.50% and 57.77%, respectively. Conclusions. The proposed 4D OBCT framework successfully reconstructed accurate phase-resolved images of the target during irradiation. By providing high-resolution volumetric images that reflect intrafractional motion, 4D OBCT could provide a robust solution for precise post-delivery dosimetric verification and adaptation.
- Open Access
- 23 May 2026
- Published by Wiley in Journal of Applied Clinical Medical Physics
- Vol. 27 (5) , e70626
Source‐free domain adaptation for multi‐institutional chest X‐ray images
Background: Machine learning models for chest X‐ray (CXR)‐based tuberculosis (TB) detection often suffer from performance drops when deployed across institutions due to domain shifts caused by differences in patient populations, imaging protocols, and equipment. Domain adaptation has been widely studied to address the problem, but privacy regulations restrict data sharing, making conventional approaches infeasible. Purpose: We aim to enhance TB detection performance when a detector model is shared across multiple clinical sites without accessing any source data or statistics. This problem is related to source‐free domain adaptation (SFDA) but requires special designs for imbalanced binary classification. Methods: We propose two methods for TB detection, which are based on Source HypOthesis Transfer (SHOT) and SFDA via source Distribution Estimation (SFDA‐DE), respectively. As the main ideas of our methods, we propose ranking‐based pseudo‐labeling and selective fine‐tuning of network layers to address severe class imbalance and prevent overfitting. Results: On six unseen target domains with high class imbalance, our proposed SFDA methods significantly improved the average F1 score and AUROC compared to non‐adapted models, all without accessing source data during adaptation. Conclusions: Our methods offer a privacy‐preserving and effective strategy to improve the generalization of TB detection models across institutions, addressing domain shifts while complying with data‐sharing restrictions.
- 23 May 2026
- Published by Wiley in Journal of Applied Clinical Medical Physics
- Vol. 27 (5) , e70626
Open AccessSource‐free domain adaptation for multi‐institutional chest X‐ray images
Background: Machine learning models for chest X‐ray (CXR)‐based tuberculosis (TB) detection often suffer from performance drops when deployed across institutions due to domain shifts caused by differences in patient populations, imaging protocols, and equipment. Domain adaptation has been widely studied to address the problem, but privacy regulations restrict data sharing, making conventional approaches infeasible. Purpose: We aim to enhance TB detection performance when a detector model is shared across multiple clinical sites without accessing any source data or statistics. This problem is related to source‐free domain adaptation (SFDA) but requires special designs for imbalanced binary classification. Methods: We propose two methods for TB detection, which are based on Source HypOthesis Transfer (SHOT) and SFDA via source Distribution Estimation (SFDA‐DE), respectively. As the main ideas of our methods, we propose ranking‐based pseudo‐labeling and selective fine‐tuning of network layers to address severe class imbalance and prevent overfitting. Results: On six unseen target domains with high class imbalance, our proposed SFDA methods significantly improved the average F1 score and AUROC compared to non‐adapted models, all without accessing source data during adaptation. Conclusions: Our methods offer a privacy‐preserving and effective strategy to improve the generalization of TB detection models across institutions, addressing domain shifts while complying with data‐sharing restrictions.
- Open Access
- 20 May 2026
- Published by IOP Publishing in Physics in Medicine & Biology
- Vol. 71 (10) , 105007
Diffusion prior-guided implicit neural representation for metal artifact reduction in sparse-view CT reconstruction
Objective . Sparse-view computed tomography (SVCT) reduces radiation exposure, but it inevitably leads to severe streak artifacts, which become even more pronounced in the presence of metallic implants. Most existing methods address SVCT and metal artifact reduction (MAR) separately, making the joint SVCT and MAR (SVMAR) problem suboptimal. While a few methods have been proposed to tackle SVMAR jointly, most are supervised and require large paired datasets that are difficult to obtain in clinical practice. To overcome these limitations, we propose a self-supervised framework that leverages a denoising diffusion probabilistic model and an implicit neural representation (INR) to address the SVMAR problem. Approach . First, an INR is optimized to produce an initial reconstruction using sparse-view measurements by sampling only rays outside the metal trace. This initial estimate is then used to accelerate the reverse diffusion process. During reverse diffusion, we alternately perform: (i) a MAR step, where diffusion priors guide the inpainting of metal-trace regions in the measurements using Poisson blending; and (ii) an SVCT step, where data fidelity is enforced by refining the INR with both diffusion priors and the MAR-corrected sinograms obtained from the MAR step. Main result . Experiments on both simulation and clinical datasets show that the proposed method reconstructs high-quality images without requiring large paired datasets. Quantitative evaluations demonstrated that the proposed method outperformed existing methods including IndudoNet+ on the OOD dataset across 160-, 80-, and 40-view settings, with peak signal-to-noise ratio (PSNR) improving from 36.28 to 45.39 dB, 32.69–44.22 dB, and 31.37–41.90 dB, and structural similarity index measure (SSIM) increasing from 0.955 to 0.988, 0.920–0.983, and 0.906–0.973, respectively. Significance . By leveraging diffusion priors within a self-supervised INR framework, the method provides a practical and generalizable solution for real-world SVMAR scenarios where ground-truth images are unavailable.
