A multicenter study on deep learning for glioblastoma auto-segmentation with prior knowledge in multimodal imaging.

A precise radiotherapy plan is crucial to ensure accurate segmentation of glioblastomas (GBMs) for radiation therapy. However, the traditional manual segmentation process is labor-intensive and heavily reliant on the experience of radiation oncologists. In this retrospective study, a novel auto-segmentation method is proposed to address these problems. To assess the method's applicability across diverse scenarios, we conducted its development and evaluation using a cohort of 148 eligible patients drawn from four multicenter datasets and retrospective data collection including noncontrast CT, multisequence MRI scans, and corresponding medical records. All patients were diagnosed with histologically confirmed high-grade glioma (HGG). A deep learning-based method (PKMI-Net) for automatically segmenting gross tumor volume (GTV) and clinical target volumes (CTV1 and CTV2) of GBMs was proposed by leveraging prior knowledge from multimodal imaging. The proposed PKMI-Net demonstrated high accuracy in segmenting, respectively, GTV, CTV1, and CTV2 in an 11-patient test set, achieving Dice similarity coefficients (DSC) of 0.94, 0.95, and 0.92; 95% Hausdorff distances (HD95) of 2.07, 1.18, and 3.95 mm; average surface distances (ASD) of 0.69, 0.39, and 1.17 mm; and relative volume differences (RVD) of 5.50%, 9.68%, and 3.97%. Moreover, the vast majority of GTV, CTV1, and CTV2 produced by PKMI-Net are clinically acceptable and require no revision for clinical practice. In our multicenter evaluation, the PKMI-Net exhibited consistent and robust generalizability across the various datasets, demonstrating its effectiveness in automatically segmenting GBMs. The proposed method using prior knowledge in multimodal imaging can improve the contouring accuracy of GBMs, which holds the potential to improve the quality and efficiency of GBMs' radiotherapy.

Semi-supervised auto-segmentation method for pelvic organ-at-risk in magnetic resonance images based on deep-learning.

BACKGROUND AND PURPOSE: In radiotherapy, magnetic resonance (MR) imaging has higher contrast for soft tissues compared to computed tomography (CT) scanning and does not emit radiation. However, manual annotation of the deep learning-based automatic organ-at-risk (OAR) delineation algorithms is expensive, making the collection of large-high-quality annotated datasets a challenge. Therefore, we proposed the low-cost semi-supervised OAR segmentation method using small pelvic MR image annotations. METHODS: We trained a deep learning-based segmentation model using 116 sets of MR images from 116 patients. The bladder, femoral heads, rectum, and small intestine were selected as OAR regions. To generate the training set, we utilized a semi-supervised method and ensemble learning techniques. Additionally, we employed a post-processing algorithm to correct the self-annotation data. Both 2D and 3D auto-segmentation networks were evaluated for their performance. Furthermore, we evaluated the performance of semi-supervised method for 50 labeled data and only 10 labeled data. RESULTS: The Dice similarity coefficient (DSC) of the bladder, femoral heads, rectum and small intestine between segmentation results and reference masks is 0.954, 0.984, 0.908, 0.852 only using self-annotation and post-processing methods of 2D segmentation model. The DSC of corresponding OARs is 0.871, 0.975, 0.975, 0.783, 0.724 using 3D segmentation network, 0.896, 0.984, 0.890, 0.828 using 2D segmentation network and common supervised method. CONCLUSION: The outcomes of our study demonstrate that it is possible to train a multi-OAR segmentation model using small annotation samples and additional unlabeled data. To effectively annotate the dataset, ensemble learning and post-processing methods were employed. Additionally, when dealing with anisotropy and limited sample sizes, the 2D model outperformed the 3D model in terms of performance.

Clinical evaluation of deep learning-based clinical target volume three-channel auto-segmentation algorithm for adaptive radiotherapy in cervical cancer.

