SEA-Net: Structure-Enhanced Attention Network for Limited-Angle CBCT Reconstruction of Clinical Projection Data

PURPOSE/OBJECTIVE(S)

Limited-angle CBCT (LA-CBCT) is of great clinical interest, because the scanning time and the patient dose are proportional to the scanning range of gantry rotation angles of CBCT. However, the image reconstruction for LA-CBCT remains technically challenging, which suffers from severe wedge artifacts and image distortions. This work aims to improve LA-CBCT by developing deep learning (DL) methods for real clinical CBCT projection data, which is the first feasibility study of clinical-projection-data-based LA-CBCT, to the best of our knowledge.

MATERIALS/METHODS

Targeting at real clinical projection data, we have explored various DL methods such as image/data/hybrid-domain methods and finally developed a so-called Structure-Enhanced Attention Network (SEA-Net) method that has the best image quality from clinical projection data among the DL methods we have implemented. Specifically, the proposed SEA-Net employs a specialized structure enhancement sub-network to promote texture preservation. Based on the observation that the distribution of wedge artifacts in reconstruction images is non-uniform, the spatial attention module is utilized to emphasize the relevant regions while ignores the irrelevant ones, which leads to more accurate texture restoration.

RESULTS

SEA-Net was validated in comparison with analytic (FDK), iterative (TV), image-domain DL (DDNet and FED-INet, data-domain DL (DCAR), dual-domain DL (Sam'Net), and various unrolling DL (hdNet, CTNet, FSR-Net, CasRedSCAN) methods. Among all methods, the SEA-Net had the best image reconstruction quality as quantified by root-mean-square error (RMSE), peak signal-to-noise ratio (PSNR), and structural similarity index (SSIM), for various LA-CBCT problems of 90°-180° projection data. In addition, LA-CBCT via SEA-Net provided comparable accuracy for both patient setup (quantified by image registration accuracy from planning CT (pCT) to CBCT) and dose calculation (see the table), with full-view CBCT.

CONCLUSION

We explored various DL methods and developed an image-domain-based method termed SEA-Net that provided the best image quality for clinical projection data. To the best of our knowledge, this is the first feasibility study of the real clinical-projection-data-based LA-CBCT. Moreover, LA-CBCT via SEA-Net can potentially provide comparable accuracy for patient setup and dose calculation, with full-view CBCT.

Induction Immunotherapy vs. Consolidation Immunotherapy for Unresectable Stage III NSCLC

PURPOSE/OBJECTIVE(S)

Consolidation immunotherapy after chemoradiotherapy (CRT) is the standard of care for unresectable stage III non-small cell lung cancer (NSCLC). However, whether upfront immunotherapy before CRT has similar benefits has not been addressed. This study aimed at exploring the efficacy and safety of induction immunotherapy for unresectable stage III NSCLC through real-world data.

MATERIALS/METHODS

Patients diagnosed with stage III NSCLC who received immunotherapy in combination with sequential (sCRT) or concurrent CRT (cCRT) between November 2018 and December 2021 were retrospectively identified. Patients were divided into induction (Ind), consolidation (Con) and induction plus consolidation (Ind+Con) immunotherapy groups. Progression-free survival (PFS) and overall survival (OS) were assessed from the initiation of treatment and estimated by Kaplan‒Meier method. The potential factors affecting PFS and OS were analyzed by univariate and multivariate Cox regression models.

RESULTS

One hundred and two patients were included, with 52 (51.0%) patients in the Ind group, 35 (34.3%) in the Con group and 15 (14.7%) in the Ind+Con group. Median PFS was 24.0 months vs. 36.0 months vs. 19.0 months in the three groups, and 2-year PFS were 43.0% vs 51.1% vs 44.4% (p = 0.940). Median OS was not reached (NR) vs. 44.0 months vs. NR, with a 2-year OS rate of 80.5% vs. 84.4% vs. 86.2% (p = 0.861). In the cCRT setting, 2-year PFS rates were 56.7% vs. 71.6% vs. 100.0% (p = 0.439), 2-year OS rates were 92.3% vs. 89.3% vs. 100.0% in the three groups (p = 0.827). In multivariate analysis, elder (HR = 0.487, p = 0.037) and cCRT (HR = 0.282, p = 0.001) were the independent factors favoring PFS, while only elder (HR = 0.088, p = 0.021) was the independent factors favoring OS. Adverse events were similar in the three arms. Further analysis found the objective response rate (ORR) and disease control rate (DCR) in the Ind and Ind+Con group after induction immunotherapy were 59.7% and 98.5%, respectively. Only 1 (1.5%) patient developed progression. Subgroup analysis showed no significant difference in PFS (p = 0.520) and OS (p = 0.116) between patients who responded to induction immunotherapy (PR+CR) and those who did not (SD+PD). Patients with <4 cycles of induction immunotherapy exhibited numerically better PFS than those with ≥4 cycles of induction immunotherapy (p = 0.113) and improved OS (p = 0.021).

