Lung cancer has been one of the leading causes of death in the world for decades. An effective computer-aided diagnosis (CAD) scheme for lung nodule analysis is critical in early detection of cancerous nodules.
Purpose
This work is dedicated to the development of a CAD system based upon deep learning to predict the likelihood of nodule malignancy in lung computed tomography (CT) images.
Methods
Because lung nodules exhibit various sizes and shapes, handcrafted texture feature maps are associated with intensity CT images for network input. Ten selected texture features computed from the local binary pattern (LBP), gray-level co-occurrence matrix (GLCM), gray-level run length matrix (GLRLM), and gray-level size zone matrix (GLSZM) methods are exploited as the concatenation candidates. A new lung nodule classification framework based upon a multiplane dense inception network (MPDINet) is investigated. The proposed model takes advantage of DenseNet for feature condensation and GoogLeNet for feature extraction. Three parallel branches in the axial, coronal, and sagittal planes, including the perinodular zone reinforce nodule characterization while maintaining computation efficacy.
Results
Our MPDINet was evaluated on the public Lung Image Database Consortium and Image Database Resource Initiative (LIDC-IDRI) challenge dataset, where 1235 (802 benign and 433 malignant) nodules were selected, based upon 10-fold cross validation. The proposed model with the inverse difference moment (IDM) feature concatenation input achieved high AUC (0.9821 ± 0.0234), sensitivity (0.9426 ± 0.0979), specificity (0.9732 ± 0.0363), and precision (0.9499 ± 0.0657) rates, which demonstrated accurate lung nodule classification.
Conclusions
The developed MPDINet architecture with the handcrafted feature concatenation input is promising in many lung nodule classification applications with CT images.
{"title":"Classification of lung nodules in CT images based upon a multiplane dense inception network","authors":"Yan-Tong Wu, Herng-Hua Chang","doi":"10.1002/mp.70316","DOIUrl":"10.1002/mp.70316","url":null,"abstract":"<div>\u0000 \u0000 \u0000 <section>\u0000 \u0000 <h3> Background</h3>\u0000 \u0000 <p>Lung cancer has been one of the leading causes of death in the world for decades. An effective computer-aided diagnosis (CAD) scheme for lung nodule analysis is critical in early detection of cancerous nodules.</p>\u0000 </section>\u0000 \u0000 <section>\u0000 \u0000 <h3> Purpose</h3>\u0000 \u0000 <p>This work is dedicated to the development of a CAD system based upon deep learning to predict the likelihood of nodule malignancy in lung computed tomography (CT) images.</p>\u0000 </section>\u0000 \u0000 <section>\u0000 \u0000 <h3> Methods</h3>\u0000 \u0000 <p>Because lung nodules exhibit various sizes and shapes, handcrafted texture feature maps are associated with intensity CT images for network input. Ten selected texture features computed from the local binary pattern (LBP), gray-level co-occurrence matrix (GLCM), gray-level run length matrix (GLRLM), and gray-level size zone matrix (GLSZM) methods are exploited as the concatenation candidates. A new lung nodule classification framework based upon a multiplane dense inception network (MPDINet) is investigated. The proposed model takes advantage of DenseNet for feature condensation and GoogLeNet for feature extraction. Three parallel branches in the axial, coronal, and sagittal planes, including the perinodular zone reinforce nodule characterization while maintaining computation efficacy.</p>\u0000 </section>\u0000 \u0000 <section>\u0000 \u0000 <h3> Results</h3>\u0000 \u0000 <p>Our MPDINet was evaluated on the public Lung Image Database Consortium and Image Database Resource Initiative (LIDC-IDRI) challenge dataset, where 1235 (802 benign and 433 malignant) nodules were selected, based upon 10-fold cross validation. The proposed model with the inverse difference moment (IDM) feature concatenation input achieved high AUC (0.9821 ± 0.0234), sensitivity (0.9426 ± 0.0979), specificity (0.9732 ± 0.0363), and precision (0.9499 ± 0.0657) rates, which demonstrated accurate lung nodule classification.</p>\u0000 </section>\u0000 \u0000 <section>\u0000 \u0000 <h3> Conclusions</h3>\u0000 \u0000 <p>The developed MPDINet architecture with the handcrafted feature concatenation input is promising in many lung nodule classification applications with CT images.</p>\u0000 </section>\u0000 </div>","PeriodicalId":18384,"journal":{"name":"Medical physics","volume":"53 2","pages":""},"PeriodicalIF":3.2,"publicationDate":"2026-01-31","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146097572","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Marco Montefiori, Luca Baldini, Maria Giuseppina Bisogni, Andrea Cavalieri, Mariagrazia Celentano, Giuseppe Felici, Faustino Gómez, Leonardo Lucchesi, Matteo Morrocchi, Leonardo Orsini, Fabiola Paiar, José Paz-Martín, Carmelo Sgrò, Fabio Di Martino
<div> <section> <h3> Background</h3> <p>Ultra-high dose-per-pulse (UHDP) dosimetry remains a key challenge in FLASH radiotherapy. Conventional ionization chambers (ICs) experience severe electric field perturbations under UHDP conditions due to high charge densities, leading to severe recombination. A novel IC design, the ALLS chamber, has been proposed to overcome these limitations by using a low-pressure noble gas, eliminating ion recombination, and enabling an analytical description of charge collection up to 40 Gy/pulse with argon at 1 hPa pressure as the active medium. However, designing such an IC requires meeting both dosimetric and mechanical constraints for low-pressure operation. Since the actual requirements for FLASH dosimetry involve DPP up to 10 Gy, less extreme de-pressures in the range of 50–1000 hPa could be applied, even though such a scenario cannot be described analytically. Numerical simulations and experimental measurements are essential to explore new gas and pressure configurations.</p> </section> <section> <h3> Purpose</h3> <p>This work presents the first experimental proof-of-concept of an ALLS-based low-pressure parallel-plate IC (PPIC) for UHDP electron beam dosimetry.</p> </section> <section> <h3> Methods</h3> <p>In order to experimentally investigate the response of the chamber to variations in the filling gas and its pressure, a custom PPIC prototype was developed in a sealed Polymethyl Methacrylate (PMMA) Di vessel with controlled depressurization. Numerical simulations of the charge transport in noble or inert gases were used to predict the chamber response. Experimental measurements were performed with UHDP electron beams. The prototype was tested in air and in nitrogen at pressures in the range of 50–1000 hPa, varying the dose per pulse (DPP) up to 9.88 Gy.</p> </section> <section> <h3> Results</h3> <p>Measurements in air showed expected saturation behavior and good agreement with commercial parallel plate chambers with similar geometry, validating the basic functionality of the prototype. In nitrogen, experimental data demonstrated good agreement with simulated charge collection predictions across all tested pressures, with residuals generally within <span></span><math> <semantics> <mrow> <mo>±</mo> <mn>5</mn> <mo>%</mo> </mrow> <annotation>$pm 5 %$</annotation> </semantics></math>. The response of the system was shown to
