Physics in medicine and biology最新文献

{"title":"Dose measurement of ophthalmic Ru-106/Rh-106 applicators with a diamond detector calibrated in a clinical megavoltage electron beam.","authors":"Assi Valve, Vappu Reijonen, Anna Rintala, Satu Strengell, Katri Nousiainen, Mikko Tenhunen","doi":"10.1088/1361-6560/adb9b1","DOIUrl":"https://doi.org/10.1088/1361-6560/adb9b1","url":null,"abstract":"<p><strong>Objective: </strong>Uveal melanomas and retinoblastomas can be treated with ophthalmic beta-emitting ruthenium-106/rhodium-106 applicators. The applicator manufacturer provides a datasheet of the dosimetric properties of each applicator set, but the source strengths and 3D dose distributions should be verified by the end user with independent measurements.</p><p><strong>Approach: </strong>The purpose of this work was to calibrate diamond detector against low energy electron beam and determine necessary correction factors in the geometry of ophthalmic applicators to be able to perform quality assurance (QA) measurements for the applicators. Two separate sets of applicators were evaluated.</p><p><strong>Main results: </strong>The results showed good agreement with manufacturers' specifications. An average agreement of 3 % to the manufacturer's reference data was observed: measured dose rate / reference = 0.97 +/- 0.04 (mean +/- SD), range 0.90 - 1.05.</p><p><strong>Significance: </strong>It can be concluded that megavoltage electron beam is suitable for calibration of a diamond detector. After calibration, detector can be used for an absolute dose measurement of a ruthenium-106/rhodium-106 applicator with sufficient performance to detect deviations larger than 10 % in the QA before clinical use.</p>","PeriodicalId":20185,"journal":{"name":"Physics in medicine and biology","volume":" ","pages":""},"PeriodicalIF":3.3,"publicationDate":"2025-02-24","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143493225","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Low-dose CT reconstruction using cross-domain deep learning with domain transfer module.","authors":"Yoseob Han","doi":"10.1088/1361-6560/adb932","DOIUrl":"https://doi.org/10.1088/1361-6560/adb932","url":null,"abstract":"<p><strong>Objective: </strong>X-ray computed tomography employing low-dose X-ray source is actively researched to reduce radiation exposure. However, the reduced photon count in low-dose X-ray sources leads to severe noise artifacts in analytic reconstruction methods like filtered backprojection. Recently, deep learning-based approaches employing uni-domain networks, either in the image-domain or projection-domain, have demonstrated remarkable effectiveness in reducing image noise and Poisson noise caused by low-dose X-ray source. Furthermore, dual-domain networks that integrate image-domain and projection-domain networks are being developed to surpass the performance of uni-domain networks. Despite this advancement, dual-domain networks require twice the computational resources of uni-domain networks, even though their underlying network architectures are not substantially different.&#xD;&#xD;Approach: The U-Net architecture, a type of Hourglass network, comprises encoder and decoder modules. The encoder extracts meaningful representations from the input data, while the decoder uses these representations to reconstruct the target data. In dual-domain networks, however, encoders and decoders are redundantly utilized due to the sequential use of two networks, leading to increased computational demands. To address this issue, this study proposes a cross-domain deep learning approach that leverages analytical domain transfer functions. These functions enable the transfer of features extracted by an encoder trained in input domain to target domain, thereby reducing redundant computations. The target data is then reconstructed using a decoder trained in the corresponding domain, optimizing resource efficiency without compromising performance.&#xD;&#xD;Main Results: The proposed cross-domain network, comprising a projection-domain encoder and an image-domain decoder, demonstrated effective performance by leveraging the domain transfer function, achieving comparable results with only half the trainable parameters of dual-domain networks. Moreover, the proposed method outperformed conventional iterative reconstruction techniques and existing deep learning approaches in reconstruction quality.