Channel-wise attention enhanced and structural similarity constrained cycleGAN for effective synthetic CT generation from head and neck MRI images

Radiation Oncology, Journal Year: 2024, Volume and Issue: 19(1)

Published: March 14, 2024

Abstract Background Magnetic resonance imaging (MRI) plays an increasingly important role in radiotherapy, enhancing the accuracy of target and organs at risk delineation, but absence electron density information limits its further clinical application. Therefore, aim this study is to develop evaluate a novel unsupervised network (cycleSimulationGAN) for unpaired MR-to-CT synthesis. Methods The proposed cycleSimulationGAN work integrates contour consistency loss function channel-wise attention mechanism synthesize high-quality CT-like images. Specially, constrains structural similarity between synthetic input images better retention characteristics. Additionally, we propose equip based on traditional generator GAN enhance feature representation capability deep extract more effective features. mean absolute error (MAE) Hounsfield Units (HU), peak signal-to-noise ratio (PSNR), root-mean-square (RMSE) index (SSIM) were calculated CT (sCT) ground truth (GT) quantify overall sCT performance. Results One hundred sixty nasopharyngeal carcinoma (NPC) patients who underwent volumetric-modulated arc radiotherapy (VMAT) enrolled study. generated our method consistent with GT compared other methods terms visual inspection. average MAE, RMSE, PSNR, SSIM over twenty 61.88 ± 1.42, 116.85 3.42, 36.23 0.52 0.985 0.002 method. four image quality assessment metrics significantly improved by approach conventional cycleGAN, produces results except bone. Conclusions We developed model that can effectively create images, making them comparable which could potentially benefit MRI-based treatment planning.

Language: Английский

---

Artificial intelligence in radiotherapy: Current applications and future trends

Diagnostic and Interventional Imaging, Journal Year: 2024, Volume and Issue: unknown

Published: June 1, 2024

Language: Английский

---

Compensation cycle consistent generative adversarial networks (Comp‐GAN) for synthetic CT generation from MR scans with truncated anatomy

Medical Physics, Journal Year: 2023, Volume and Issue: 50(7), P. 4399 - 4414

Published: Jan. 26, 2023

MR scans used in radiotherapy can be partially truncated due to the limited field of view (FOV), affecting dose calculation accuracy MR-based radiation treatment planning.

Language: Английский

---

A Systematic Collection of Medical Image Datasets for Deep Learning

ACM Computing Surveys, Journal Year: 2023, Volume and Issue: 56(5), P. 1 - 51

Published: Aug. 16, 2023

The astounding success made by artificial intelligence in healthcare and other fields proves that it can achieve human-like performance. However, always comes with challenges. Deep learning algorithms are data dependent require large datasets for training. Many junior researchers face a lack of variety reasons. Medical image acquisition, annotation, analysis costly, their usage is constrained ethical restrictions. They also several resources, such as professional equipment expertise. That makes difficult novice non-medical to have access medical data. Thus, comprehensively possible, this article provides collection associated challenges deep research. We collected the information approximately 300 mainly reported between 2007 2020 categorized them into four categories: head neck, chest abdomen, pathology blood, others. purpose our work provide list, up-to-date complete be used reference easily find related these datasets.

Language: Английский

---

The Use of MR-Guided Radiation Therapy for Head and Neck Cancer and Recommended Reporting Guidance

Seminars in Radiation Oncology, Journal Year: 2023, Volume and Issue: 34(1), P. 69 - 83

Published: Dec. 15, 2023

Although magnetic resonance imaging (MRI) has become standard diagnostic workup for head and neck malignancies is currently recommended by most radiological societies pharyngeal oral carcinomas, its utilization in radiotherapy been heterogeneous during the last decades. However, few would argue that implementing MRI annotation of target volumes organs at risk provides several advantages, so implementation modality this purpose widely accepted. Today, term MR-guidance received a much broader meaning, including adaptive treatments, MR-gating tracking application, MR-features as biomarkers finally MR-only workflows. First studies on treatment cancer commercially available dedicated hybrid-platforms (MR-linacs), with distinct common features but also differences amongst them, have recently reported, well "biological adaptation" based evaluation early response via functional MRI-sequences such diffusion weighted ones. Yet, all these approaches towards remain their infancy, especially when compared to other indications. Moreover, lack standardization reporting MR-guided major obstacle both further progress field conduct compare clinical trials. Goals article present explain different aspects cancer, summarize evidence, possible advantages challenges method provide comprehensive guidance use routine