{"title":"First experimental characterization of a low-pressure nitrogen filled parallel-plate ionization chamber for UHDP electron beam dosimetry","authors":"Marco Montefiori, Luca Baldini, Maria Giuseppina Bisogni, Andrea Cavalieri, Mariagrazia Celentano, Giuseppe Felici, Faustino Gómez, Leonardo Lucchesi, Matteo Morrocchi, Leonardo Orsini, Fabiola Paiar, José Paz-Martín, Carmelo Sgrò, Fabio Di Martino","doi":"10.1002/mp.70287","DOIUrl":"10.1002/mp.70287","url":null,"abstract":"<div>\u0000 \u0000 \u0000 <section>\u0000 \u0000 <h3> Background</h3>\u0000 \u0000 <p>Ultra-high dose-per-pulse (UHDP) dosimetry remains a key challenge in FLASH radiotherapy. Conventional ionization chambers (ICs) experience severe electric field perturbations under UHDP conditions due to high charge densities, leading to severe recombination. A novel IC design, the ALLS chamber, has been proposed to overcome these limitations by using a low-pressure noble gas, eliminating ion recombination, and enabling an analytical description of charge collection up to 40 Gy/pulse with argon at 1 hPa pressure as the active medium. However, designing such an IC requires meeting both dosimetric and mechanical constraints for low-pressure operation. Since the actual requirements for FLASH dosimetry involve DPP up to 10 Gy, less extreme de-pressures in the range of 50–1000 hPa could be applied, even though such a scenario cannot be described analytically. Numerical simulations and experimental measurements are essential to explore new gas and pressure configurations.</p>\u0000 </section>\u0000 \u0000 <section>\u0000 \u0000 <h3> Purpose</h3>\u0000 \u0000 <p>This work presents the first experimental proof-of-concept of an ALLS-based low-pressure parallel-plate IC (PPIC) for UHDP electron beam dosimetry.</p>\u0000 </section>\u0000 \u0000 <section>\u0000 \u0000 <h3> Methods</h3>\u0000 \u0000 <p>In order to experimentally investigate the response of the chamber to variations in the filling gas and its pressure, a custom PPIC prototype was developed in a sealed Polymethyl Methacrylate (PMMA) Di vessel with controlled depressurization. Numerical simulations of the charge transport in noble or inert gases were used to predict the chamber response. Experimental measurements were performed with UHDP electron beams. The prototype was tested in air and in nitrogen at pressures in the range of 50–1000 hPa, varying the dose per pulse (DPP) up to 9.88 Gy.</p>\u0000 </section>\u0000 \u0000 <section>\u0000 \u0000 <h3> Results</h3>\u0000 \u0000 <p>Measurements in air showed expected saturation behavior and good agreement with commercial parallel plate chambers with similar geometry, validating the basic functionality of the prototype. In nitrogen, experimental data demonstrated good agreement with simulated charge collection predictions across all tested pressures, with residuals generally within <span></span><math>\u0000 <semantics>\u0000 <mrow>\u0000 <mo>±</mo>\u0000 <mn>5</mn>\u0000 <mo>%</mo>\u0000 </mrow>\u0000 <annotation>$pm 5 %$</annotation>\u0000 </semantics></math>. The response of the system was shown to","PeriodicalId":18384,"journal":{"name":"Medical physics","volume":"53 2","pages":""},"PeriodicalIF":3.2,"publicationDate":"2026-01-31","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC12860524/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146097563","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Naonori Hu, Ryo Kakino, Akinori Sasaki, Mai Nojiri, Kazuhiko Akita, Syuushi Yoshikawa, Yasushi Kohigashi, Yuki Yoshino, Satoshi Takeno, Teruhito Aihara, Takushi Takata, Hiroki Tanaka, Keiji Nihei, Koji Ono
<div> <section> <h3> Background</h3> <p>Boron neutron capture therapy (BNCT) enables selective tumor irradiation by exploiting the high-linear energy transfer particles generated from neutron interactions with <sup>10</sup>B atoms. BNCT has been approved as an insurance-covered medical treatment for recurrent head and neck cancer in Japan. Unlike photon radiotherapy, neutrons that come out of the collimator have an angular distribution. Therefore, it is necessary to keep the distance between the collimator and the patient as short as possible. However, for head and neck cancer treatments, patient anatomy often limits proximity to the collimator, creating an unwanted air gap. This ultimately increases the neutron exposure to surrounding healthy tissue.</p> </section> <section> <h3> Purpose</h3> <p>To develop a freely deformable LiF-polyethylene neutron shield and assess its impact on neutron/gamma attenuation and clinical organ at-risk sparing in head and neck BNCT.</p> </section> <section> <h3> Methods</h3> <p>A freely deformable neutron shielding device was constructed using polyethylene beads loaded with lithium fluoride encapsulated in a vacuum-sealed cushion. Neutron and gamma-ray attenuation were measured in a water phantom under clinical conditions using an accelerator-based BNCT system (NeuCure®, Kansai BNCT Medical Center). Measurements were compared with a solid LiF-polyethylene block and validated through Monte Carlo–based simulations in a commercial treatment planning system. Three representative head and neck cases were further simulated to assess clinical dosimetric effects.</p> </section> <section> <h3> Results</h3> <p>The deformable shielding device reduced the thermal neutron flux by approximately 50%, compared with 60% for the solid LiF-polyethylene block. Simulated head and neck treatments demonstrated significant OAR dose reductions (up to 46.6% in pharyngeal mucosa <i>D</i><sub>50%</sub>) without compromising tumor dose coverage (<i>D</i><sub>80%</sub> ≥ 20 Gy-eq). Treatment delivery times were minimally affected (< 2 min difference) across all plans. A 5 mm positional perturbation analysis showed ≤ 0.5 Gy-eq variation in GTV <i>D</i><sub>min</sub> and pharyngeal mucosa <i>D</i><sub>50</sub> and <i>D</i><sub>max</sub>.</p> </section> <section> <h3> Conclusions</h3> <p>The freely deformable LiF-based neutron shielding device effectively attenuated stray neutron dose while maintaining target coverage in BNCT. Its adaptability a
{"title":"Development and dosimetric evaluation of a freely deformable 6Li-based neutron shield for boron neutron capture therapy","authors":"Naonori Hu, Ryo Kakino, Akinori Sasaki, Mai Nojiri, Kazuhiko Akita, Syuushi Yoshikawa, Yasushi Kohigashi, Yuki Yoshino, Satoshi Takeno, Teruhito Aihara, Takushi Takata, Hiroki Tanaka, Keiji Nihei, Koji Ono","doi":"10.1002/mp.70319","DOIUrl":"10.1002/mp.70319","url":null,"abstract":"<div>\u0000 \u0000 \u0000 <section>\u0000 \u0000 <h3> Background</h3>\u0000 \u0000 <p>Boron neutron capture therapy (BNCT) enables selective tumor irradiation by exploiting the high-linear energy transfer particles generated from neutron interactions with <sup>10</sup>B atoms. BNCT has been approved as an insurance-covered medical treatment for recurrent head and neck cancer in Japan. Unlike photon radiotherapy, neutrons that come out of the collimator have an angular distribution. Therefore, it is necessary to keep the distance between the collimator and the patient as short as possible. However, for head and neck cancer treatments, patient anatomy often limits proximity to the collimator, creating an unwanted air gap. This ultimately increases the neutron exposure to surrounding healthy tissue.</p>\u0000 </section>\u0000 \u0000 <section>\u0000 \u0000 <h3> Purpose</h3>\u0000 \u0000 <p>To develop a freely deformable LiF-polyethylene neutron shield and assess its impact on neutron/gamma attenuation and clinical organ at-risk sparing in head and neck BNCT.</p>\u0000 </section>\u0000 \u0000 <section>\u0000 \u0000 <h3> Methods</h3>\u0000 \u0000 <p>A freely deformable neutron shielding device was constructed using polyethylene beads loaded with lithium fluoride encapsulated in a vacuum-sealed cushion. Neutron and gamma-ray attenuation were measured in a water phantom under clinical conditions using an accelerator-based BNCT system (NeuCure®, Kansai BNCT Medical Center). Measurements were compared with a solid LiF-polyethylene block and validated through Monte Carlo–based simulations in a commercial treatment planning system. Three representative head and neck cases were further simulated to assess clinical dosimetric effects.