&#xD;&#xD;Significance: The proposed network leverages the transfer function to bypass redundant encoder and decoder modules, enabling direct connections between different domains. This approach not only surpasses the performance of dual-domain networks but also significantly reduces the number of required parameters. By facilitating the transfer of primal representations across domains, the method achieves synergistic effects, delivering high quality reconstruction images with reduced radiation doses.</p>","PeriodicalId":20185,"journal":{"name":"Physics in medicine and biology","volume":" ","pages":""},"PeriodicalIF":3.3,"publicationDate":"2025-02-21","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143472829","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Power absorption and temperature rise in deep learning based head models for local radiofrequency exposures.","authors":"Sachiko Kodera, Reina Yoshida, Essam A Rashed, Yinliang Diao, Hiroyuki Takizawa, Akimasa Hirata","doi":"10.1088/1361-6560/adb935","DOIUrl":"https://doi.org/10.1088/1361-6560/adb935","url":null,"abstract":"<p><strong>Objective: </strong>Computational uncertainty and variability of power absorption and temperature rise in humans for radiofrequency (RF) exposure is a critical factor in ensuring human protection. This aspect has been emphasized as a priority. However, accurately modeling head tissue composition and assigning tissue dielectric and thermal properties remains a challenging task. This study investigated the impact of segmentation-based versus segmentation-free models for assessing localized RF exposure. &#xD;Approach: Two computational head models were compared: one employing traditional tissue segmentation and the other leveraging deep learning to estimate tissue dielectric and thermal properties directly from magnetic resonance images. The finite-difference time-domain method and the bioheat transfer equation was solved to assess temperature rise for local exposure. Inter-subject variability and dosimetric uncertainties were analyzed across multiple frequencies.&#xD;Main Results: The comparison between the two methods for head modeling demonstrated strong consistency, with differences in peak temperature rise of 7.6±6.4%. The segmentation-free model showed reduced inter-subject variability, particularly at higher frequencies where superficial heating dominates. The maximum relative standard deviation in the inter-subject variability of heating factor was 15.0% at 3 GHz and decreased with increasing frequencies.&#xD;Significance: This study highlights the advantages of segmentation-free deep-learning models for RF dosimetry, particularly in reducing inter-subject variability and improving computational efficiency. While the differences between the two models are relatively small compared to overall dosimetric uncertainty, segmentation-free models offer a promising approach for refining individual-specific exposure assessments. These findings contribute to improving the accuracy and consistency of human protection guidelines against RF electromagnetic field exposure.</p>","PeriodicalId":20185,"journal":{"name":"Physics in medicine and biology","volume":" ","pages":""},"PeriodicalIF":3.3,"publicationDate":"2025-02-21","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143472644","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Nuclear interaction correction based on dual-energy computed tomography in carbon-ion radiotherapy.","authors":"Yushi Wakisaka, Masashi Yagi, Yuki Tominaga, Shinichi Shimizu, Teiji Nishio, Kazuhiko Ogawa","doi":"10.1088/1361-6560/adaad4","DOIUrl":"10.1088/1361-6560/adaad4","url":null,"abstract":"<p><p><i>Objective.</i>Accurate dose predictions are crucial to maximizing the benefits of carbon-ion therapy (CIT). Carbon beams incident on the human body cause nuclear interactions with tissues, resulting in changes in the constituent nuclides and leading to dose errors that are conventionally corrected using conventional single-energy computed tomography (SECT). Dual-energy computed tomography (DECT) has frequently been used for stopping power estimation in particle therapy and is well suited for correcting nuclear reactions because of its detailed body-tissue elemental information. This study proposes a correction method for the absolute dose in CIT that considers changes in nuclide composition resulting from nuclear reactions with body tissues, as a novel application of DECT.<i>Approach.