- 20 May 2026
- Published by IOP Publishing in Physics in Medicine & Biology
- Vol. 71 (10) , 105007
Open AccessDiffusion prior-guided implicit neural representation for metal artifact reduction in sparse-view CT reconstruction
Objective . Sparse-view computed tomography (SVCT) reduces radiation exposure, but it inevitably leads to severe streak artifacts, which become even more pronounced in the presence of metallic implants. Most existing methods address SVCT and metal artifact reduction (MAR) separately, making the joint SVCT and MAR (SVMAR) problem suboptimal. While a few methods have been proposed to tackle SVMAR jointly, most are supervised and require large paired datasets that are difficult to obtain in clinical practice. To overcome these limitations, we propose a self-supervised framework that leverages a denoising diffusion probabilistic model and an implicit neural representation (INR) to address the SVMAR problem. Approach . First, an INR is optimized to produce an initial reconstruction using sparse-view measurements by sampling only rays outside the metal trace. This initial estimate is then used to accelerate the reverse diffusion process. During reverse diffusion, we alternately perform: (i) a MAR step, where diffusion priors guide the inpainting of metal-trace regions in the measurements using Poisson blending; and (ii) an SVCT step, where data fidelity is enforced by refining the INR with both diffusion priors and the MAR-corrected sinograms obtained from the MAR step. Main result . Experiments on both simulation and clinical datasets show that the proposed method reconstructs high-quality images without requiring large paired datasets. Quantitative evaluations demonstrated that the proposed method outperformed existing methods including IndudoNet+ on the OOD dataset across 160-, 80-, and 40-view settings, with peak signal-to-noise ratio (PSNR) improving from 36.28 to 45.39 dB, 32.69–44.22 dB, and 31.37–41.90 dB, and structural similarity index measure (SSIM) increasing from 0.955 to 0.988, 0.920–0.983, and 0.906–0.973, respectively. Significance . By leveraging diffusion priors within a self-supervised INR framework, the method provides a practical and generalizable solution for real-world SVMAR scenarios where ground-truth images are unavailable.
- Open Access
- 4 February 2026
- Published by Elsevier in SSRN Electronic Journal
A 3D scan–driven method for activity estimation under complex source–detector geometries
In gamma-ray spectrometry, efficiency calibration using geometry-matched certified reference materials is often impractical for irregularly shaped radioactive sources and off-axis setups. Semi-empirical efficiency calibration software packages such as the ISOCS and ANGLE have been developed to address this issue; however, because these tools rely on simplified assumptions and user-defined parameters, they have limitations in covering complex source–detector geometries, leading to significant deviations in activity estimation. This study presents a three-dimensional (3D) scanner–based method that directly models the complete source–detector geometry and incorporates their relative positions into Monte Carlo simulations for efficiency calibration. This framework enables precise activity evaluation for complex geometries without relying on geometric simplifications. Experimental validation was performed using fabricated showerhead- and turbine-shaped reference materials, and the results were compared with those obtained using ANGLE. The proposed method reproduced certified activity values with deviations of up to ±15%, whereas ANGLE exhibited larger deviations depending on the assumed model dimensions. These findings demonstrate that an accurate geometry modeling can markedly improve agreement with the certified values. Compared with the semi-empirical efficiency transfer software such as ANGLE, the proposed 3D scanner–based approach provides enhanced accuracy and robustness for complex source–detector configurations. This highlights the potential of the proposed method for reliable in-situ gamma spectrometry of irregularly shaped materials and nonstandard measurement conditions.