</p>\u0000 </section>\u0000 \u0000 <section>\u0000 \u0000 <h3> Results</h3>\u0000 \u0000 <p>The deformable shielding device reduced the thermal neutron flux by approximately 50%, compared with 60% for the solid LiF-polyethylene block. Simulated head and neck treatments demonstrated significant OAR dose reductions (up to 46.6% in pharyngeal mucosa <i>D</i><sub>50%</sub>) without compromising tumor dose coverage (<i>D</i><sub>80%</sub> ≥ 20 Gy-eq). Treatment delivery times were minimally affected (< 2 min difference) across all plans. A 5 mm positional perturbation analysis showed ≤ 0.5 Gy-eq variation in GTV <i>D</i><sub>min</sub> and pharyngeal mucosa <i>D</i><sub>50</sub> and <i>D</i><sub>max</sub>.</p>\u0000 </section>\u0000 \u0000 <section>\u0000 \u0000 <h3> Conclusions</h3>\u0000 \u0000 <p>The freely deformable LiF-based neutron shielding device effectively attenuated stray neutron dose while maintaining target coverage in BNCT. Its adaptability a","PeriodicalId":18384,"journal":{"name":"Medical physics","volume":"53 2","pages":""},"PeriodicalIF":3.2,"publicationDate":"2026-01-31","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC12860538/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146097545","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Maximilian E. Lochschmidt, Theresa Urban, Lennard Kaster, Rafael C. Schick, Thomas Koehler, Daniela Pfeiffer, Franz Pfeiffer
<div> <section> <h3> Background</h3> <p>The microstructure of material at a <span></span><math> <semantics> <mrow> <mspace></mspace> <mi>μ</mi> <mi>m</mi> </mrow> <annotation>$,umumathrm{m}$</annotation> </semantics></math> length scale leads to ultra-small-angle scattering of X-rays, which typically occurs, e.g., for lung tissue or some plastic foams. When using an interferometer, this effect alters the visibility of the fringe pattern, which can be detected and resolved by the detector. Thus, the ultra-small-angle scattering can be represented as a dark-field image. For a polychromatic source, the hardening of the source spectrum changes visibility as well, generating an additional fake dark-field signal by the attenuation of the material on top of the real ultra-small-angle scatter-related dark-field signal. Consequently, even homogeneous materials without microstructure typically exhibit a change in visibility.</p> </section> <section> <h3> Purpose</h3> <p>The objective of this study is to develop a fast, simple, and robust method to correct dark-field signals and bony structures present due to beam hardening on dark-field chest radiographs of study participants.</p> </section> <section> <h3> Methods</h3> <p>The method is based on calibration measurements and image processing. BH by bones and soft tissue is modeled by aluminum and water, respectively, which have no microstructure and thus only generate an artificial dark-field signal. Look-up tables were then created for both. By using a weighted mean of these, forming a single LUT, and using the attenuation images, the artificial dark-field signal and thus the bone structures present are reduced for study participants.</p> </section> <section> <h3> Results</h3> <p>It was found that applying a correction using a weighted LUT leads to a significant reduction of bone structures in the dark-field image. The weighting of the aluminum component has a substantial impact on the degree to which bone structures remain visible in the dark-field image. Furthermore, a large negative bias in the dark-field image–dependent on the aluminum weighting–was successfully corrected.</p> </section> <section> <h3> Conclusions</h3> <p>BH-induced signal in the dark-field images was successfully reduced using the method describe
背景：材料在μ m $,umu mathm {m}$长度尺度上的微观结构导致x射线的超小角度散射，这通常发生在肺组织或某些塑料泡沫中。当使用干涉仪时，这种效应改变了条纹图案的可见性，这可以被检测器检测和解决。因此，超小角度散射可以表示为暗场图像。对于多色光源，源光谱的硬化也会改变可见性，在真实的超小角散射相关的暗场信号之上，通过材料的衰减产生额外的假暗场信号。因此，即使是没有微观结构的均匀材料通常也会表现出可见性的变化。目的：本研究的目的是开发一种快速、简单、可靠的方法来纠正研究参与者的暗场胸片上由于光束硬化而出现的暗场信号和骨结构。方法：基于标定测量和图像处理的方法。骨骼和软组织的BH分别由铝和水模拟，铝和水没有微观结构，因此只能产生人工暗场信号。然后为两者创建了查询表。通过使用这些的加权平均值，形成一个单一的LUT，并使用衰减图像，研究参与者的人工暗场信号和存在的骨结构被减少。结果：发现使用加权LUT进行校正导致暗场图像中骨结构的显着减少。铝组件的重量对骨骼结构在暗场图像中保持可见的程度有重大影响。此外，一个大的负偏差在暗场图像-依赖于铝的重量-被成功地纠正。结论：该方法可有效地降低暗场图像中的bh诱导信号。应根据具体的临床问题，对铝加重抑制肋骨结构的选择，以及偏置矫正的选择进行评估。
{"title":"Look-up table correction for beam hardening-induced signal of clinical dark-field chest radiographs","authors":"Maximilian E. Lochschmidt, Theresa Urban, Lennard Kaster, Rafael C. Schick, Thomas Koehler, Daniela Pfeiffer, Franz Pfeiffer","doi":"10.1002/mp.70307","DOIUrl":"10.1002/mp.70307","url":null,"abstract":"<div>\u0000 \u0000 \u0000 <section>\u0000 \u0000 <h3> Background</h3>\u0000 \u0000 <p>The microstructure of material at a <span></span><math>\u0000 <semantics>\u0000 <mrow>\u0000 <mspace></mspace>\u0000 <mi>μ</mi>\u0000 <mi>m</mi>\u0000 </mrow>\u0000 <annotation>$,umumathrm{m}$</annotation>\u0000 </semantics></math> length scale leads to ultra-small-angle scattering of X-rays, which typically occurs, e.g., for lung tissue or some plastic foams. When using an interferometer, this effect alters the visibility of the fringe pattern, which can be detected and resolved by the detector. Thus, the ultra-small-angle scattering can be represented as a dark-field image. For a polychromatic source, the hardening of the source spectrum changes visibility as well, generating an additional fake dark-field signal by the attenuation of the material on top of the real ultra-small-angle scatter-related dark-field signal. Consequently, even homogeneous materials without microstructure typically exhibit a change in visibility.</p>\u0000 </section>\u0000 \u0000 <section>\u0000 \u0000 <h3> Purpose</h3>\u0000 \u0000 <p>The objective of this study is to develop a fast, simple, and robust method to correct dark-field signals and bony structures present due to beam hardening on dark-field chest radiographs of study participants.</p>\u0000 </section>\u0000 \u0000 <section>\u0000 \u0000 <h3> Methods</h3>\u0000 \u0000 <p>The method is based on calibration measurements and image processing. BH by bones and soft tissue is modeled by aluminum and water, respectively, which have no microstructure and thus only generate an artificial dark-field signal. Look-up tables were then created for both. By using a weighted mean of these, forming a single LUT, and using the attenuation images, the artificial dark-field signal and thus the bone structures present are reduced for study participants.</p>\u0000 </section>\u0000 \u0000 <section>\u0000 \u0000 <h3> Results</h3>\u0000 \u0000 <p>It was found that applying a correction using a weighted LUT leads to a significant reduction of bone structures in the dark-field image. The weighting of the aluminum component has a substantial impact on the degree to which bone structures remain visible in the dark-field image. Furthermore, a large negative bias in the dark-field image–dependent on the aluminum weighting–was successfully corrected.</p>\u0000 </section>\u0000 \u0000 <section>\u0000 \u0000 <h3> Conclusions</h3>\u0000 \u0000 <p>BH-induced signal in the dark-field images was successfully reduced using the method describe","PeriodicalId":18384,"journal":{"name":"Medical physics","volume":"53 2","pages":""},"PeriodicalIF":3.2,"publicationDate":"2026-01-31","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC12860541/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146097683","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Kuo Men, Bining Yang, Yuxiang Liu, Yuan Tang, Ningning Lu, Jianrong Dai
Background
Segmentation is the most effort-consuming step for magnetic resonance imaging guided adaptive radiotherapy (MRIgART). Although the segment anything model (SAM) exhibits impressive capabilities, its application in medical imaging necessitates clicks, bounding boxes, or providing mask prompts on each target image, which would still require complex human interactions.