OBJECTIVES: Accurate contouring of the clinical target volume (CTV) is a key element of radiotherapy in cervical cancer. We validated a novel deep learning (DL)-based auto-segmentation algorithm for CTVs in cervical cancer called the three-channel adaptive auto-segmentation network (TCAS). METHODS: A total of 107 cases were collected and contoured by senior radiation oncologists (ROs). Each case consisted of the following: (1) contrast-enhanced CT scan for positioning, (2) the related CTV, (3) multiple plain CT scans during treatment and (4) the related CTV. After registration between (1) and (3) for the same patient, the aligned image and CTV were generated. Method 1 is rigid registration, method 2 is deformable registration, and the aligned CTV is seen as the result. Method 3 is rigid registration and TCAS, method 4 is deformable registration and TCAS, and the result is generated by a DL-based method. RESULTS: From the 107 cases, 15 pairs were selected as the test set. The dice similarity coefficient (DSC) of method 1 was 0.8155 ± 0.0368; the DSC of method 2 was 0.8277 ± 0.0315; the DSCs of method 3 and 4 were 0.8914 ± 0.0294 and 0.8921 ± 0.0231, respectively. The mean surface distance and Hausdorff distance of methods 3 and 4 were markedly better than those of method 1 and 2. CONCLUSIONS: The TCAS achieved comparable accuracy to the manual delineation performed by senior ROs and was significantly better than direct registration.

Hyper-Heuristic Capacitance Array Method for Multi-Metal Wear Debris Detection.

Online detection of fatigued wear debris in the lubricants of aero-engines can provide warning of engine failure during flight, thus having great economic and social benefits. In this paper, we propose a capacitance array sensor and a hyper-heuristic partial differential equation (PDE) inversion method for detecting multiple micro-scale metal debris, combined with self-adaptive cellular genetic (SA-CGA) and morphological algorithms. Firstly, different from the traditional methods, which are limited in multi-induction-Dirac-boundary-inversion, a mathematical model with non-local boundary conditions is established. Furthermore, a hyper-heuristic method based on prior knowledge is also proposed to extract the wear character. Moreover, a 12-plate array circulating sensor and corresponding detection system are designed. The experimental results were compared with the optical microscopy. The results show that under the conditions of 1~3 wear debris with diameters of between 250⁻900 µm, the accuracy of the proposed method is 10⁻38% higher than those of the traditional methods. The recognition error of the wear debris counts decreases to 0.

Direct measurement of the sound velocity in seawater based on the pulsed acousto-optic effect between the frequency comb and the ultrasonic pulse.

We present a new method to measure the velocity of sound in pure water and seawater using the Raman-Nath diffraction caused by acousto-optic effect between the optical frequency comb and the ultrasonic pulse. In the Mach-Zehnder interferometry system we established, the measurement and reference arms are tagged with sharp negative pulses caused by the pulsed ultrasound passing through them. The difference in optical path between the two parallel beams is twice the flight distance of the ultrasonic waves. The span between the two negative pulses reflects the time interval. At the same time, the distance between the two arms can be measured precisely using the femtosecond laser interferometry. Consequently, the time interval and the distance can be used to measure the sound velocity. The experimental results show that, the uncertainty of the sound speed measurement can achieve 0.03m/s@1482m/s in pure water and 0.029m/s@1527m/s in seawater, respectively, compared with the commercial sound velocity profiler (SVP). More importantly, benefiting from the faster and cleaner response of the acousto-optic effect than the piezoelectric effect which is widely adopted in direct sound velocity measurement method, our method provides a new idea for the metrology of sound velocity in seawater.

Ultrasonic Phased Array Compressive Imaging in Time and Frequency Domain: Simulation, Experimental Verification and Real Application.

Embracing the fact that one can recover certain signals and images from far fewer measurements than traditional methods use, compressive sensing (CS) provides solutions to huge amounts of data collection in phased array-based material characterization. This article describes how a CS framework can be utilized to effectively compress ultrasonic phased array images in time and frequency domains. By projecting the image onto its Discrete Cosine transform domain, a novel scheme was implemented to verify the potentiality of CS for data reduction, as well as to explore its reconstruction accuracy. The results from CIVA simulations indicate that both time and frequency domain CS can accurately reconstruct array images using samples less than the minimum requirements of the Nyquist theorem. For experimental verification of three types of artificial flaws, although a considerable data reduction can be achieved with defects clearly preserved, it is currently impossible to break Nyquist limitation in the time domain. Fortunately, qualified recovery in the frequency domain makes it happen, meaning a real breakthrough for phased array image reconstruction. As a case study, the proposed CS procedure is applied to the inspection of an engine cylinder cavity containing different pit defects and the results show that orthogonal matching pursuit (OMP)-based CS guarantees the performance for real application.