Language: Английский

---

Anatomical evaluation of deep-learning synthetic computed tomography images generated from male pelvis cone-beam computed tomography

Physics and Imaging in Radiation Oncology, Journal Year: 2023, Volume and Issue: 25, P. 100416 - 100416

Published: Jan. 1, 2023

To improve cone-beam computed tomography (CBCT), deep-learning (DL)-models are being explored to generate synthetic CTs (sCT). The sCT evaluation is mainly focused on image quality and CT number accuracy. However, correct representation of daily anatomy the CBCT also important for sCTs in adaptive radiotherapy. aim this study was emphasize importance anatomical correctness by quantitatively assessing scans generated from using different paired unpaired dl-models.

Language: Английский

---

Comprehensive dose evaluation of a Deep Learning based synthetic Computed Tomography algorithm for pelvic Magnetic Resonance-only radiotherapy

Radiotherapy and Oncology, Journal Year: 2023, Volume and Issue: 184, P. 109692 - 109692

Published: May 6, 2023

Background and PurposeMagnetic Resonance (MR)-only radiotherapy enables the use of MR without uncertainty MR–Computed Tomography (CT) registration. This requires a synthetic CT (sCT) for dose calculations, which can be facilitated by novel Zero Echo Time (ZTE) sequence where bones are visible images acquired in 65 seconds. study evaluated calculation accuracy pelvic sites ZTE-based Deep Learning sCT algorithm developed GE Healthcare.Materials MethodsZTE were 56 patients position. A 2D U-net convolutional neural network was trained using pairs deformably registered ZTE from 36 patients. In remaining 20 dosimetric assessed cylindrical dummy Planning Target Volumes (PTVs) positioned at four different central axial locations, as well clinical treatment plans (for prostate (n = 10), rectum 4) anus 6) cancers). The rigidly registered, plan recalculated doses compared mean differences gamma analysis.ResultsMean to PTV D98% ≤ 0.5% all PTVs (rigid registration). Mean pass rates 1%/1 mm 98.0 ± 0.4% (rigid) 100.0 0.0% (deformable), 96.5 0.8% 99.8 0.1%, 95.4 0.6% 99.4 prostate, respectively.ConclusionsA with high throughout pelvis has been developed. suggests is sufficiently accurate MR-only sites.

Language: Английский

---

CT synthesis from MR images using frequency attention conditional generative adversarial network

Computers in Biology and Medicine, Journal Year: 2024, Volume and Issue: 170, P. 107983 - 107983

Published: Jan. 21, 2024

Language: Английский

---

Joint synthesis and registration network for deformable MR-CBCT image registration for neurosurgical guidance

Physics in Medicine and Biology, Journal Year: 2022, Volume and Issue: 67(12), P. 125008 - 125008

Published: May 24, 2022

Abstract Objective. The accuracy of navigation in minimally invasive neurosurgery is often challenged by deep brain deformations (up to 10 mm due egress cerebrospinal fluid during neuroendoscopic approach). We propose a learning-based deformable registration method address such between preoperative MR and intraoperative CBCT. Approach. uses joint image synthesis network (denoted JSR) simultaneously synthesize CBCT images the CT domain perform using multi-resolution pyramid. JSR was first trained simulated dataset (simulated deformations) then refined on real clinical via transfer learning. performance compared single-resolution architecture as well series alternative methods (symmetric normalization (SyN), VoxelMorph, synthesis-based methods). Main results. achieved median Dice coefficient (DSC) 0.69 structures target error (TRE) 1.94 simulation dataset, with improvement from (median DSC = 0.68 TRE 2.14 mm). Additionally, superior methods—e.g. SyN 0.54, 2.77 mm), VoxelMorph 0.52, 2.66 mm) provided runtime less than 3 s. Similarly 0.72 2.05 mm. Significance. resolved other state-of-the-art methods. support translation further studies high-precision neurosurgery.

Language: Английский

---

A high-performance method of deep learning for prostate MR-only radiotherapy planning using an optimized Pix2Pix architecture

Physica Medica, Journal Year: 2022, Volume and Issue: 103, P. 108 - 118

Published: Oct. 19, 2022

Language: Английский

---