Purpose
This study introduces SAM-ART, a large vision model that integrates personalized information to enhance the segmentation accuracy of MRIgART.
Methods
This study utilized planning computed tomography (pCT), approved contours, and daily MRI (dMRI) from 38 patients with prostate cancer and 10 patients with rectal cancer. SAM-ART comprises an image encoder, a prompt encoder, and a mask decoder. Using mask and box prompts, SAM-ART propagates contours from pCT to dMRI using deformable image registration (DIR) and employs them as mask prompts, providing patient-specific information. The box prompts are used in slices prone to false negative (FN) predictions. A 5-fold cross-validation was then conducted, comparing SAM-ART with DIR, traditional deep learning (tDL), and SAM-ART using other manual prompts (point or box).
Results
The proposed SAM-ART exhibited a mean dice similarity coefficient of 0.934 ± 0.023 for the regions of interest, surpassing DIR (0.873 ± 0.063) and tDL (0.887 ± 0.056). Moreover, the proposed mask/box prompts also outperformed the other modes (point: 0.910 ± 0.027, and box: 0.921 ± 0.025). Mask/box prompts effectively mitigated FN predictions with minimal manual intervention. The ratio of acceptable slices (using the criteria of dice ≥ 0.85, 95th percentile of Hausdorff distance ≤ 5 mm, and mean distance to agreement ≤ 1.5 mm) was 89.38% with the proposed method, which means that the segmentations on about 90% of the slices did not require manual modification.
Conclusions
This study proposed a novel method that integrates personalized information and manual prompts into a SAM-based segmentation model. It outperformed the baseline methods, with only a few contours needing to be revised for clinical use.
{"title":"A segmentation method with a large vision model for magnetic resonance imaging-guided adaptive radiotherapy","authors":"Kuo Men, Bining Yang, Yuxiang Liu, Yuan Tang, Ningning Lu, Jianrong Dai","doi":"10.1002/mp.70257","DOIUrl":"10.1002/mp.70257","url":null,"abstract":"<div>\u0000 \u0000 \u0000 <section>\u0000 \u0000 <h3> Background</h3>\u0000 \u0000 <p>Segmentation is the most effort-consuming step for magnetic resonance imaging guided adaptive radiotherapy (MRIgART). Although the segment anything model (SAM) exhibits impressive capabilities, its application in medical imaging necessitates clicks, bounding boxes, or providing mask prompts on each target image, which would still require complex human interactions.</p>\u0000 </section>\u0000 \u0000 <section>\u0000 \u0000 <h3> Purpose</h3>\u0000 \u0000 <p>This study introduces SAM-ART, a large vision model that integrates personalized information to enhance the segmentation accuracy of MRIgART.</p>\u0000 </section>\u0000 \u0000 <section>\u0000 \u0000 <h3> Methods</h3>\u0000 \u0000 <p>This study utilized planning computed tomography (pCT), approved contours, and daily MRI (dMRI) from 38 patients with prostate cancer and 10 patients with rectal cancer. SAM-ART comprises an image encoder, a prompt encoder, and a mask decoder. Using mask and box prompts, SAM-ART propagates contours from pCT to dMRI using deformable image registration (DIR) and employs them as mask prompts, providing patient-specific information. The box prompts are used in slices prone to false negative (FN) predictions. A 5-fold cross-validation was then conducted, comparing SAM-ART with DIR, traditional deep learning (tDL), and SAM-ART using other manual prompts (point or box).</p>\u0000 </section>\u0000 \u0000 <section>\u0000 \u0000 <h3> Results</h3>\u0000 \u0000 <p>The proposed SAM-ART exhibited a mean dice similarity coefficient of 0.934 ± 0.023 for the regions of interest, surpassing DIR (0.873 ± 0.063) and tDL (0.887 ± 0.056). Moreover, the proposed mask/box prompts also outperformed the other modes (point: 0.910 ± 0.027, and box: 0.921 ± 0.025). Mask/box prompts effectively mitigated FN predictions with minimal manual intervention. The ratio of acceptable slices (using the criteria of dice ≥ 0.85, 95th percentile of Hausdorff distance ≤ 5 mm, and mean distance to agreement ≤ 1.5 mm) was 89.38% with the proposed method, which means that the segmentations on about 90% of the slices did not require manual modification.</p>\u0000 </section>\u0000 \u0000 <section>\u0000 \u0000 <h3> Conclusions</h3>\u0000 \u0000 <p>This study proposed a novel method that integrates personalized information and manual prompts into a SAM-based segmentation model. It outperformed the baseline methods, with only a few contours needing to be revised for clinical use.</p>\u0000 </section>\u0000 </div>","PeriodicalId":18384,"journal":{"name":"Medical physics","volume":"53 2","pages":""},"PeriodicalIF":3.2,"publicationDate":"2026-01-30","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146088702","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Zhao Peng, Chunbo Liu, Du Tang, Shuzhou Li, Xiaoyu Yang, Qigang Shao, Ying Cao, Yuchen Song, Wanli Huo, Zhen Yang
<div> <section> <h3> Background</h3> <p>In computed tomography (CT)-guided cervical cancer brachytherapy, the manual contouring for the high-risk clinical target volume (HR-CTV) is a time-consuming and expertise-dependent process. Furthermore, automated approaches struggle with ambiguous boundaries of HR-CTV.</p> </section> <section> <h3> Purpose</h3> <p>We aimed to develop a clinically efficient interactive segmentation framework integrating deep learning with clinician expertise.</p> </section> <section> <h3> Methods and materials</h3> <p>We propose a slice-prompted interactive segmentation method (SPSeg) for HR-CTV delineation in CT-guided cervical cancer brachytherapy. Clinicians provided sparse prompts by manually outlining HR-CTV on key slices, which were then encoded into a 3D U-Net architecture to guide full-volume segmentation. We investigated two architectural variants: SPSeg-Mono, which jointly processes the CT images and the prompt masks with a single encoder; and SPSeg-Dual, which employs two separate encoders for image and prompt, fusing their features at a deeper level. The model was trained on 640 CT scans (from 160 patients) and validated on 160 scans (40 patients) from a single center, and externally tested on three multi-center cohorts: 400 scans (100 patients), 115 scans (40 patients), and 150 scans (30 patients), respectively. Evaluation included Dice Similarity Coefficient (DSC), 95% Hausdorff Distance (HD95), a 5-point Likert scale for clinical acceptability, time efficiency, and inter-observer agreement.</p> </section> <section> <h3> Results</h3> <p>Performance consistently improved with the addition of prompt slices, with SPSeg-Dual outperforming SPSeg-Mono. Without prompts, the model yielded DSCs of 0.83, 0.76, and 0.76, and HD95s of 7.5, 10.1, and 11.6 mm for Test Sets 1, 2, and 3, respectively. With the addition of just three prompt slices, DSCs increased significantly to 0.95, 0.92, and 0.91, while HD95s decreased to 2.1, 3.1, and 3.2 mm, respectively (all <i>p</i> < 0.001). Qualitative scores confirmed high clinical acceptability (mean Likert scores > 3), and the interactive method substantially reduced contouring time for both clinicians (from 11.7 to 1.7 min for Clinician A, and from 9.9 to 1.5 min for Clinician B). It also improved inter-observer agreement, with DSC increasing from 0.88 to 0.93 and HD95 decreasing from 3.2 to 2.5 mm (<i>p</i> < 0.001).</p> </section> <section> <h3> Conclusions</h3> <p>The proposed SPSeg method effectively integrates clinical expertise with deep learning, offering a highly prec