CONCLUSION

Induction immunotherapy may achieve similar survival benefits to consolidation immunotherapy, and the combination of induction and consolidation immunotherapy with cCRT appears to achieve better outcomes. It seems feasible and safe to upfront immunotherapy before CRT, and further investigations on the combination of induction immunotherapy and CRT are warranted.

Radical Chemoradiotherapy vs. Radical Surgery plus Adjuvant Chemotherapy or Chemoradiotherapy in Locally Advanced Primary Small Cell Carcinoma of the Esophagus: A Multicenter Retrospective Study in China

PURPOSE/OBJECTIVE(S)

The purpose was to compare the treatment outcomes of radical chemoradiotherapy versus radical surgery plus adjuvant chemotherapy or chemoradiotherapy in locally advanced primary small cell carcinoma of the esophagus (LA-PSCCE). The hypothesis was that radical chemoradiotherapy had better overall survival (OS) than radical surgery plus adjuvant chemotherapy or chemoradiotherapy.

MATERIALS/METHODS

This large-scale multicenter retrospective study in China enrolled patients with newly diagnosed LA-PSCCE (T3-4N0M0 or TanyN+M0, AJCC 8th edition) from 2008 to 2021. According to different curative treatment approaches, patients were divided into two groups: radical chemoradiotherapy (group: CRT), and radical surgery following adjuvant chemotherapy or chemoradiotherapy (group: S + CT/CRT). The propensity score match (PSM) was applied to reduce the effect of confounding biases in clinicopathological characteristics (age, gender, KPS, tumor location, tumor length, and cTNM stage). Univariate Cox-regression analysis and Kaplan-Meier curve were calculated for OS. Statistical results were summarized as hazard ratio (HR), 95% confidence interval (CI) and P value. A two-sided P < 0.05 was regarded to be statistically significant.

RESULTS

A total of 291 patients with a median follow-up of 4.3 years were retrospectively enrolled. After PSM analysis, 94 and 94 patients were eventually included in group CRT and S + CT/CRT, respectively. Group CRT demonstrated a significantly superior survival than group S + CT/CRT (HR, 0.63; 95% CI, 0.43-0.91; P = 0.01), with a 3-year OS of 49.5% and 27.8% (P = 0.02), respectively. In secondary analysis, patients treated with radical chemoradiotherapy consistently showed significant survival benefit than those with radical surgery plus adjuvant chemoradiotherapy (HR, 0.4; 95% CI, 0.21-0.79; P = 0.008).

CONCLUSION

For patients with newly diagnosed LA-PSCCE, radical chemoradiotherapy should be a preferred recommendation in real-world clinical practice.

Deep Learning for Automated Contouring of Primary Gross Tumor Volumes by MRI for Radiation Therapy of Brain Metastasis

PURPOSE/OBJECTIVE(S)

Radiotherapy is one of the most effective methods for the treatment of brain metastases (BMs). Traditional manual delineation of primary gross tumor volumes (GTV) of multiple BMs (especially small metastases) in radiotherapy practice is extremely labor intensive and highly dependent on oncologists' experience, achieving the precise and efficient automatic delineation of BMs is of great significance for efficient and homogeneous one-stop adaptive radiotherapy.

MATERIALS/METHODS

We retrospectively collected 62 MRI (non-enhanced T1-weighted sequences) sequences of 50 patients with BMs from January 2020 to July 2021. An automatic model (BUC-Net) for automatic delineation BMs was proposed in this work, which was based on deep learning by combining 3D bottler layer module and the cascade architecture to improve the accuracy and efficient of BMs' automatic delineation, especially for small metastases with tiny size and relatively low contrast. The prosed method was compared with the existing 3D U-Net (U-Net) and 3D U-Net Cascade (U-Net Cascade). The performance of our proposed method was evaluated by Dice similarity coefficient (DSC), 95% Hausdorff distance (HD95) and average surface distance (ASD) with human experts.