</i>The change in dose associated with nuclear reactions is determined by correcting each integrated depth dose component of the carbon beam using a nuclear interaction correction factor. This factor is determined based on the stopping power, mass density, and nuclear interaction cross-section in body tissue. The stopping power and mass density were calculated using established methods, whereas the nuclear interaction cross-section was newly defined through a conversion equation derived from the effective atomic number.<i>Main results.</i>Nuclear interaction correction factors and corrected doses were determined for 85 body tissues with known compositions, comparing them with existing SECT-based methods. The root-mean-square errors of the SECT- and DECT-based nuclear interaction correction factors relative to theoretical values were 0.66% and 0.39%, respectively.<i>Significance.</i>This indicates a notable enhancement in the estimation accuracy with DECT. The dose calculations in uniform body tissues derived from SECT showed slight over-correction in adipose and bone tissues, whereas those based on DECT were almost consistent with theoretical values. Our proposed method demonstrates the potential of DECT for enhancing dose calculation accuracy in CIT, complementing its established role in stopping power estimation.</p>","PeriodicalId":20185,"journal":{"name":"Physics in medicine and biology","volume":" ","pages":""},"PeriodicalIF":3.3,"publicationDate":"2025-02-21","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143010007","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Validation of patient-specific deep learning markerless lung tumor tracking aided by 4DCBCT.","authors":"Lili Huang, Adrian Thummerer, Christianna Iris Papadopoulou, Stefanie Corradini, Claus Belka, Marco Riboldi, Christopher Kurz, Guillaume Landry","doi":"10.1088/1361-6560/adb89c","DOIUrl":"https://doi.org/10.1088/1361-6560/adb89c","url":null,"abstract":"<p><strong>Objective: </strong>Tracking tumors with multi-leaf collimators and X-ray imaging can be a cost-effective motion management method to reduce internal target volume margins for lung cancer patients, sparing normal tissues while ensuring target coverage. To realize that, accurate tumor localization on X-ray images is essential. We aimed to develop a systematic method for automatically generating tumor segmentation ground truth (GT) on CBCT projections and use it to help refine and validate our patient-specific AI-based tumor localization model.</p><p><strong>Approach: </strong>To obtain the tumor segmentation GT on CBCT projections, we propose a 4DCBCT-aided GT generation pipeline consisting of three steps: breathing phase extraction and 10-phase 4DCBCT reconstruction, manual segmentation on phase 50%, deformable contour propagation to other phases, and forward projection of the 3D segmentation to the CBCT projection of the corresponding phase. We then used the CBCT projections from one fraction in the angular range of [-10°, 10°] and [80°, 100°] to refine a Retina U-Net baseline model, which was pretrained on 1140231 digitally reconstructed radiographs generated from a public lung dataset for automatic tumor delineation on projections, and used later-fraction CBCT projections in the same angular range for testing. Six LMU University Hospital patient CBCT projection sets were reserved for validation and 11 for testing. Tracking accuracy was evaluated as the center-of-mass (COM) error and the Dice similarity coefficient (DSC) between the predicted and ground-truth segmentations.</p><p><strong>Main results: </strong>Over the 11 testing patients, each with around 40 CBCT projections tested, the patient-refined models had a mean COM error of 2.3±0.9mm / 4.2±1.7mm and a mean DSC of 0.83±0.06 / 0.72±0.13 for angles within [-10°, 10°] / [80°, 100°]. The mean inference time was 68 ms/frame. The patient-specific training segmentation loss was found to be correlated to the segmentation performance at [-10°, 10°].</p><p><strong>Significance: </strong>Our proposed approach allows patient-specific real-time markerless lung tumor tracking, which could be validated thanks to the novel 4DCBCT-aided GT generation approach.</p>","PeriodicalId":20185,"journal":{"name":"Physics in medicine and biology","volume":" ","pages":""},"PeriodicalIF":3.3,"publicationDate":"2025-02-20","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143468800","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"IConDiffNet: an unsupervised inverse-consistent diffeomorphic network for medical image registration.","authors":"Rui Liao, Jeffrey F Williamson, Tianyu Xia, Tao Ge, Joseph A O'Sullivan","doi":"10.1088/1361-6560/ada516","DOIUrl":"10.1088/1361-6560/ada516","url":null,"abstract":"&lt;p&gt;&lt;p&gt;&lt;i&gt;Objective.