背景：在计算机断层扫描（CT）引导下的宫颈癌近距离放射治疗中，高危临床靶体积（HR-CTV）的人工轮廓是一个耗时且依赖专业知识的过程。此外，自动化方法与HR-CTV的模糊边界作斗争。目的：我们的目标是开发一个临床有效的交互式分割框架，将深度学习与临床医生的专业知识相结合。方法和材料：我们提出了一种用于ct引导下宫颈癌近距离放疗HR-CTV划定的切片提示交互式分割方法（SPSeg）。临床医生通过手动勾勒关键切片上的HR-CTV来提供稀疏提示，然后将其编码为3D U-Net架构，以指导全体积分割。我们研究了两种架构变体：SPSeg-Mono，它用单个编码器联合处理CT图像和提示掩码；SPSeg-Dual采用两个独立的图像和提示编码器，在更深层次上融合了它们的特征。该模型在来自160名患者的640次CT扫描上进行了训练，并在来自单个中心的160次扫描（40名患者）上进行了验证，并在三个多中心队列中进行了外部测试：400次扫描（100名患者），115次扫描（40名患者）和150次扫描（30名患者）。评估包括骰子相似系数（DSC）、95%豪斯多夫距离（HD95）、临床可接受性、时间效率和观察者间一致性的5点李克特量表。结果：随着提示片的加入，性能不断提高，SPSeg-Dual优于SPSeg-Mono。在没有提示的情况下，对于测试集1、2和3，模型产生的dsc分别为0.83、0.76和0.76，hd95分别为7.5、10.1和11.6 mm。仅增加三个提示切片，dsc显著增加到0.95、0.92和0.91，而hd95分别下降到2.1、3.1和3.2 mm（均为p 3），交互式方法大大减少了两位临床医生的轮廓时间（临床医生A从11.7分钟减少到1.7分钟，临床医生B从9.9分钟减少到1.5分钟）。该方法还提高了观察者间的一致性，DSC从0.88增加到0.93，HD95从3.2下降到2.5 mm (p)。结论：提出的SPSeg方法有效地将临床专业知识与深度学习相结合，为宫颈癌近距离治疗的HR-CTV划定提供了高度精确和高效的解决方案。
{"title":"Slice-prompted HR-CTV interactive segmentation for cervical cancer brachytherapy: A multi-center study","authors":"Zhao Peng, Chunbo Liu, Du Tang, Shuzhou Li, Xiaoyu Yang, Qigang Shao, Ying Cao, Yuchen Song, Wanli Huo, Zhen Yang","doi":"10.1002/mp.70305","DOIUrl":"10.1002/mp.70305","url":null,"abstract":"<div>\u0000 \u0000 \u0000 <section>\u0000 \u0000 <h3> Background</h3>\u0000 \u0000 <p>In computed tomography (CT)-guided cervical cancer brachytherapy, the manual contouring for the high-risk clinical target volume (HR-CTV) is a time-consuming and expertise-dependent process. Furthermore, automated approaches struggle with ambiguous boundaries of HR-CTV.</p>\u0000 </section>\u0000 \u0000 <section>\u0000 \u0000 <h3> Purpose</h3>\u0000 \u0000 <p>We aimed to develop a clinically efficient interactive segmentation framework integrating deep learning with clinician expertise.</p>\u0000 </section>\u0000 \u0000 <section>\u0000 \u0000 <h3> Methods and materials</h3>\u0000 \u0000 <p>We propose a slice-prompted interactive segmentation method (SPSeg) for HR-CTV delineation in CT-guided cervical cancer brachytherapy. Clinicians provided sparse prompts by manually outlining HR-CTV on key slices, which were then encoded into a 3D U-Net architecture to guide full-volume segmentation. We investigated two architectural variants: SPSeg-Mono, which jointly processes the CT images and the prompt masks with a single encoder; and SPSeg-Dual, which employs two separate encoders for image and prompt, fusing their features at a deeper level. The model was trained on 640 CT scans (from 160 patients) and validated on 160 scans (40 patients) from a single center, and externally tested on three multi-center cohorts: 400 scans (100 patients), 115 scans (40 patients), and 150 scans (30 patients), respectively. Evaluation included Dice Similarity Coefficient (DSC), 95% Hausdorff Distance (HD95), a 5-point Likert scale for clinical acceptability, time efficiency, and inter-observer agreement.</p>\u0000 </section>\u0000 \u0000 <section>\u0000 \u0000 <h3> Results</h3>\u0000 \u0000 <p>Performance consistently improved with the addition of prompt slices, with SPSeg-Dual outperforming SPSeg-Mono. Without prompts, the model yielded DSCs of 0.83, 0.76, and 0.76, and HD95s of 7.5, 10.1, and 11.6 mm for Test Sets 1, 2, and 3, respectively. With the addition of just three prompt slices, DSCs increased significantly to 0.95, 0.92, and 0.91, while HD95s decreased to 2.1, 3.1, and 3.2 mm, respectively (all <i>p</i> < 0.001). Qualitative scores confirmed high clinical acceptability (mean Likert scores > 3), and the interactive method substantially reduced contouring time for both clinicians (from 11.7 to 1.7 min for Clinician A, and from 9.9 to 1.5 min for Clinician B). It also improved inter-observer agreement, with DSC increasing from 0.88 to 0.93 and HD95 decreasing from 3.2 to 2.5 mm (<i>p</i> < 0.001).</p>\u0000 </section>\u0000 \u0000 <section>\u0000 \u0000 <h3> Conclusions</h3>\u0000 \u0000 <p>The proposed SPSeg method effectively integrates clinical expertise with deep learning, offering a highly prec","PeriodicalId":18384,"journal":{"name":"Medical physics","volume":"53 2","pages":""},"PeriodicalIF":3.2,"publicationDate":"2026-01-30","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146088728","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pengyu Chen, Zixue Zeng, Xiaoyan Zhao, Xin Meng, Zhiyu Sheng, Maryam Satapour, John M. Cormack, Allison C. Bean, Ryan Nussbaum, Maya Maurer, Emily Landis-Walkenhorst, Kang Kim, Ajay D. Wasan, Jiantao Pu
<div> <section> <h3> Background</h3> <p>Chronic low back pain (cLBP) profoundly impacts quality of life, yet its underlying mechanisms remain poorly understood. Three-dimensional (3D) ultrasound imaging offers valuable insights into cLBP but poses challenges of manual annotation due to large data volume and poor image quality.</p> </section> <section> <h3> Objectives</h3> <p>We aim to develop and validate a novel approach called Cross-Scan Fusion Network (CSFN) for segmenting anatomical tissue layers in 3D ultrasound images.</p> </section> <section> <h3> Materials and Methods</h3> <p>We analyzed 3D B-mode ultrasound volumes of the lumbar region, with six tissue layers annotated: dermis, superficial fat, superficial fascial membrane, deep fat, deep fascial membrane, and muscle. The dataset included 69 labeled scans from 29 subjects and 30 unlabeled scans from 10 subjects. Labeled scans were split into training (<i>n =</i> 19), validation (<i>n =</i> 10), and independent test (<i>n =</i> 40) sets. For annotation efficiency analysis, 5, 10, 15, and 19 scans in the training set were separately treated as annotated, with the remainder considered unannotated. CSFN leverages a VoxelMorph-style elastic registration network trained with a novel Projected Hausdorff Distance Loss (PHDL) to accurately register anatomical tissue layers across 3D scans. This supports both sample-efficient learning (CSFN-SEL) and semi-supervised learning (CSFN-SSL). In the CSFN-SEL, controlled wrap ratios are applied to pairs of labeled scans to synthesize realistic image-mask pairs, while in the CSFN-SSL, labeled scans are registered onto unlabeled scans to generate high-quality synthesized scans for augmentation. Both real and synthetic data are then combined to train an nnU-Net segmentor, enabling robust segmentation with minimal manual annotations. CSFN was compared to fully-supervised nnU-Net with and without augmentation, and SimCLR (nnU-Net backbone). Model segmentation performance was evaluated using the Dice coefficient, while registration quality was assessed using Dice and Average Symmetric Surface Distance (ASSD). Results were reported as mean ± standard deviation(SD) and compared using paired two-tailed t-tests on class-wise subject averages. Significance was set at 0.05 and adjusted for multiple comparisons.</p> </section> <section> <h3> Results</h3> <p>CSFN-SEL consistently outperformed the fully supervised nnU-Net baseline across varying numbers of labeled training samples, improving the mean Dice coefficient from 69.34% to 74.62% (+5.28%, <i>p</i>-value < 0.05/3). CSFN-SSL further improved performance, achieving 79.33% (±1.96%) and 81.14% (±1