RESULTS

The automatic segmentation results of BUC-Net evaluated with 310 BMs in 13 test patients was summarized in Table 1. These BMs in each test patient were automatically delineated by two types of contours: as a whole tumor contour (Whole-delineation) and the multiple tumor contours (Multiple-delineation). BUC-Net performed the best mean DSC and HD95, which is significantly outperformed U-Net (Whole-delineation: 0.911 & 0.894 of DSC, Multiple-delineation: 0.794 & 0.754 of DSC, P < 0.05 for both) and U-Net cascade (Whole-delineation: 0.947 & 7.141 of HD95, Multiple-delineation: 0.902 & 1.171 of HD95, P < 0.05 for both); Additionally, BUC-Net achieved the best mean ASD for Whole-delineation and comparable ASD (0.290 & 0.277, P > 0) for Multiple-delineation with U-Net Cascade.

CONCLUSION

Our results showed that the proposed approach is promising for the automatic delineation of BMs in MRI, which can be integrated into a radiotherapy workflow to significantly shorten segmentation time.

Detection of Circulating Tumor DNA Correlates with Recurrence and Survival in Localized Non-Small-Cell Lung Cancer: A Meta-Analysis

PURPOSE/OBJECTIVE(S)

Circulating tumor DNA (ctDNA) is approved recently to use in clinical practice of solid tumor. Several large-scale prospective studies also revealed that minimal residual disease based on ctDNA (ctDNA-MRD) is a potential predictive and prognostic biomarker of localized non-small-cell lung cancer (NSCLC) receiving curative treatment (surgery or radiotherapy). However, there are still barriers to clinical management of ctDNA/ctDNA-MRD in localized NSCLC, and the most significant is effectiveness and detection times of ctDNA/ctDNA-MRD. This meta-analysis aims to clarify the prognostic value of the ctDNA and ctDNA-MRD in predicting the disease recurrence and survival of localized NSCLC.

MATERIALS/METHODS

Electronic databases (Pubmed/MEDLINE, Web of Science, Cochrane Library, meeting abstracts) were comprehensively searched for eligible studies from January 2001 to January 2023. The Hazard ratio (HR) from relevant reports was extracted to better evaluate the correlation of ctDNA and ctDNA-MRD detected in plasma with clinical outcomes among patients with localized NSCLC. Pooled results including ctDNA detection rate, disease-/relapse-/progression- free survival (DFS/RFS/PFS) and overall survival (OS) were obtained and analyzed by Review Manager 5.4.1.

RESULTS

A total of 18 eligible studies with 2692 patients were enrolled in the final analysis. The pooled analysis revealed that ctDNA detection in pretreatment plasma indicated poor prognosis for DFS/RFS/PFS (HR 3.82, 95% CI 2.85 - 5.12, p < .00001; Figure 1) with a long-term effect on OS (HR 4.88, 95% CI 3.29 - 7.24, p < .00001; Figure 2). The same result was also observed in patients with positive posttreatment ctDNA-MRD who have shorter DFS/RFS/PFS (HR 7.15, 95% CI 5.50 - 9.31, p < .00001; Figure 3) and OS (HR 4.34, 95% CI: 2.51-7.51, p < .00001; Figure 4) compared to negative groups.

CONCLUSION

Based on the results from our meta-analysis, the presence of pretreatment ctDNA or posttreatment ctDNA-MRD in radically treated localized NSCLC is associated with higher risk of recurrence and poorer survival. This finding is supportive of ctDNA/ctDNA-MRD becoming a widespread prognostic biomarker in localized NSCLC.

Deep Learning and GAN-Synthesis for Auto-Segmentation of Pancreatic Cancer by Non-Enhanced CT for Adaptive Radiotherapy

PURPOSE/OBJECTIVE(S)

In conventional adaptive radiotherapy (ART) for pancreatic cancer, contrast-enhanced CT (CECT) helps to more precisely delineate primary gross tumor volume (GTV) than non-enhanced CT (NECT). However, frequent use of contrast medium can damage kidneys and prolong treatment time. Moreover, traditional manual delineation is labor-intensive and highly dependent on the experience of oncologists. Currently, automatic delineation based on deep learning with Generative Adversarial Networks (GAN)-based CT synthesis is one of the most feasible solutions to these problems.

MATERIALS/METHODS

A dataset of 35 pancreatic cancer patients was retrospectively collected from May 2021 to December 2022. All patients consist of a pair of NECT and CECT. We designed and developed an automatic delineation framework (Proposed) for GTV of pancreatic cancer based on Trans-cycleGAN and a modified 3D U-Net. TranscycleGAN can not only synthesize CECT from NECT, but can also augment the amount of CT images; then all real and synthesized CT images were used to train the modified 3D U-Net for automatic delineation of GTV; finally, our framework was able to automatically delineate GTV by NECT, but not only by CECT. Our framework was evaluated by dice similarity coefficient (DSC), 95% Harsdorff distance (95HD) and average surface distance (ASD) with oncologists' manual delineation ("gold standard").