&lt;/i&gt;Deformable image registration (DIR) is critical in many medical imaging applications. Diffeomorphic transformations, which are smooth invertible mappings with smooth inverses preserve topological properties and are an anatomically plausible means of constraining the solution space in many settings. Traditional iterative optimization-based diffeomorphic DIR algorithms are computationally costly and are not able to consistently resolve large and complicated deformations in medical image registration. Convolutional neural network implementations can rapidly estimate the transformation in through a pre-trained model. However, the structure design of most neural networks for DIR fails to systematically enforce diffeomorphism and inverse consistency. In this paper, a novel unsupervised neural network structure is proposed to perform a fast, accurate, and inverse-consistent diffeomorphic DIR.&lt;i&gt;Approach.&lt;/i&gt;This paper introduces a novel unsupervised inverse-consistent diffeomorphic registration network termed IConDiffNet, which incorporates an energy constraint that minimizes the total energy expended during the deformation process. The IConDiffNet architecture consists of two symmetric paths, each employing multiple recursive cascaded updating blocks (neural networks) to handle different virtual time steps parameterizing the path from the initial undeformed image to the final deformation. These blocks estimate velocities corresponding to specific time steps, generating a series of smooth time-dependent velocity vector fields. Simultaneously, the inverse transformations are estimated by corresponding blocks in the inverse path. By integrating these series of time-dependent velocity fields from both paths, optimal forward and inverse transformations are obtained, aligning the image pair in both directions.&lt;i&gt;Main result.&lt;/i&gt;Our proposed method was evaluated on a three-dimensional inter-patient image registration task with a large-scale brain MRI image dataset containing 375 subjects. The proposed IConDiffNet achieves fast and accurate DIR with better DSC, lower Hausdorff distance metric, and lower total energy spent during the deformation in the test dataset compared to competing state-of-the-art deep-learning diffeomorphic DIR approaches. Visualization shows that IConDiffNet produces more complicated transformations that better align structures than the VoxelMorph-Diff, SYMNet, and ANTs-SyN methods.&lt;i&gt;Significance.&lt;/i&gt;The proposed IConDiffNet represents an advancement in unsupervised deep-learning-based DIR approaches. By ensuring inverse consistency and diffeomorphic properties in the outcome transformations, IConDiffNet offers a pathway for improved registration accuracy, particularly in clinical settings where diffeomorphic properties are crucial. Furthermore, the generality of IConDiffNet's network structure supports direct extension to diverse 3D image registration challenges. This adaptability is facilitated by the flexibili","PeriodicalId":20185,"journal":{"name":"Physics in medicine and biology","volume":" ","pages":""},"PeriodicalIF":3.3,"publicationDate":"2025-02-20","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142922653","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Unravelling the dynamics of coated nanobubbles and low frequency ultrasound using the Blake threshold and modified surface tension model.","authors":"Ilia Mezdrokhin, Tali Ilovitsh","doi":"10.1088/1361-6560/adb3e9","DOIUrl":"10.1088/1361-6560/adb3e9","url":null,"abstract":"<p><p><i>Objective.</i>To develop a model that accurately describes the behavior of nanobubbles (NBs) under low-frequency ultrasound (US) insonation (<250 kHz), addressing the limitations of existing numerical models, such as the Marmottant model and Blake's Threshold model, in predicting NB behavior.<i>Approach.</i>A modified surface tension model, derived from empirical data, was introduced to capture the surface tension behavior of NBs as a function of bubble radius. This model was integrated into the Marmottant framework and combined with the Blake threshold to predict cavitation thresholds at low pressures, providing a comprehensive approach to understanding NB dynamics.