{"title":"Cross-scan fusion network: A registration-based framework for annotation-efficient 3D ultrasound segmentation in low back pain assessment","authors":"Pengyu Chen, Zixue Zeng, Xiaoyan Zhao, Xin Meng, Zhiyu Sheng, Maryam Satapour, John M. Cormack, Allison C. Bean, Ryan Nussbaum, Maya Maurer, Emily Landis-Walkenhorst, Kang Kim, Ajay D. Wasan, Jiantao Pu","doi":"10.1002/mp.70304","DOIUrl":"10.1002/mp.70304","url":null,"abstract":"<div>\u0000 \u0000 \u0000 <section>\u0000 \u0000 <h3> Background</h3>\u0000 \u0000 <p>Chronic low back pain (cLBP) profoundly impacts quality of life, yet its underlying mechanisms remain poorly understood. Three-dimensional (3D) ultrasound imaging offers valuable insights into cLBP but poses challenges of manual annotation due to large data volume and poor image quality.</p>\u0000 </section>\u0000 \u0000 <section>\u0000 \u0000 <h3> Objectives</h3>\u0000 \u0000 <p>We aim to develop and validate a novel approach called Cross-Scan Fusion Network (CSFN) for segmenting anatomical tissue layers in 3D ultrasound images.</p>\u0000 </section>\u0000 \u0000 <section>\u0000 \u0000 <h3> Materials and Methods</h3>\u0000 \u0000 <p>We analyzed 3D B-mode ultrasound volumes of the lumbar region, with six tissue layers annotated: dermis, superficial fat, superficial fascial membrane, deep fat, deep fascial membrane, and muscle. The dataset included 69 labeled scans from 29 subjects and 30 unlabeled scans from 10 subjects. Labeled scans were split into training (<i>n =</i> 19), validation (<i>n =</i> 10), and independent test (<i>n =</i> 40) sets. For annotation efficiency analysis, 5, 10, 15, and 19 scans in the training set were separately treated as annotated, with the remainder considered unannotated. CSFN leverages a VoxelMorph-style elastic registration network trained with a novel Projected Hausdorff Distance Loss (PHDL) to accurately register anatomical tissue layers across 3D scans. This supports both sample-efficient learning (CSFN-SEL) and semi-supervised learning (CSFN-SSL). In the CSFN-SEL, controlled wrap ratios are applied to pairs of labeled scans to synthesize realistic image-mask pairs, while in the CSFN-SSL, labeled scans are registered onto unlabeled scans to generate high-quality synthesized scans for augmentation. Both real and synthetic data are then combined to train an nnU-Net segmentor, enabling robust segmentation with minimal manual annotations. CSFN was compared to fully-supervised nnU-Net with and without augmentation, and SimCLR (nnU-Net backbone). Model segmentation performance was evaluated using the Dice coefficient, while registration quality was assessed using Dice and Average Symmetric Surface Distance (ASSD). Results were reported as mean ± standard deviation(SD) and compared using paired two-tailed t-tests on class-wise subject averages. Significance was set at 0.05 and adjusted for multiple comparisons.</p>\u0000 </section>\u0000 \u0000 <section>\u0000 \u0000 <h3> Results</h3>\u0000 \u0000 <p>CSFN-SEL consistently outperformed the fully supervised nnU-Net baseline across varying numbers of labeled training samples, improving the mean Dice coefficient from 69.34% to 74.62% (+5.28%, <i>p</i>-value < 0.05/3). CSFN-SSL further improved performance, achieving 79.33% (±1.96%) and 81.14% (±1","PeriodicalId":18384,"journal":{"name":"Medical physics","volume":"53 2","pages":""},"PeriodicalIF":3.2,"publicationDate":"2026-01-30","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146088745","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Nathan Shaffer, Avinash Reddy Mudireddy, Joel St-Aubin
<div> <section> <h3> Background</h3> <p>Volumetric modulated arc therapy (VMAT) machine parameter optimization (MPO) is a complex, high-dimensional problem typically solved with inverse planning solutions that are both temporally and computationally expensive. While machine learning techniques have been explored to automate this process, they often supplement rather than replace conventional optimizers and are fundamentally limited by the quality and diversity of training data. Reinforcement learning (RL) offers a promising alternative, finding optimal strategies through trial-and-error by maximizing a narrowly tailored reward function, which can potentially discover novel solutions beyond mimicking features present in existing plans.</p> </section> <section> <h3> Purpose</h3> <p>The purpose of this study was to develop and validate a deep reinforcement learning-based VMAT MPO algorithm capable of automatically generating clinically comparable treatment plans for prostate cancer that meet machine constraints, entirely independent of a commercial treatment planning system (TPS) optimizer.</p> </section> <section> <h3> Methods</h3> <p>A dataset comprised of 100 prostate cancer patients planned using the criteria from PACE-B SBRT arm serve as the basis for network training using a 70-10-20 training/validation/testing split. An RL framework using a Proximal Policy Optimization (PPO) algorithm was developed to train two tandem convolutional neural networks that sequentially optimize multi-leaf collimator (MLC) positions and monitor units (MUs) using current dose, contoured structure masks, and current machine parameters as inputs. Training was designed to predict MLC positions and MUs that maximize a dose-volume histogram (DVH)-based reward function tailored to prioritize meeting clinical objectives. The fully trained networks were executed on a test set of 20 patients and compared to reference plans optimized with a commercial TPS.</p> </section> <section> <h3> Results</h3> <p>The RL algorithm generated plans in an average of 6.3 ± 4.7 s. Compared to the reference plans, the RL-generated plans demonstrated improved sparing for both the bladder and rectum across their respective dosimetric endpoints. When normalizing to 95% coverage, the RL generated plans resulted in a statistically significant increase in the PTV <span></span><math> <semantics> <msub> <mi>D</mi> <mrow> <mn>2</mn> <mo>%</mo>
{"title":"A tandem reinforcement learning framework for localized prostate cancer treatment planning and machine parameter optimization","authors":"Nathan Shaffer, Avinash Reddy Mudireddy, Joel St-Aubin","doi":"10.1002/mp.70306","DOIUrl":"10.1002/mp.70306","url":null,"abstract":"<div>\u0000 \u0000 \u0000 <section>\u0000 \u0000 <h3> Background</h3>\u0000 \u0000 <p>Volumetric modulated arc therapy (VMAT) machine parameter optimization (MPO) is a complex, high-dimensional problem typically solved with inverse planning solutions that are both temporally and computationally expensive. While machine learning techniques have been explored to automate this process, they often supplement rather than replace conventional optimizers and are fundamentally limited by the quality and diversity of training data. Reinforcement learning (RL) offers a promising alternative, finding optimal strategies through trial-and-error by maximizing a narrowly tailored reward function, which can potentially discover novel solutions beyond mimicking features present in existing plans.</p>\u0000 </section>\u0000 \u0000 <section>\u0000 \u0000 <h3> Purpose</h3>\u0000 \u0000 <p>The purpose of this study was to develop and validate a deep reinforcement learning-based VMAT MPO algorithm capable of automatically generating clinically comparable treatment plans for prostate cancer that meet machine constraints, entirely independent of a commercial treatment planning system (TPS) optimizer.