RESULTS

The evaluation results were summarized in Table 1. The proposed framework achieved the best automatic delineation results by NECT, which was superior to that of CECT: 0.917 & 0.903 of DSC, 2.498mm & 3.029mm of HD95, 0.481mm & 0.534mm of ASD, p < 0.05 for DSC and HD95. Specifically, it is significantly superior to the automatic delineation results using U-Net by CECT 0.917 & 0.818 of DSC, 2.498mm & 13.228mm of HD95, 0.481mm & 3.633mm of ASD, p < 0.05 for DSC.

CONCLUSION

We proposed an automatic delineation framework for contouring GTV in ART of pancreatic cancer based on deep learning and Trans-cycleGAN network. This framework could automatically delineate GTV and achieve better performance with NECT compared to CECT. Our method could not only reduce the use of contrast medium, but also increase the precision and effectiveness of tumor delineation, which could have a positive impact on precision radiotherapy.

All-in-One Online Radiotherapy for Nasopharyngeal Carcinoma: Preliminary Results of Treatment Time, Contouring Accuracy, Treatment Plan Quality and Patient Compliance

PURPOSE/OBJECTIVE(S)

To explore the feasibility of Fan-beam CT (FBCT)-based all in one (AIO) online workflow for nasopharyngeal carcinoma (NPC) in radical radiotherapy setting, and to preliminarily describe the timing of different steps in the process, contouring accuracy of regions of interest (ROIs), target coverage, organs at risk (OARs) dose and patient compliance.

MATERIALS/METHODS

From March 16, 2022 to January 04, 2023, 25 NPC patients (22/25 diagnosed as phase III/IV disease according to 8th edition of the AJCC/UICC staging system) consecutively treated with AIO radiotherapy were prospectively enrolled. All patients received mask fixation and MRI simulation scan in advance. Primary gross tumor volume (GTVp) of nasopharynx was automatically delineated by AI and edited manually on MRI images. AIO online workflow started with an integrated KV-level CT in a CT-integrated linear accelerator. After that GTVp was registrated to CT images and other ROIs was contoured automatically and then modified manually as needed. Subsequently automatic treatment plan was calculated and optimized until the dose of target and OARs was evaluated satisfactory by physicians and physicists. Finally, treatment was delivered using volumetric modulated arc treatment (VMAT), with prescribed dose of 6996 cGy/ 33 fractions to the GTVp.

RESULTS

Twenty-four patients (24/25, 96%) completed the AIO radiotherapy workflow successfully, with average treatment time of 28.3 min (range: 19.9-42.4 min). the AI-assisted ROIs automatically contouring took 1.55 min in average (range: 1.32-1.77 min), with an average DICE of 97.7% compared with modified contouring, and the average DICE was 95.7% for clinical tumor volume 1 (CTV1), 88.6% for CTV2, 73.6% for GTVn (cervical lymph node), 99.3% for 30 OARs. The automatic treatment plan averagely needed 3.5 min, and the pass rate of radiotherapy planning was 91.7% (22/24). The target coverage for PTVs for GTVp, CTV1, and CTV2 was 99.3%, 99.8%, 98.0% respectively. As for the dose of OARs, the average Dmax of brainstem was 5,583cGy; the Dmax of spinal cord was 3,467cGy; the Dmean of parotid was 3,285 cGy. The average monitor units of all patients was 643 MU and the delivery took 2.93 min. Patient compliance with respect to AIO workflow and total treatment time was excellent.

CONCLUSION

The AIO online radiotherapy was promising for NPC patients, with clinically acceptable AI assisted ROIs contouring and treatment planning, as well as favorable patient compliance to the AIO online workflow.

A Machine Learning Method to Predict Pathological Complete Response of Esophageal Cancer after Neoadjuvant Chemoradiotherapy with Clinicohematological Markers and MR Radiomics: A Multi-Center Study

PURPOSE/OBJECTIVE(S)

Nearly 30% of patients with local advanced esophageal cancer achieved pathological complete response (pCR) after neoadjuvant chemoradiotherapy (nCRT), who may benefit from organ-preservation strategy under accurate prediction of pCR. We aimed to develop and validate machine learning models based on clinicohematological markers and MR radiomics to accurately predict pCR of esophageal cancer after nCRT.