- 4 February 2026
- Published by Elsevier in SSRN Electronic Journal
Open AccessA 3D scan–driven method for activity estimation under complex source–detector geometries
In gamma-ray spectrometry, efficiency calibration using geometry-matched certified reference materials is often impractical for irregularly shaped radioactive sources and off-axis setups. Semi-empirical efficiency calibration software packages such as the ISOCS and ANGLE have been developed to address this issue; however, because these tools rely on simplified assumptions and user-defined parameters, they have limitations in covering complex source–detector geometries, leading to significant deviations in activity estimation. This study presents a three-dimensional (3D) scanner–based method that directly models the complete source–detector geometry and incorporates their relative positions into Monte Carlo simulations for efficiency calibration. This framework enables precise activity evaluation for complex geometries without relying on geometric simplifications. Experimental validation was performed using fabricated showerhead- and turbine-shaped reference materials, and the results were compared with those obtained using ANGLE. The proposed method reproduced certified activity values with deviations of up to ±15%, whereas ANGLE exhibited larger deviations depending on the assumed model dimensions. These findings demonstrate that an accurate geometry modeling can markedly improve agreement with the certified values. Compared with the semi-empirical efficiency transfer software such as ANGLE, the proposed 3D scanner–based approach provides enhanced accuracy and robustness for complex source–detector configurations. This highlights the potential of the proposed method for reliable in-situ gamma spectrometry of irregularly shaped materials and nonstandard measurement conditions.
- Open Access
- 24 October 2025
- Published by Wiley in Journal of Applied Clinical Medical Physics
- Vol. 26 (11) , e70317
Metal artifact reduction algorithm with conditional latent diffusion model for dental cone‐beam CT
Background: Metal artifacts in computed tomography pose great challenges to diagnosis and treatment planning. Various metal artifact reduction techniques have been developed tackling the artifacts in the sinogram, projection, and image domains. Purpose: We aimed to reduce the metal artifacts via latent diffusion network and improve normalized metal artifact reduction (NMAR) scheme with metal segmentation network and secondary artifact correction network in dental cone‐beam computed tomography (CBCT). Methods: We first produced a metal‐artifact‐reduced image through a latent diffusion model (LDM) with the metal‐artifact‐corrupted image as the condition. A combination of mean squared error (MSE) and learned perceptual image patch similarity (LPIPS) loss were used as the objective function to train the network. To resolve the concerns of an image from the generative model such as hallucination, we used the image as a prior for a modified normalized metal artifact reduction (NMAR) process which is a well‐known analytic scheme. The modified NMAR in this work includes an automatic metal segmentation network and a secondary artifact correction network to enhance the MAR performance. Results: The proposed method showed significant outperformance over the models such as classical NMAR and a state‐of‐the‐art convolutional‐neural‐network‐based MAR (CNNMAR) approach for CBCT. The major improvements to metal artifact reduction are attributed to the improved NMAR prior estimated by the LDM. The proposed method improved the image quality compared to CNNMAR in terms of RMSE, PSNR, and SSIM from 34.78 × 10 −4 to 19.30 × 10 −4 , 49.3–54.3, and 91.2–97.2, respectively. Its clinical dental implementation was also explored and showcased success in reducing metal artifacts while preserving other tissue structures. Conclusions: The proposed method has shown its practical utility in dental CBCT for reducing metal artifacts and is believed to contribute to dental diagnosis and treatment.
- 24 October 2025
- Published by Wiley in Journal of Applied Clinical Medical Physics
- Vol. 26 (11) , e70317
Open AccessMetal artifact reduction algorithm with conditional latent diffusion model for dental cone‐beam CT
Background: Metal artifacts in computed tomography pose great challenges to diagnosis and treatment planning. Various metal artifact reduction techniques have been developed tackling the artifacts in the sinogram, projection, and image domains. Purpose: We aimed to reduce the metal artifacts via latent diffusion network and improve normalized metal artifact reduction (NMAR) scheme with metal segmentation network and secondary artifact correction network in dental cone‐beam computed tomography (CBCT). Methods: We first produced a metal‐artifact‐reduced image through a latent diffusion model (LDM) with the metal‐artifact‐corrupted image as the condition. A combination of mean squared error (MSE) and learned perceptual image patch similarity (LPIPS) loss were used as the objective function to train the network. To resolve the concerns of an image from the generative model such as hallucination, we used the image as a prior for a modified normalized metal artifact reduction (NMAR) process which is a well‐known analytic scheme. The modified NMAR in this work includes an automatic metal segmentation network and a secondary artifact correction network to enhance the MAR performance. Results: The proposed method showed significant outperformance over the models such as classical NMAR and a state‐of‐the‐art convolutional‐neural‐network‐based MAR (CNNMAR) approach for CBCT. The major improvements to metal artifact reduction are attributed to the improved NMAR prior estimated by the LDM. The proposed method improved the image quality compared to CNNMAR in terms of RMSE, PSNR, and SSIM from 34.78 × 10 −4 to 19.30 × 10 −4 , 49.3–54.3, and 91.2–97.2, respectively. Its clinical dental implementation was also explored and showcased success in reducing metal artifacts while preserving other tissue structures. Conclusions: The proposed method has shown its practical utility in dental CBCT for reducing metal artifacts and is believed to contribute to dental diagnosis and treatment.