Synthetic CT generation for pelvic cases based on deep learning in multi-center datasets.

BACKGROUND AND PURPOSE: To investigate the feasibility of synthesizing computed tomography (CT) images from magnetic resonance (MR) images in multi-center datasets using generative adversarial networks (GANs) for rectal cancer MR-only radiotherapy. MATERIALS AND METHODS: Conventional T2-weighted MR and CT images were acquired from 90 rectal cancer patients at Peking University People's Hospital and 19 patients in public datasets. This study proposed a new model combining contrastive learning loss and consistency regularization loss to enhance the generalization of model for multi-center pelvic MRI-to-CT synthesis. The CT-to-sCT image similarity was evaluated by computing the mean absolute error (MAE), peak signal-to-noise ratio (SNRpeak), structural similarity index (SSIM) and Generalization Performance (GP). The dosimetric accuracy of synthetic CT was verified against CT-based dose distributions for the photon plan. Relative dose differences in the planning target volume and organs at risk were computed. RESULTS: Our model presented excellent generalization with a GP of 0.911 on unseen datasets and outperformed the plain CycleGAN, where MAE decreased from 47.129 to 42.344, SNRpeak improved from 25.167 to 26.979, SSIM increased from 0.978 to 0.992. The dosimetric analysis demonstrated that most of the relative differences in dose and volume histogram (DVH) indicators between synthetic CT and real CT were less than 1%. CONCLUSION: The proposed model can generate accurate synthetic CT in multi-center datasets from T2w-MR images. Most dosimetric differences were within clinically acceptable criteria for photon radiotherapy, demonstrating the feasibility of an MRI-only workflow for patients with rectal cancer.

Initial Experience of CT-Based Online Adaptive Radiotherapy for Nasopharyngeal Carcinoma With a Novel Integrated Platform: A Case Report.

Intensity-modulated radiation therapy (IMRT) improves tumor control and reduces long-term radiation-induced complications of patients with nasopharyngeal carcinoma (NPC), contingent upon accurate contouring and precise delivery of treatment plans. Online adaptive radiotherapy (ART) involves real-time treatment plan modification based on the variations in targets and organs at risk (OARs) to uphold treatment planning accuracy. This study describes the first reported case of fan beam computed tomography (FBCT)-guided online ART for NPC using a novel integrated platform. Online ART was performed at the 25th fraction in this case, as tumors and the patient's anatomy were observed to regress inter-fractionally, necessitating adjustments to the contours based on the anatomy of the day. Online ART plan optimized target volume coverage while reducing doses to OARs. Notably, online ART significantly improved radiotherapy efficiency. This patient achieved a clinical complete response 12 weeks post-treatment, with Epstein-Barr virus DNA levels reduced to 0 copies/ml. Currently, the patient is alive without evidence of high-grade toxicity or local recurrence at approximately 10 months post-treatment. This case confirms the feasibility and dosimetric benefit of online ART for NPC using a novel integrated platform. Further research is needed to confirm its clinical benefits.

Machine learning-based ensemble prediction model for the gamma passing rate of VMAT-SBRT plan.

PURPOSE: The purpose of this study was to accurately predict or classify the beam GPR with an ensemble model by using machine learning for SBRT(VMAT) plans. METHODS: A total of 128 SBRT VMAT plans with 330 arc beams were retrospectively selected, and 216 radiomics and 34 plan complexity features were calculated for each arc beam. Three models for GPR prediction and classification using support vector machine algorithm were as follows: (1) plan complexity feature-based model (plan model); (2) radiomics feature-based model (radiomics model); and (3) an ensemble model combining the two models (ensemble model). The prediction performance was evaluated by calculating the mean absolute error (MAE), root mean square error (RMSE), and Spearman's correlation coefficient (SC), and the classification performance was measured by calculating the area under the receiver operating characteristic curve (AUC), accuracy, specificity, and sensitivity. RESULTS: The MAE, RMSE and SC at the 2 %/2 mm gamma criterion in the test dataset were 1.4 %, 2.57 %, and 0.563, respectively, for the plan model; 1.42 %, and 2.51 %, and 0.508, respectively, for the radiomics model; and 1.33 %, 2.49 %, and 0.611, respectively, for the ensemble model. The accuracy, specificity, sensitivity, and AUC at the 2 %/2 mm gamma criterion in the test dataset were 0.807, 0.824, 0.681, and 0.854, respectively, for the plan model; 0.860, 0.893, 0.624, and 0.877, respectively, for the radiomics model; and 0.852, 0.871, 0.710, and 0.896, respectively, for the ensemble model. CONCLUSIONS: The ensemble model can improve the prediction and classification performance for the GPR of SBRT (VMAT).