MATERIALS/METHODS

In this multi-center study, eligible patients with esophageal cancer who received baseline MR scan (T2-weighted image) and nCRT plus surgery were enrolled between September 2014 and September 2022 at institution 1 (training set) and between December 2017 and August 2021 at institution 2 (testing set). Pre-nCRT and post-nCRT blood test results were collected to calculate hematological markers. Models were constructed by machine learning based on clinicohematological markers and MR radiomics to predict pCR. Area under the curve (AUC) and cut-off analysis were used to evaluate model performances.

RESULTS

Totally 154 patients (81 in the training set and 73 in the testing set) were enrolled. The combined model integrating pre-nCRT monocyte-to-lymphocyte ratio and 6 radiomics features achieved AUC of 0.800 (95% CI 0.671-0.918) in the testing set, with sensitivity of 79.2% (95% CI 62.5%-95.8%), specificity of 83.7% (95% CI 73.5%-93.9%), positive predictive value of 76.0% (95% CI 62.5%-90.0%), and negative predictive value of 89.6% (95% CI 82.0%-95.8%).

CONCLUSION

A machine learning model based on clinicohematological markers and MR radiomics to predict pCR after nCRT for patients with esophageal cancer was developed and validated, providing a novel tool for personalized treatment. It is necessary to further validate in more large datasets.

Automated Contouring of Cervical Lymph Nodes and Clinical Target Volumes for Nasopharyngeal Carcinoma Based on Deep Learning and Experience Constraints

PURPOSE/OBJECTIVE(S)

Application of artificial intelligence (AI) for automated contouring of tumor volumes and organs at risk (OARs) for radiotherapy of nasopharyngeal carcinoma (NPC) leads to improved contouring accuracy and efficiency. However, few studies have involved the automated contouring of gross tumor volume of cervical lymph nodes (GTVn) and clinical target volumes (CTVs). In this work, we proposed an AI automated contouring tool for GTVn and CTVs for radiotherapy of NPC on the plain scans of planning compute tomography (CT).

MATERIALS/METHODS

In this retrospective study, plain scan datasets of planning CT covering the nasopharynx and neck from 139 patients with NPC between March 2022 and December 2022 were collected and divided into training, validation, and testing cohorts of 95, 24, and 20 patients, respectively. Ground truth contours of primary gross tumor volume (GTVp), GTVn (divided into GTVn_L in left neck and GTVn_R in right neck), CTVs (including high risk CTV1 contains GTVp and low risk CTV2 contains GTVp and cervical nodal levels) and OARs were delineated and were defined by consensus of two experts. We first proposed a three-dimensional (3D) U-net using GTVp and OARs as experience constrains to guide the automated delineation of GTVn and CTVs. The average Dice similarity coefficient (DSC) and average surface distance (ASD) were used to quantify the performance of the AI tool. Next, five prospective patients were enrolled for clinical evaluation of our AI tool. DSC between automated contours and radiation oncologist-revised contours and time consuming of the revision were record.

RESULTS

Clinical characteristics of 139 retrospective and 5 prospective patients are list in Table 1. In the independent testing set of 20 patients, our AI tool showed high performance in GTVn and CTVs contouring when compared with the ground truth contours. The mean DSC were 0.73 ± 0.07, 0.74 ± 0.05, 0.93 ± 0.03, and 0.88 ± 0.03, and the mean ASD were 1.01 ± 0.43 mm, 1.14 ± 0.61 mm, 0.51 ± 0.13 mm, 1.17 ± 0.43 mm for GTVn_L, GTVn_R, CTV1 and CTV2, respectively. In the five prospective patients, mean DSC were 0.74 ± 0.07, 0.74 ± 0.10, 0.95 ± 0.01 and 0.89 ± 0.04, respectively. The median time consuming for GTVn and CTVs revision was 2minutes and 10 seconds (range, 1 minutes to 3 minutes).

CONCLUSION

The proposed AI tool integrating clinical experience as constrains showed high accuracy for contouring GTVn and CTVs of NPC. With the assistance of AI contours, contouring efficiency could be probably increased, which is promising in online adaptive radiotherapy of NPC.

First Implementation of Full-Workflow Automation for Online Adaptive Radiotherapy of Nasopharyngeal Carcinoma

PURPOSE/OBJECTIVE(S)

The aim of this work is to established the technical characteristics and implementation procedures of an artificial intelligence (AI)-powered radiotherapy workflow that enables full-process automation for online adaptive radiotherapy (ART); and evaluate its feasibility and performance implemented for ART of nasopharyngeal carcinoma (NPC).