- Open Access
- 19 September 2025
- Published by Public Library of Science (PLoS) in PLOS ONE
- Vol. 20 (9) , e0332603
Improving segmentation precision in prostate cancer adaptive radiation therapy with a patient-specific network
Adaptive radiotherapy (ART) enhances prostate cancer treatment by accounting for daily anatomical variations, but clinical implementation remains limited due to the need for accurate and efficient auto segmentation; manual corrections after automated contouring often hinder workflow efficiency. To address this, we propose a patient-specific network (PSN) approach for clinical target volume (CTV) segmentation using cone-beam computed tomography (CBCT). This retrospective study included 26 prostate cancer patients treated with CBCT-guided online ART using the Ethos therapy system, comprising 119 retrospectively exported fractions. The PSN framework uses a two-stage strategy: initial pre-training followed by patient-specific fine-tuning via PSNadaptive or PSNsequence, implemented with the Swin UNETR architecture. This approach is distinct from static personalization methods as it continuously adapts to daily anatomical changes. Segmentation performance was compared against deformable registration and generalized deep learning models using the Dice similarity coefficient (DSC), 95th percentile Hausdorff distance (HD), and mean surface distance (MSD). PSN significantly improved segmentation performance, with PSNadaptive achieving a DSC of 0.978 ± 0.005, HD of 1.681 ± 0.743 mm, and MSD of 0.510 ± 0.035 mm by the fifth fraction, with accuracy improving across sequential fractions. Visual assessments confirmed high agreement with physician contours, especially in anatomically complex regions. These findings support the PSN framework as a clinically feasible and accurate solution for patient-specific segmentation in prostate ART, potentially reducing the need for manual editing, streamlining workflow efficiency, and enhancing the precision of adaptive treatment delivery.
- 19 September 2025
- Published by Public Library of Science (PLoS) in PLOS ONE
- Vol. 20 (9) , e0332603
Open AccessImproving segmentation precision in prostate cancer adaptive radiation therapy with a patient-specific network
Adaptive radiotherapy (ART) enhances prostate cancer treatment by accounting for daily anatomical variations, but clinical implementation remains limited due to the need for accurate and efficient auto segmentation; manual corrections after automated contouring often hinder workflow efficiency. To address this, we propose a patient-specific network (PSN) approach for clinical target volume (CTV) segmentation using cone-beam computed tomography (CBCT). This retrospective study included 26 prostate cancer patients treated with CBCT-guided online ART using the Ethos therapy system, comprising 119 retrospectively exported fractions. The PSN framework uses a two-stage strategy: initial pre-training followed by patient-specific fine-tuning via PSNadaptive or PSNsequence, implemented with the Swin UNETR architecture. This approach is distinct from static personalization methods as it continuously adapts to daily anatomical changes. Segmentation performance was compared against deformable registration and generalized deep learning models using the Dice similarity coefficient (DSC), 95th percentile Hausdorff distance (HD), and mean surface distance (MSD). PSN significantly improved segmentation performance, with PSNadaptive achieving a DSC of 0.978 ± 0.005, HD of 1.681 ± 0.743 mm, and MSD of 0.510 ± 0.035 mm by the fifth fraction, with accuracy improving across sequential fractions. Visual assessments confirmed high agreement with physician contours, especially in anatomically complex regions. These findings support the PSN framework as a clinically feasible and accurate solution for patient-specific segmentation in prostate ART, potentially reducing the need for manual editing, streamlining workflow efficiency, and enhancing the precision of adaptive treatment delivery.
- 10 July 2025
- Published by SPIE-Intl Soc Optical Eng in Journal of Medical Imaging
- Vol. 12 (S2) , S22016-16
Can scout scan help choose a scanning angle in digital breast tomosynthesis?