<i>Main results.</i>Experimentally, inertial cavitation for NBs with a radius of 85 nm was observed at peak negative pressures of 200 kPa at 80 kHz and 1000 kPa at 250 kHz. The Marmottant model significantly overestimated these thresholds (1600 kPa). The modified surface tension model improved predictions at 250 kHz, while combining it with the Blake threshold accurately aligned cavitation thresholds at both frequencies (∼150 kPa at low pressures) with experimental results.<i>Significance.</i>This work bridges a critical gap in understanding the acoustic behavior of NBs at low US frequencies and offers a new theoretical framework for predicting cavitation thresholds of NBs at low US frequencies, advancing their application in biomedical US technologies.</p>","PeriodicalId":20185,"journal":{"name":"Physics in medicine and biology","volume":" ","pages":""},"PeriodicalIF":3.3,"publicationDate":"2025-02-19","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143370937","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Demonstration of ultra-high dose rate electron irradiation at FLASH<i>lab</i>@PITZ.","authors":"X-K Li, Z Amirkhanyan, A Grebinyk, M Gross, Y Komar, F Riemer, A Asoyan, P Boonpornprasert, P Borchert, H Davtyan, D Dmytriiev, M Frohme, A Hoffmann, M Krasilnikov, G Loisch, Z Lotfi, F Müller, M Schmitz, F Obier, A Oppelt, S Philipp, C Richard, G Vashchenko, D Villani, S Worm, F Stephan","doi":"10.1088/1361-6560/adb276","DOIUrl":"10.1088/1361-6560/adb276","url":null,"abstract":"<p><p><i>Objective.</i>The photo injector test facility at DESY in Zeuthen (PITZ) is building up an R&D platform, known as FLASH<b><i>lab</i></b>@PITZ, for systematically studying the FLASH effect in cancer treatment with its high-brightness electron beams, which can provide a uniquely large dose parameter range for radiation experiments. In this paper, we demonstrate the capabilities by experiments with a reduced parameter range on a startup beamline and study the potential performance of the full beamline by simulations.<i>Approach.</i>To measure the dose, Gafchromic films are installed both in front of and after the samples; Monte Carlo simulations are conducted to predict the dose distribution during beam preparation and help understand the dose distribution inside the sample. Plasmid DNA is irradiated under various doses at conventional and ultra-high dose rate (UHDR) to study the DNA damage by radiations. Start-to-end simulations are performed to verify the performance of the full beamline.<i>Main results.</i>On the startup beamline, reproducible irradiation has been established with optimized electron beams and the delivered dose distributions have been measured with Gafchromic films and compared to FLUKA simulations. The functionality of this setup has been further demonstrated in biochemical experiments at conventional dose rate of 0.05 Gy s^-1and UHDR of several 10⁵ Gy s^-1and a varying dose up to 60 Gy, with the UHDR experiments finished within a single RF pulse (less than 1 millisecond); the observed conformation yields of the irradiated plasmid DNA revealed its dose-dependent radiation damage. The upgrade to the full FLASH<b><i>lab</i></b>@PITZ beamline is justified by simulations with homogeneous radiation fields generated by both pencil beam scanning and scattering beams.<i>Significance.</i>With the demonstration of UHDR irradiation and the simulated performance of the new beamline, FLASH<b><i>lab</i></b>@PITZ will serve as a powerful platform for studying the FLASH effects in cancer treatment.</p>","PeriodicalId":20185,"journal":{"name":"Physics in medicine and biology","volume":" ","pages":""},"PeriodicalIF":3.3,"publicationDate":"2025-02-19","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143189693","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"<i>In-vitro</i>and microdosimetric study of proton boron capture therapy and neutron capture enhanced proton therapy.","authors":"Villads Jacobsen, Vladimir A Pan, Linh T Tran, James Vohradsky, Jonas Bønnelykke, Cecilie Schmidt Herø, Jacob G Johansen, Anders Tobias Frederiksen, Brita Singers Sørensen, Morten Busk, Wolfgang A G Sauerwein, Anatoly B Rosenfeld, Niels Bassler","doi":"10.1088/1361-6560/adb199","DOIUrl":"10.1088/1361-6560/adb199","url":null,"abstract":"<p><p><i>Objective.