MATERIALS/METHODS

This single center, prospective study has been approved by the ethical committee of the institution. The online ART workflow was developed based on a CT-integrated linear accelerator. During the course of radiotherapy, the patient underwent daily pre-treatment fan-beam CT (FBCT) scan. Then the FBCT was automatically registered to the original planning CT and used to assess the need for the patient to implement ART according to radiation oncologist's discretionary. The online ART workflow incorporates critical radiotherapy procedures from re-simulation, auto-segmentation by integrating image fusion and deep learning method, auto-replanning, beam delivery, and in vivo quality assurance (QA) into one scheme, while the patient is on the treatment couch during the whole process.

RESULTS

From 2th April 2022 to 5th January 2023, 20 patients with newly-diagnosed, non-metastatic NPC were enrolled in this study. Only one-time online ART was performed for each patient, because that the appropriate timing for triggering online ART was explored in parallel with this study. According to radiation oncologists' discretionary, the median fraction for performing online ART was at 21 fractions (interquartile range, 19-24 fractions). All patients were well tolerated and successfully completed the treatment. For tumor targets contouring, minor revisions were required for automated contours of the primary gross tumor volume (GTVp) and clinical target volumes (CTVs, including CTV1 and CTV2), with the mean DSC between before and after revision of 0.91±0.042, 0.94 ± 0.042 and 0.91 ± 0.061, respectively; and much more revisions for the automated contours of cervical lymph nodes GTV (GTVn), with the mean DSC of 0.74 ± 0.28. The automated contours of normal tissues were clinically acceptable with little modifications. Median time consuming for auto-segmentation and revision was 9.5 minutes (min). For treatment planning, 18 automated plans (90%) were passed at their first auto-optimization and two plans (10%) were passed after further optimization of the dose coverage of CTVs by physicist; and the median time consuming for auto-planning was 6.2 min. Time consuming for other procedures were as follows: re-simulation, 2.3 min; plan evaluation, 3.3 min; beam delivery, 4.6 min; and the duration of the entire process was 25.9 min, range from 19.4 min to 32.5 min.

CONCLUSION

We successfully established an AI-powered online ART workflow for adaptive radiotherapy of NPC, and confirmed that current auto-segmentation and auto-replanning methods are powered enough to support the clinical application of its online ART.

Preserving Structural Consistency in the Generation of Synthetic CT in Pelvic MR-Only Radiation Treatment Planning

PURPOSE/OBJECTIVE(S)

MR-based synthetic CT (sCT) generation is necessary for MR-only radiotherapy to assist in radiation dose calculation, owing to no electronic density information in MR images. This study investigated the feasibility of synthesizing CT images from magnetic resonance (MR) images using generation antagonism networks (GANs) for MR radiotherapy of rectal cancer. Meanwhile, the transformer module and the contrast learning loss were introduced to improve the sCT.

MATERIALS/METHODS

The data set used in this study was the T2-weighted MR and CT image data of 108 patients with rectal cancer. Three-fold cross-validation was performed on all data sets. The transformer module was introduced into the plain CycleGAN, and the improved Patch Noise Contrastive Estimation (PatchNCE) loss was used as the loss function. The improved PatchNCE loss maintained the structural consistency of the MR and the synthetic CT by ensuring the consistency of the distribution of image patches on the MR-sCT image pair. The 2.5D images were taken as the input of our model, which refers to taking two consecutive adjacent layers in a specific layer. The CT-to-sCT image similarity was evaluated by metrics of mean absolute error (MAE), peak signal-to-noise ratio (PSNR), and Structure Similarity Index Measure (SSIM). The sCT dosimetric accuracy 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

The evaluation indicators of sCT images generated by our model were superior to the plain CycleGAN in the results of the three-fold cross-validation. MAE, PSNR and SSIM of our model were 42.850HU, 26.486 and 0.988, respectively, which were superior to 47.129HU, 25.167 and 0.978 of the plain CycleGAN. In addition, sCT generated by our model exhibited good continuity in the axial direction compared with plain CycleGAN. Furthermore, most of the relative differences in the DVH indicators were less than 1%.

CONCLUSION

The accuracy of sCT can be effectively improved by introducing a transformer module and comparative learning loss function. Moreover, all dosimetric differences were within clinically acceptable criteria for photon radiotherapy, demonstrating the feasibility of the MRI-only workflow for patients with rectal cancer.

Cost-Effectiveness of Short-Course Radiotherapy Based Total Neoadjuvant Therapy for Locally Advanced Rectal Cancer in China

PURPOSE/OBJECTIVE(S)

The phase III STELLAR (NCT02533271) trial demonstrated that four cycles of chemotherapy after short-course radiotherapy (SCRT-TNT) were not inferior to the standard care of long-course concurrent radiotherapy (LCRT) in patients with locally advanced rectal cancer (LARC). This study assessed the cost-effectiveness of SCRT-TNT versus LCRT in locally advanced rectal cancer in China on the basis of the STELLAR trial.