PurposePrevious studies have shown that wide angular range (AR) digital breast tomosynthesis (DBT) is more effective for detecting larger but low-contrast masses, whereas narrow one for smaller but high-contrast structures. We propose a method for adaptively adjusting DBT AR based on the mass-like lesion probability predicted by a scout scan, which is performed for automatic exposure control in most DBT scanners.ApproachWe conducted a model observer analysis using anthropomorphic phantoms to assess detectability across different AR settings focusing on lesion type, breast density, and compression thickness. We developed a scout scan synthesis network initially trained on VICTRE simulation data. To reduce the domain gap, the network was further refined using paired clinical images obtained from a clinical trial. A mass classifier was then trained using synthesized scout scans derived from the open full-field digital mammography dataset.ResultsOur model observer analysis identified lesion type as the most significant factor influencing detectability. As AR switched from narrow to wide, the signal-to-noise ratio for mass detection improved from 2.538 to 3.019 (p=0.007). The mass classifier, trained with the output of the scout scan synthesis network, achieved a recall rate of 74.8%. This indicates that a significant number of mass-positive cases could be scanned using wide AR DBT and therefore would benefit from improved mass detectability.ConclusionsWe demonstrate the feasibility of lesion type–based AR adjustment in DBT offering patient-tailored imaging configuration for improved diagnostic accuracy.
- 10 July 2025
- Published by SPIE-Intl Soc Optical Eng in Journal of Medical Imaging
- Vol. 12 (S2) , S22016-16
Can scout scan help choose a scanning angle in digital breast tomosynthesis?
PurposePrevious studies have shown that wide angular range (AR) digital breast tomosynthesis (DBT) is more effective for detecting larger but low-contrast masses, whereas narrow one for smaller but high-contrast structures. We propose a method for adaptively adjusting DBT AR based on the mass-like lesion probability predicted by a scout scan, which is performed for automatic exposure control in most DBT scanners.ApproachWe conducted a model observer analysis using anthropomorphic phantoms to assess detectability across different AR settings focusing on lesion type, breast density, and compression thickness. We developed a scout scan synthesis network initially trained on VICTRE simulation data. To reduce the domain gap, the network was further refined using paired clinical images obtained from a clinical trial. A mass classifier was then trained using synthesized scout scans derived from the open full-field digital mammography dataset.ResultsOur model observer analysis identified lesion type as the most significant factor influencing detectability. As AR switched from narrow to wide, the signal-to-noise ratio for mass detection improved from 2.538 to 3.019 (p=0.007). The mass classifier, trained with the output of the scout scan synthesis network, achieved a recall rate of 74.8%. This indicates that a significant number of mass-positive cases could be scanned using wide AR DBT and therefore would benefit from improved mass detectability.ConclusionsWe demonstrate the feasibility of lesion type–based AR adjustment in DBT offering patient-tailored imaging configuration for improved diagnostic accuracy.
- 12 June 2025
- Published by SPIE-Intl Soc Optical Eng in Journal of Medical Imaging
- Vol. 12 (S2) , 1-18
Asymmetric scatter kernel estimation neural network for digital breast tomosynthesis
PurposeVarious deep learning (DL) approaches have been developed for estimating scatter radiation in digital breast tomosynthesis (DBT). Existing DL methods generally employ an end-to-end training approach, overlooking the underlying physics of scatter formation. We propose a deep learning approach inspired by asymmetric scatter kernel superposition to estimate scatter in DBT.ApproachWe use the network to generate the scatter amplitude distribution as well as the scatter kernel width and asymmetric factor map. To account for variations in local breast thickness and shape in DBT projection data, we integrated the Euclidean distance map and projection angle information into the network design for estimating the asymmetric factor.ResultsSystematic experiments on numerical phantom data and physical experimental data demonstrated the outperformance of the proposed approach to UNet-based end-to-end scatter estimation and symmetric kernel-based approaches in terms of signal-to-noise ratio and structure similarity index measure of the resulting scatter corrected images.ConclusionsThe proposed method is believed to have achieved significant advancement in scatter estimation of DBT projections, allowing a robust and reliable physics-informed scatter correction.