</i>The clinical advantage of proton therapy, compared to other types of irradiations, lies in its reduced dose to normal tissue. Still, proton therapy faces challenges of normal tissue toxicity and radioresistant tumors. To combat these challenges, proton boron capture therapy (PBCT) and neutron capture enhanced particle therapy (NCEPT) were proposed to introduce high-LET radiation in the target volume.<i>Approach</i>. In this work, we performed<i>in-vitro</i>experiments with a V79 cell line to validate PBCT and introduced a novel approach to use NCEPT in proton therapy. We quantified the effectiveness of PBCT and NCEPT with microdosimetric measurements, Monte-Carlo simulations and microdosimetric kinetic RBE model (MKM).<i>Main results</i>. No RBE increase was observed for PBCT. With the use of a tungsten spallation source, enough neutrons were generated in the incoming proton beam to measure significant neutron capture in the microdosimeter. However, no significant increase of RBE was detected when conventional<i>in vitro</i>protocol was followed. The resulting cell deactivation based RBE for NCEPT was found to be heavily dependent on which criteria was used to determine surviving colonies.<i>Significance</i>. PBCT and NCEPT are two proposed treatment modalities that may have the potential to expand the cases in which proton therapy can be beneficial. Understanding the scope of these treatment methods and developing measurement protocols to evaluate and understand their RBE impact are the first step to quantify their potential in clinical context.</p>","PeriodicalId":20185,"journal":{"name":"Physics in medicine and biology","volume":" ","pages":""},"PeriodicalIF":3.3,"publicationDate":"2025-02-18","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143123239","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Impact of pectoral muscle removal on deep-learning-based breast cancer risk prediction.","authors":"Zan Klanecek, Yao-Kuan Wang, Tobias Wagner, Lesley Cockmartin, Nicholas Marshall, Brayden Schott, Ali Deatsch, Andrej Studen, Katja Jarm, Mateja Krajc, Miloš Vrhovec, Hilde Bosmans, Robert Jeraj","doi":"10.1088/1361-6560/adb367","DOIUrl":"10.1088/1361-6560/adb367","url":null,"abstract":"<p><p><i>Objective.</i>State-of-the-art breast cancer risk (BCR) prediction models have been originally trained on mammograms with pectoral muscle (PM) included. This study investigated whether excluding PM during training/fine-tuning improves the model's BCR discrimination performance, calibration, and robustness.<i>Approach.</i>First, the Original deep learning model (MIRAI), trained on the US (Massachusetts General Hospital) data, was validated, and the relative contribution of PM to BCR predictions was evaluated using saliency maps. Additionally, 23 792 mammograms from the Slovenian screening program were collected and two datasets were created, with and without screening positive exams. The original MIRAI was then fine-tuned on the training/fine-tuning set of Slovenian mammograms with and without PM, creating Fine-tuned MIRAI models. In total, four models (Original MIRAI with PM, Original MIRAI without PM, Fine-tuned MIRAI with PM, Fine-tuned MIRAI without PM) were compared on a test set in terms of discrimination performance for 1-5 Year BCR (evaluating area under the curve), calibration performance (measured with expected calibration error-ECE) and robustness to incremental PM removals/additions, and to incremental breast tissue removals.<i>Results.</i>The relative contribution of PM to the BCR prediction on the Original MIRAI model was low (∼5%); however, there were significant outliers where the relative contribution was more than 50%. The removal of PM did not impact the 1-5 Year BCR discrimination performance of the Original MIRAI (with screening positive exams: 0.77-0.91, without screening positive exams: 0.64-0.67). Fine-tuned MIRAI on mammograms with PM removed achieved significantly higher 1-5 Year BCR discrimination performance (with screening positive exams: 0.82-0.93, without screening positive exams: 0.71-0.79). After recalibration, all models had similar ECE (with screening positive exams: 0.04-0.05, without screening positive exams: 0.02-0.03).<i>Significance.</i>Improved BCR discrimination performance can be achieved when the model is trained/fine-tuned on mammograms with PM removed.</p>","PeriodicalId":20185,"journal":{"name":"Physics in medicine and biology","volume":" ","pages":""},"PeriodicalIF":3.3,"publicationDate":"2025-02-18","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143365705","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}