MATERIALS/METHODS

A Markov model was used to synthesize the healthcare costs and benefits of LARC patients based on results from the STELLAR trial. The model assumes that LARC who meet the inclusion criteria of the STELLAR trial experience four possible states: No Evidence of Disease (NED), locally recurrence, distant metastases, or any death from rectal cancer or other unrelated causes, where local recurrence continues to be classified as resectable and unresectable. The transition status period is 3 month, and 5 years is used to calculate direct medical costs and health benefits. The probabilities of states transition after SCRT-TNT or LCRT were derived from the results of the STELLAR trial and previous published article (Table.1). Costs were evaluated from the Chinese payer's perspective reported in early 2022 US dollars (US$1 = 6.78 Chinese Yuan). Sensitivity analyses were performed for key variables. Cost-effectiveness was evaluated using the incremental cost-effectiveness ratio and net monetary benefits. Effectiveness was defined as quality-adjusted life-years (QALYs). Willingness-to-pay (WTP) threshold was set at $43500/QALY. Data were collected from October 3, 2020, to September 20, 2021, and analyzed from November 15, 2020, to October 25, 2021.

RESULTS

During the 5-year horizon, for the base case scenario, SCRT-TNT incurred a lower total cost and higher QALYs compared with LCCRT. The total cost was $65767 and QALYs were 1.77 for SCRT-TNT; for LCCRT, the total cost was $72802 and QALYs were 1.64. This resulted in an ICER of -$ 55470.69 per QALY. Therefore, SCRT-TNT was a cost-saving and dominating treatment strategy compared with LCRT. Sensitivity analysis showed that ICERs were most sensitive to the parameters of distant metastases risk after treatment.

CONCLUSION

SCRT-TNT in locally advanced rectal cancer can be a cost-effective alternative to LCRT in China, and should be considered in appropriately selected patients.

A Multicenter Study on Deep Learning for Glioblastoma Auto-Segmentation with Prior Knowledge in Multimodal Imaging

PURPOSE/OBJECTIVE(S)

A precise radiotherapy plan is required to ensure accurate delineation of gross tumor volumes (GTV) and clinical target volumes (CTV1 and CTV2) of glioblastomas (GBMs). However, traditional manual delineation is labor intensive and highly dependent on oncologists' experience. To construct and evaluate a deep learning-based automatic delineation method using prior knowledge in multimodal medical imaging to automate precise GTV, CTV1 and CTV2 contouring in GBM patients.

MATERIALS/METHODS

We retrospectively collected the CT and MRI scans of 55 eligible patients with histologically proven high-grade glioma (HGG) from an institute, these scans were performed with non-enhanced CT (CT), contrast-enhanced T1-weighted (T1C) and T2-FLAIR (T2F) sequences. We proposed a two-stage automatic segmentation framework (PKMI-Net) for GTV, CTV1 and CTV2 based on deep learning using prior knowledge in multimodal medical imaging, and its segmentation performance was evaluated with dice similarity coefficient (DSC), 95% Harsdorff distance (HD95), average surface distance (ASD) and relative volume difference (RVD). To further investigate the generalizability of our method, we designed and conducted two evaluation strategies (Mix and Cross) on four multicenter datasets (including 55 patients, 37 patients, 21 patients and 35 patients).

RESULTS

The evaluation results with an 11-patient test set from the single institute were summarized in Table 1, the proposed method demonstrated the best accuracy in segmenting, respectively, GTV, CTV1 and CTV, achieving a DSC of 0.94, 0.95 and 0.92; HD95 of 2.07 mm, 1.18 mm and 3.80 mm; ASD of 0.69 mm, 0.39 mm and 1.13 mm and RVE of 5.50%, 3.97% and 9.68%. In the multicenter evaluation, the segmentation performance of our method implemented with the Cross strategy was comparable to that with the Mix strategy, demonstrating that our method had high and stable generalizability across multicenter datasets in automatically segmenting GTV, CTV1 and CTV2.

CONCLUSION

Our proposed method achieved promising results in automatically segmenting gliomas across various datasets, which could improve the quality and efficiency of glioblastoma radiotherapy.