Key technologies and challenges in online adaptive radiotherapy for lung cancer.

ABSTRACT: Definitive treatment of lung cancer with radiotherapy is challenging, as respiratory motion and anatomical changes can increase the risk of severe off-target effects during radiotherapy. Online adaptive radiotherapy (ART) is an evolving approach that enables timely modification of a treatment plan during the interfraction of radiotherapy, in response to physiologic or anatomic variations, aiming to improve the dose distribution for precise targeting and delivery in lung cancer patients. The effectiveness of online ART depends on the seamless integration of multiple components: sufficient quality of linear accelerator-integrated imaging guidance, deformable image registration, automatic recontouring, and efficient quality assurance and workflow. This review summarizes the present status of online ART for lung cancer, including key technologies, as well as the challenges and areas of active research in this field.

Analysis of Individualized Silicone Rubber Bolus Using Fan Beam Computed Tomography in Postmastectomy Radiotherapy: A Dosimetric Evaluation and Skin Acute Radiation Dermatitis Survey.

Objective: To investigate the dosimetric effects of using individualized silicone rubber (SR) bolus on the target area and organs at risk (OARs) during postmastectomy radiotherapy (PMRT), as well as evaluate skin acute radiation dermatitis (ARD). Methods: A retrospective study was performed on 30 patients with breast cancer. Each patient was prepared with an individualized SR bolus of 3 mm thickness. Fan-beam computed tomography (FBCT) was performed at the first and second fractions, and then once a week for a total of 5 times. Dosimetric metrics such as homogeneity index (HI), conformity index (CI), skin dose (SD), and OARs including the heart, lungs, and spinal cord were compared between the original plan and the FBCTs. The acute side effects were recorded. Results: In targets' dosimetric metrics, there were no significant differences in Dmean and V105% between planning computed tomography (CT) and actual treatments (P > .05), while the differences in D95%, V95%, HI, and CI were statistically significant (P < .05). In OARs, there were no significant differences between the Dmean, V5, and V20 of the affected lung, V5 of the heart and Dmax of the spinal cord (P > .05) except the V30 of affected lung, which was slightly lower than the planning CT (P < .05). In SD, both Dmax and Dmean in actual treatments were increased than plan A, and the difference was statistically significant (P < .05), while the skin-V20 and skin-V30 has no difference. Among the 30 patients, only one patient had no skin ARD, and 5 patients developed ARD of grade 2, while the remaining 24 patients were grade 1. Conclusion: The OR bolus showed good anastomoses and high interfraction reproducibility with the chest wall, and did not cause deformation during irradiation. It ensured accurate dose delivery of the target and OARs during the treatment, which may increase SD by over 101%. In this study, no cases of grade 3 skin ARD were observed. However, the potential of using OR bolus to reduce grade 1 and 2 skin ARD warrants further investigation with a larger sample size.

Study of peripheral dose from low-dose CT to adaptive radiotherapy of postoperative prostate cancer.