</p>\u0000 </section>\u0000 \u0000 <section>\u0000 \u0000 <h3> Methods</h3>\u0000 \u0000 <p>A dataset comprised of 100 prostate cancer patients planned using the criteria from PACE-B SBRT arm serve as the basis for network training using a 70-10-20 training/validation/testing split. An RL framework using a Proximal Policy Optimization (PPO) algorithm was developed to train two tandem convolutional neural networks that sequentially optimize multi-leaf collimator (MLC) positions and monitor units (MUs) using current dose, contoured structure masks, and current machine parameters as inputs. Training was designed to predict MLC positions and MUs that maximize a dose-volume histogram (DVH)-based reward function tailored to prioritize meeting clinical objectives. The fully trained networks were executed on a test set of 20 patients and compared to reference plans optimized with a commercial TPS.</p>\u0000 </section>\u0000 \u0000 <section>\u0000 \u0000 <h3> Results</h3>\u0000 \u0000 <p>The RL algorithm generated plans in an average of 6.3 ± 4.7 s. Compared to the reference plans, the RL-generated plans demonstrated improved sparing for both the bladder and rectum across their respective dosimetric endpoints. When normalizing to 95% coverage, the RL generated plans resulted in a statistically significant increase in the PTV <span></span><math>\u0000 <semantics>\u0000 <msub>\u0000 <mi>D</mi>\u0000 <mrow>\u0000 <mn>2</mn>\u0000 <mo>%</mo>\u0000 ","PeriodicalId":18384,"journal":{"name":"Medical physics","volume":"53 2","pages":""},"PeriodicalIF":3.2,"publicationDate":"2026-01-30","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC12857532/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146088768","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
<div> <section> <h3> Background</h3> <p>It has been shown that rupture of vulnerable plaques of atherosclerosis (AS) is one of the main causes of ischemic stroke. Increasing evidence suggests that the inflammatory changes of perivascular adipose tissue (PVAT) were independently associated with both vulnerable plaque characteristics and cerebrovascular symptoms. Therefore, it is essential to conduct non-invasive joint assessment of carotid AS plaques and PVAT characteristics for better stratifying the risk of ischemic cerebrovascular events.</p> </section> <section> <h3> Purpose</h3> <p>To develop a simultaneous magnetic resonance (MR) imaging technique for carotid artery PVAT and vessel wall, and determine its feasibility and repeatability.</p> </section> <section> <h3> Methods</h3> <p>This study developed an MR sequence for simultaneous imaging carotid vessel wall and PVAT. Seventeen healthy subjects and nine patients with carotid AS were recruited for MR imaging experiments, of whom five healthy subjects were selected for the repeatability test. All participants underwent bilateral carotid three-dimensional MR imaging by acquiring the proposed iMSDE-mDIXON, MERGE, and mDIXON sequences. To evaluate the reliability of the proposed iMSDE-mDIXON sequence, we analyzed its agreement in measuring morphology (lumen area, wall area, mean wall thickness, and normalized wall index) of carotid wall with the reference sequence of MERGE, as well as its agreement in quantifying carotid PVAT (PVAT area, proton density fat fraction [PDFF], area index, and volume index) with the reference sequence of mDIXON. The interclass correlation coefficient (ICC) and Bland–Altman plots were conducted in statistical analysis.</p> </section> <section> <h3> Results</h3> <p>The proposed iMSDE-mDIXON technique demonstrated high reliability in quantifying carotid vessel wall morphology (healthy subjects: ICC = 0.903–0.997; patients: ICC = 0.928–0.999) and PVAT morphology (healthy subjects: ICC = 0.906–0.988; patients: ICC = 0.957–0.996). Although iMSDE-mDIXON sequence showed potential in assessing carotid AS, there was a substantial bias (>20%) in PDFF quantification. Nevertheless, moderate to excellent agreement was maintained between iMSDE-mDIXON and mDIXON in measuring PVAT PDFF both in healthy subjects (ICC: left, 0.782; right, 0.740) and AS patients (ICC: left, 0.790; right, 0.628). In addition, the proposed sequence showed excellent agreement in quantifying carotid vessel wall (ICC = 0.845–0.999) and PVAT morphology (ICC = 0.841–0.989) between the repeated scans.</p> </section> <section> <h3> Conclu
{"title":"Simultaneous magnetic resonance imaging of carotid artery perivascular adipose tissue and vessel wall: A feasibility and repeatability study","authors":"Shuwan Yu, Shuo Chen, Zihan Ning, Huiyu Qiao, Zechen Zhou, Xueyi Chen, Ning Xu, Zixuan Lin, Xihai Zhao","doi":"10.1002/mp.70302","DOIUrl":"10.1002/mp.70302","url":null,"abstract":"<div>\u0000 \u0000 \u0000 <section>\u0000 \u0000 <h3> Background</h3>\u0000 \u0000 <p>It has been shown that rupture of vulnerable plaques of atherosclerosis (AS) is one of the main causes of ischemic stroke. Increasing evidence suggests that the inflammatory changes of perivascular adipose tissue (PVAT) were independently associated with both vulnerable plaque characteristics and cerebrovascular symptoms. Therefore, it is essential to conduct non-invasive joint assessment of carotid AS plaques and PVAT characteristics for better stratifying the risk of ischemic cerebrovascular events.</p>\u0000 </section>\u0000 \u0000 <section>\u0000 \u0000 <h3> Purpose</h3>\u0000 \u0000 <p>To develop a simultaneous magnetic resonance (MR) imaging technique for carotid artery PVAT and vessel wall, and determine its feasibility and repeatability.</p>\u0000 </section>\u0000 \u0000 <section>\u0000 \u0000 <h3> Methods</h3>\u0000 \u0000 <p>This study developed an MR sequence for simultaneous imaging carotid vessel wall and PVAT. Seventeen healthy subjects and nine patients with carotid AS were recruited for MR imaging experiments, of whom five healthy subjects were selected for the repeatability test. All participants underwent bilateral carotid three-dimensional MR imaging by acquiring the proposed iMSDE-mDIXON, MERGE, and mDIXON sequences. To evaluate the reliability of the proposed iMSDE-mDIXON sequence, we analyzed its agreement in measuring morphology (lumen area, wall area, mean wall thickness, and normalized wall index) of carotid wall with the reference sequence of MERGE, as well as its agreement in quantifying carotid PVAT (PVAT area, proton density fat fraction [PDFF], area index, and volume index) with the reference sequence of mDIXON. The interclass correlation coefficient (ICC) and Bland–Altman plots were conducted in statistical analysis.</p>\u0000 </section>\u0000 \u0000 <section>\u0000 \u0000 <h3> Results</h3>\u0000 \u0000 <p>The proposed iMSDE-mDIXON technique demonstrated high reliability in quantifying carotid vessel wall morphology (healthy subjects: ICC = 0.903–0.997; patients: ICC = 0.928–0.999) and PVAT morphology (healthy subjects: ICC = 0.906–0.988; patients: ICC = 0.957–0.996). Although iMSDE-mDIXON sequence showed potential in assessing carotid AS, there was a substantial bias (>20%) in PDFF quantification. Nevertheless, moderate to excellent agreement was maintained between iMSDE-mDIXON and mDIXON in measuring PVAT PDFF both in healthy subjects (ICC: left, 0.782; right, 0.740) and AS patients (ICC: left, 0.790; right, 0.628). In addition, the proposed sequence showed excellent agreement in quantifying carotid vessel wall (ICC = 0.845–0.999) and PVAT morphology (ICC = 0.841–0.989) between the repeated scans.</p>\u0000 </section>\u0000 \u0000 <section>\u0000 \u0000 <h3> Conclu","PeriodicalId":18384,"journal":{"name":"Medical physics","volume":"53 2","pages":""},"PeriodicalIF":3.2,"publicationDate":"2026-01-30","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146088761","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Elakkyen Murugesu, Qinghui Liu, Till Schellhorn, Espen S Kristoffersen, Inge Groote, Atle Bjornerud, Karoline Skogen, Bradley J MacIntosh