- 12 June 2025
- Published by SPIE-Intl Soc Optical Eng in Journal of Medical Imaging
- Vol. 12 (S2) , 1-18
Asymmetric scatter kernel estimation neural network for digital breast tomosynthesis
PurposeVarious deep learning (DL) approaches have been developed for estimating scatter radiation in digital breast tomosynthesis (DBT). Existing DL methods generally employ an end-to-end training approach, overlooking the underlying physics of scatter formation. We propose a deep learning approach inspired by asymmetric scatter kernel superposition to estimate scatter in DBT.ApproachWe use the network to generate the scatter amplitude distribution as well as the scatter kernel width and asymmetric factor map. To account for variations in local breast thickness and shape in DBT projection data, we integrated the Euclidean distance map and projection angle information into the network design for estimating the asymmetric factor.ResultsSystematic experiments on numerical phantom data and physical experimental data demonstrated the outperformance of the proposed approach to UNet-based end-to-end scatter estimation and symmetric kernel-based approaches in terms of signal-to-noise ratio and structure similarity index measure of the resulting scatter corrected images.ConclusionsThe proposed method is believed to have achieved significant advancement in scatter estimation of DBT projections, allowing a robust and reliable physics-informed scatter correction.
- 28 February 2025
- Published by Elsevier in Computers in Biology and Medicine
- Vol. 189, 109900
Sparse-view CBCT reconstruction using meta-learned neural attenuation field and hash-encoding regularization
- 28 February 2025
- Published by Elsevier in Computers in Biology and Medicine
- Vol. 189, 109900
Sparse-view CBCT reconstruction using meta-learned neural attenuation field and hash-encoding regularization
- Open Access
- 12 November 2024
- Published by IOP Publishing in Physics in Medicine & Biology
- Vol. 69 (22) , 225012
TMAA-net: tensor-domain multi-planal anti-aliasing network for sparse-view CT image reconstruction
Objective. Among various deep-network-based sparse-view CT image reconstruction studies, the sinogram upscaling network has been predominantly employed to synthesize additional view information. However, the performance of the sinogram-based network is limited in terms of removing aliasing streak artifacts and recovering low-contrast small structures. In this study, we used a view-by-view back-projection (VVBP) tensor-domain network to overcome such limitations of the sinogram-based approaches. Approach. The proposed method offers advantages of addressing the aliasing artifacts directly in the 3D tensor domain over the 2D sinogram. In the tensor-domain network, the multi-planal anti-aliasing modules were used to remove artifacts within the coronal and sagittal tensor planes. In addition, the data-fidelity-based refinement module was also implemented to successively process output images of the tensor network to recover image sharpness and textures. Main result. The proposed method showed outperformance in terms of removing aliasing artifacts and recovering low-contrast details compared to other state-of-the-art sinogram-based networks. The performance was validated for both numerical and clinical projection data in a circular fan-beam CT configuration. Significance. We observed that view-by-view aliasing artifacts in sparse-view CT exhibit distinct patterns within the tensor planes, making them effectively removable in high-dimensional representations. Additionally, we demonstrated that the co-domain characteristics of tensor space processing offer higher generalization performance for aliasing artifact removal compared to conventional sinogram-domain processing.
- 12 November 2024
- Published by IOP Publishing in Physics in Medicine & Biology
- Vol. 69 (22) , 225012
Open AccessTMAA-net: tensor-domain multi-planal anti-aliasing network for sparse-view CT image reconstruction
Objective. Among various deep-network-based sparse-view CT image reconstruction studies, the sinogram upscaling network has been predominantly employed to synthesize additional view information. However, the performance of the sinogram-based network is limited in terms of removing aliasing streak artifacts and recovering low-contrast small structures. In this study, we used a view-by-view back-projection (VVBP) tensor-domain network to overcome such limitations of the sinogram-based approaches. Approach. The proposed method offers advantages of addressing the aliasing artifacts directly in the 3D tensor domain over the 2D sinogram. In the tensor-domain network, the multi-planal anti-aliasing modules were used to remove artifacts within the coronal and sagittal tensor planes. In addition, the data-fidelity-based refinement module was also implemented to successively process output images of the tensor network to recover image sharpness and textures. Main result. The proposed method showed outperformance in terms of removing aliasing artifacts and recovering low-contrast details compared to other state-of-the-art sinogram-based networks. The performance was validated for both numerical and clinical projection data in a circular fan-beam CT configuration. Significance. We observed that view-by-view aliasing artifacts in sparse-view CT exhibit distinct patterns within the tensor planes, making them effectively removable in high-dimensional representations. Additionally, we demonstrated that the co-domain characteristics of tensor space processing offer higher generalization performance for aliasing artifact removal compared to conventional sinogram-domain processing.