Objectives: The increasing use of computed tomography (CT) for adaptive radiotherapy (ART) has raised concerns about the peripheral radiation dose. This study investigates the feasibility of low-dose CT (LDCT) for postoperative prostate cancer ART to reduce the peripheral radiation dose, and evaluates the peripheral radiation dose of different imaging techniques and propose an image enhancement method based on deep learning for LDCT. Materials and methods: A linear accelerator integrated with a 16-slice fan-beam CT from UIH (United Imaging Healthcare, China) was utilized for prostate cancer ART. To reduce the tube current of CT for ART, LDCT was acquired. Peripheral doses of normal-dose CT (NDCT), LDCT, and mega-voltage computed tomography (MV-CT) were measured using a cylindrical Virtual Water™ phantom and an ion chamber. A deep learning model of LDCT for abdominal and pelvic-based cycle-consistent generative adversarial network was employed to enhance the image quality of LDCT. Six postoperative prostate cancer patients were selected to evaluate the feasibility of low-dose CT network restoration images (RCT) by the deep learning model for ART. The three aspects among NDCT, LDCT, and RCT were compared: the Hounsfield Unit (HU) of the tissue, the Dice Similarity Coefficient (DSC) criterion of target and organ, and dose calculation differences. Results: In terms of peripheral dose, the LDCT had a surface measurement point dose of approximately 1.85 mGy at the scanning field, while the doses of NDCT and MV-CT were higher at 22.85 mGy and 29.97 mGy, respectively. However, the image quality of LDCT was worse than NDCT. When compared to LDCT, the tissue HU value of RCT showed a significant improvement and was closer to that of NDCT. The DSC results for target CTV between RCT and NDCT were also impressive, reaching up to 94% for bladder and femoral heads, 98% for rectum, and 94% for the target organ. Additionally, the dose calculation differences for the ART plan based on LDCT and NDCT were all within 1%. Overall, these findings suggest that RCT can provide an effective alternative to NDCT and MV-CT with similar or better outcomes in HU values of tissue and organ damage. More testing is required before clinical application.

Total marrow lymphoid irradiation IMRT treatment using a novel CT-linac.

BACKGROUND: A novel CT-linac (kilovolt fan-beam CT-linac) has been introduced into total marrow and lymphoid irradiation (TMLI) treatment. Its integrated kilovolt fan-beam CT (kV FBCT) can be used not only for image guidance (IGRT) but also to re-calculate the dose. PURPOSE: This study reported our clinical routine on performing TMIL treatment on the CT-linac, as well as dose distribution comparison between planned and re-calculated based on IGRT FBCT image sets. METHODS: 11 sets of data from 5 male and 6 female patients who had underwent the TMLI treatment with uRT-linac 506c were selected for this study. The planning target volumes consist of all skeletal bones exclusion of the mandible and lymphatic sanctuary sites. A planned dose of 10 Gy was prescribed to all skeletal bones exclusion of the mandible in two fractions and 12 Gy in two fractions was prescribed to lymphatic sanctuary sites. Each TMLI plan contained two sub-plans, one dynamic IMRT for the upper body and the other VMAT for the lower extremity. Two attempts were made to obtain homogeneous dose in the overlapping region, i.e., applying two plans with different isocenters for the treatment of two fractions, and using a dose gradient matching scheme. The CT scans, including planning CT and IGRT FBCT, were stitched to a whole body CT scan for dose distribution evaluation. RESULTS: The average beam-on time of Planupper is 30.6 min, ranging from 24.9 to 37.5 min, and the average beam-on time of Planlower is 6.3 min, ranging from 5.7 to 8.2 min. For the planned dose distribution, the 94.79% of the PTVbone is covered by the prescription dose of 10 Gy (V10), and the 94.68% of the PTVlymph is covered by the prescription dose of 12 Gy (V12). For the re-calculated dose distribution, the 92.17% of the PTVbone is covered by the prescription dose of 10 Gy (V10), and the 90.07% of the PTVlymph is covered by the prescription dose of 12 Gy (V12). The results showed that there is a significant difference (p < 0.05) between planning V10, V12 and delivery V10, V12. There is no significant difference (p > 0.05) between planned dose and re-calculated dose on selected organs, except for right lens (p < 0.05, Dmax). The actual delivered maximum dose of right lens is apparently larger than the planned dose of it. CONCLUSION: TMLI treatment can be performed on the CT-linac with clinical acceptable quality and high efficiency. Evaluation of the recalculated dose on IGRT FBCT suggests the treatment was delivered with adequate target coverage.

Transfer Learning-Based Autosegmentation of Primary Tumor Volumes of Glioblastomas Using Preoperative MRI for Radiotherapy Treatment.