<div> <section> <h3> Background</h3> <p>Segmentation of intracranial hemorrhage (ICH) alongside the brain's ventricles can provide crucial information in the management stroke or traumatic brain injury (TBI). Automated computed tomography (CT) based segmentation is time-efficient and desirable when manual annotation is not feasible.</p> </section> <section> <h3> Purpose</h3> <p>To develop a fully automated and simultaneous segmentation method capable of quantifying ICH and the lateral ventricle volumes. Segmentation results are evaluated in patients with a range of TBI conditions and patients with hemorrhage in the ventricles, known as intraventricular hemorrhage (IVH).</p> </section> <section> <h3> Methods</h3> <p>Five deep learning models were trained using 154 non-contrast CT images with manual annotations of the ICH and lateral ventricles. Model performance was evaluated using Dice and Hausdorff Distance metrics. A top performing model was applied to three clinical samples: 1) <i>N</i> = 591 patients with mild TBI without ICH; 2) <i>N</i> = 91 moderate-to-severe TBI with baseline and follow-up CT; and 3) <i>N</i> = 5 patients with IVH and repeat CT. Statistical tests assessed model performance, and one model was selected to: test for relationship between lateral ventricle volume and age and sex, and investigate volume estimates from baseline to follow-up.</p> </section> <section> <h3> Results</h3> <p>Significant segmentation differences were observed between models for the Dice score (<i>F</i> = 3.3, <i>p</i> = 0.012), but not Hausdorff Distance (<i>F</i> = 0.58, <i>p</i> = 0.68). The 3D nnU-Net model had the highest performance with a mean Dice score of 0.92 ± 0.05 for the lateral ventricles and 0.88 ± 0.05 for ICH. In mild TBI, a quadratic association between lateral ventricle volume and age was found for the whole sample and after stratifying by sex. There were six ICH false positives out of 591 with mild TBI. In moderate-to-severe TBI patients, volume changes in both lateral ventricle and ICH were observed from baseline and follow-up (<i>p</i> < 0.04, Wilcoxon signed rank test). Follow-up lateral ventricle volume was significantly associated with the baseline estimates, but this was not the case for serial ICH estimates. The IVH case series demonstrated the feasibility of measuring lateral ventricle volume changes despite the presence of IVH.</p> </section> <section> <h3> Conclusions</h3> <p>Automated dual segmentation of lateral ventricles and ICH using deep learning provides a reliable method to monitor TBI severity over time. The approac
背景：脑室旁颅内出血（ICH）的分割可以为脑卒中或创伤性脑损伤（TBI）的治疗提供重要信息。当手工标注不可行时，基于自动计算机断层扫描（CT）的分割是有效的和可取的。目的：建立一种能够定量脑出血和侧脑室体积的全自动、同步分割方法。分割结果在一系列TBI患者和脑室出血患者（称为脑室内出血（IVH））中进行评估。方法：使用154张非对比CT图像对脑室和侧脑室进行人工标注，训练5个深度学习模型。使用Dice和Hausdorff距离指标评估模型性能。将最佳模型应用于3个临床样本：1)N = 591例轻度TBI无脑出血患者；2) 91例中重度TBI伴基线及随访CT；3) 5例IVH伴重复CT。统计测试评估模型的性能，并选择一个模型：测试侧脑室容量与年龄和性别之间的关系，并调查从基线到随访的容量估计值。结果：模型之间的Dice评分（F = 3.3, p = 0.012）存在显著的分割差异，但Hausdorff距离（F = 0.58, p = 0.68）不存在显著的分割差异。三维nnU-Net模型表现最好，侧脑室的平均Dice评分为0.92±0.05，脑室的平均Dice评分为0.88±0.05。在轻度TBI中，侧脑室容量与年龄之间的二次相关关系在整个样本和按性别分层后被发现。591例轻度TBI患者中有6例脑出血假阳性。在中重度TBI患者中，从基线和随访中观察到侧脑室和脑出血的体积变化(p)。结论：使用深度学习对侧脑室和脑出血进行自动双重分割提供了一种随时间监测TBI严重程度的可靠方法。该方法获得了临床相关信息，包括脑出血和侧脑室容积变化。这种方法提供了脑出血的CT图像的有效分析，特别是发展为IVH的患者。
{"title":"Monitoring the lateral ventricles in the presence of intracranial hemorrhage using automated dual segmentation","authors":"Elakkyen Murugesu, Qinghui Liu, Till Schellhorn, Espen S Kristoffersen, Inge Groote, Atle Bjornerud, Karoline Skogen, Bradley J MacIntosh","doi":"10.1002/mp.70308","DOIUrl":"10.1002/mp.70308","url":null,"abstract":"<div>\u0000 \u0000 \u0000 <section>\u0000 \u0000 <h3> Background</h3>\u0000 \u0000 <p>Segmentation of intracranial hemorrhage (ICH) alongside the brain's ventricles can provide crucial information in the management stroke or traumatic brain injury (TBI). Automated computed tomography (CT) based segmentation is time-efficient and desirable when manual annotation is not feasible.</p>\u0000 </section>\u0000 \u0000 <section>\u0000 \u0000 <h3> Purpose</h3>\u0000 \u0000 <p>To develop a fully automated and simultaneous segmentation method capable of quantifying ICH and the lateral ventricle volumes. Segmentation results are evaluated in patients with a range of TBI conditions and patients with hemorrhage in the ventricles, known as intraventricular hemorrhage (IVH).</p>\u0000 </section>\u0000 \u0000 <section>\u0000 \u0000 <h3> Methods</h3>\u0000 \u0000 <p>Five deep learning models were trained using 154 non-contrast CT images with manual annotations of the ICH and lateral ventricles. Model performance was evaluated using Dice and Hausdorff Distance metrics. A top performing model was applied to three clinical samples: 1) <i>N</i> = 591 patients with mild TBI without ICH; 2) <i>N</i> = 91 moderate-to-severe TBI with baseline and follow-up CT; and 3) <i>N</i> = 5 patients with IVH and repeat CT. Statistical tests assessed model performance, and one model was selected to: test for relationship between lateral ventricle volume and age and sex, and investigate volume estimates from baseline to follow-up.</p>\u0000 </section>\u0000 \u0000 <section>\u0000 \u0000 <h3> Results</h3>\u0000 \u0000 <p>Significant segmentation differences were observed between models for the Dice score (<i>F</i> = 3.3, <i>p</i> = 0.012), but not Hausdorff Distance (<i>F</i> = 0.58, <i>p</i> = 0.68). The 3D nnU-Net model had the highest performance with a mean Dice score of 0.92 ± 0.05 for the lateral ventricles and 0.88 ± 0.05 for ICH. In mild TBI, a quadratic association between lateral ventricle volume and age was found for the whole sample and after stratifying by sex. There were six ICH false positives out of 591 with mild TBI. In moderate-to-severe TBI patients, volume changes in both lateral ventricle and ICH were observed from baseline and follow-up (<i>p</i> < 0.04, Wilcoxon signed rank test). Follow-up lateral ventricle volume was significantly associated with the baseline estimates, but this was not the case for serial ICH estimates. The IVH case series demonstrated the feasibility of measuring lateral ventricle volume changes despite the presence of IVH.</p>\u0000 </section>\u0000 \u0000 <section>\u0000 \u0000 <h3> Conclusions</h3>\u0000 \u0000 <p>Automated dual segmentation of lateral ventricles and ICH using deep learning provides a reliable method to monitor TBI severity over time. The approac","PeriodicalId":18384,"journal":{"name":"Medical physics","volume":"53 2","pages":""},"PeriodicalIF":3.2,"publicationDate":"2026-01-30","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146088717","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
扫码关注我们
求助内容：
应助结果提醒方式：
京公网安备 11010802042870号