Objectives: Glioblastoma is the most common primary malignant brain tumor in adults and can be treated with radiation therapy. However, tumor target contouring for head radiation therapy is labor-intensive and highly dependent on the experience of the radiation oncologist. Recently, autosegmentation of the tumor target has been playing an increasingly important role in the development of radiotherapy plans. Therefore, we established a deep learning model and improved its performance in autosegmenting and contouring the primary gross tumor volume (GTV) of glioblastomas through transfer learning. Methods: The preoperative MRI data of 20 patients with glioblastomas were collected from our department (ST) and split into a training set and testing set. We fine-tuned a deep learning model for autosegmentation of the hippocampus on separate MRI scans (RZ) through transfer learning and trained this deep learning model directly using the training set. Finally, we evaluated the performance of both trained models in autosegmenting glioblastomas using the testing set. Results: The fine-tuned model converged within 20 epochs, compared to over 50 epochs for the model trained directly by the same training set, and demonstrated better autosegmentation performance [Dice similarity coefficient (DSC) 0.9404 ± 0.0117, 95% Hausdorff distance (95HD) 1.8107 mm ±0.3964mm, average surface distance (ASD) 0.6003 mm ±0.1287mm] than the model trained directly (DSC 0.9158±0.0178, 95HD 2.5761 mm ± 0.5365mm, ASD 0.7579 mm ± 0.1468mm) with the same test set. The DSC, 95HD, and ASD values of the two models were significantly different (P<0.05). Conclusion: A model developed with semisupervised transfer learning and trained on independent data achieved good performance in autosegmenting glioblastoma. The autosegmented volume of glioblastomas is sufficiently accurate for radiotherapy treatment, which could have a positive impact on tumor control and patient survival.

Deep learning-based low-dose CT for adaptive radiotherapy of abdominal and pelvic tumors.

The shape and position of abdominal and pelvic organs change greatly during radiotherapy, so image-guided radiation therapy (IGRT) is urgently needed. The world's first integrated CT-linac platform, equipped with fan beam CT (FBCT), can provide a diagnostic-quality FBCT for achieve adaptive radiotherapy (ART). However, CT scans will bring the risk of excessive scanning radiation dose. Reducing the tube current of the FBCT system can reduce the scanning dose, but it will lead to serious noise and artifacts in the reconstructed images. In this study, we proposed a deep learning method, Content-Noise Cycle-Consistent Generative Adversarial Network (CNCycle-GAN), to improve the image quality and CT value accuracy of low-dose FBCT images to meet the requirements of adaptive radiotherapy. We selected 76 patients with abdominal and pelvic tumors who received radiation therapy. The patients received one low-dose CT scan and one normal-dose CT scan in IGRT mode during different fractions of radiotherapy. The normal dose CT images (NDCT) and low dose CT images (LDCT) of 70 patients were used for network training, and the remaining 6 patients were used to validate the performance of the network. The quality of low-dose CT images after network restoration (RCT) were evaluated in three aspects: image quality, automatic delineation performance and dose calculation accuracy. Taking NDCT images as a reference, RCT images reduced MAE from 34.34 ± 5.91 to 20.25 ± 4.27, PSNR increased from 34.08 ± 1.49 to 37.23 ± 2.63, and SSIM increased from 0.92 ± 0.08 to 0.94 ± 0.07. The P value is less than 0.01 of the above performance indicators indicated that the difference were statistically significant. The Dice similarity coefficients (DCS) between the automatic delineation results of organs at risk such as bladder, femoral heads, and rectum on RCT and the results of manual delineation by doctors both reached 0.98. In terms of dose calculation accuracy, compared with the automatic planning based on LDCT, the difference in dose distribution between the automatic planning based on RCT and the automatic planning based on NDCT were smaller. Therefore, based on the integrated CT-linac platform, combined with deep learning technology, it provides clinical feasibility for the realization of low-dose FBCT adaptive radiotherapy for abdominal and pelvic tumors.

Characterization of pulsed ultrasound using optical detection in Raman-Nath regime.

In this paper, we describe an optical detection method for the characterization of pulsed ultrasound based on acousto-optic interaction. We deduce the relationship between the ultrasound and the diffracted light from the principle of acousto-optic diffraction in the Raman-Nath regime, which is verified experimentally. Five ultrasonic transducers with different central frequencies and different focusing types are measured to show the method's performance regarding linearity, sound pressure measurement, phase measurement, frequency response, and spatial resolution. The experimental results show a good agreement with simulation data by CIVA (ultrasonic simulation software, M2M NDT, Inc.) and the pulse-echo method.