Virtual-Collimator Based Spatial Dose Modulation for Proton GRID Therapy

PURPOSE/OBJECTIVE(S)

Compared to conventional proton therapy, the proton GRID therapy can substantially improve normal tissue protection (with the delivery of spatially-modulated peak-valley dose pattern to normal tissues) while maintaining the tumor control efficacy (with the delivery of uniform dose pattern to tumor targets). The realization of proton GRID often relies on the use of physical collimators to shape the spatial dose distribution. However, the physical collimator may increase neutron dose, decrease delivery efficiency, and limit the freedom for patient positioning. Here we propose a virtual-collimator (VC) method for proton GRID. This new approach can generate peak-to-valley pattern with high peak-to-valley dose ratio (PVDR), without using a physical collimator.

MATERIALS/METHODS

The principle behind the VC method to modulate the spatial dose distribution consists of two major steps: (1) the primary beam is essentially halved, i.e., the beamlets are interleaved, so that the organ-at-risk (OAR) plane has the peak-valley dose pattern, while the target plane also has the valley dose; (2) the complementary beam is added with half complementary beamlets to fill in the previously valley-dose positions at the target plane, so that the target dose is uniform, while on the other hand, the complementary beam is angled slightly from the primary beam, so that the OAR still has the peak-valley dose pattern. Moreover, on top of VC, we also utilize sparsity regularization method using total variation and L1 sparsity (TVL1) to further jointly optimize PVDR and dose objectives, namely VC-TVL1.

RESULTS

VC and VC-TVL1 were validated in comparison with conventional proton GRID treatment planning method via IMPT ("CONV") and TVL1-based proton GRID treatment planning method without VC ("TVL1"), for a prostate case with single-beam (270° only) or two-beam (90° and 270°) scenarios. As shown in the table, the results show that VC can indeed modulate spatial dose with higher PVDR than CONV or even TVL1. VC had higher spatial modulation frequency with smaller peak-to-peak distance than TVL1. Moreover, VC+TVL1, as the synergy of VC and TVL1, further improved PVDR from VC or TVL1 alone.

CONCLUSION

A new way to deliver proton GRID therapy without a physical collimator is developed using the VC method. The VC method can be synergized with TVL1 optimization algorithm to further jointly optimize PVDR and dose objectives.

Prediction of Response to Neoadjuvant Chemoradiotherapy Combined with Pembrolizumab in Esophageal Squamous Cell Carcinoma with CT/FDG PET Radiomic Signatures Based on Machine Learning Classification

PURPOSE/OBJECTIVE(S)

PALACE-1 trial has confirm that the addition of pembrolizumab to neoadjuvant chemoradiotherapy (NCRT) improves the pathological complete response(pCR) for esophageal squamous cell carcinoma (ESCC), which might be a novel treatment strategy for ESCC. In the present study, we aim to establish a machine learning model to predict the local response to NCRT+ pembrolizumab for ESCC by using pretreatment 18-fluorodeoxyglucose positron emission tomography (FDG PET) and contrast-enhanced plan CT images.

MATERIALS/METHODS

A total of 65 cases treated with NCRT+ pembrolizumab followed by surgery were prospectively enrolled for analysis from 2019-2022. Each patient contains a contrast-enhanced plan CT and FDG PET images. 52 patients were randomly divided into training set and 13 patients were used as test set. The Extraction of radiomics features was performed using an open-source Python library PyRadiomics automatically. Features were computed according to the radiologist-drawn ROIs on both CT and PET images. In the feature selection stage least absolute shrinkage and selection operator (LASSO) was utilized on CT features and PET features separately. Four different machine learning models were implemented: Support Vector Machine (SVM), Logistic Regression (LR), Random Forest (RF) and XGBoost (XGB). The features selected by LASSO regression were used as model input and the output of the model is "pCR" or "non-pCR". To find the optimal parameter, the 5-fold cross-validation method was used in the training stage. In this study, we use accuracy, sensitivity and specificity as the metrics to evaluate the performance of the model on the testing cohort. The predictive performance of the model was assessed using the area under curve (AUC) of the receiver operating characteristics curve (ROC).

RESULTS

Of the 65 cases treated with NCRT+pembrolizumab, 35 patients archived pCR (53.8%), and 30 archived non-pCR. 1684 radiomics features were extracted from each case, and half of them (842 features) were from CT and others were from PET. Among the machine learning models mentioned above SVM achieves the most promising performance on the evaluation metrics. Accuracy, sensitivity, specificity and AUC score on test set were 0.692, 0.833, 0.571 and 0.786 for CT features and 0.615, 0.667, 0.571 and 0.762 for PET features, respectively. For CT+FDG PET fused features accuracy, sensitivity, specificity and AUC score on test set were 0.769, 0.667, 0.857 and 0.833.

CONCLUSION

In this study, we performed several different machine learning models to predict the response to NCRT+ pembrolizumab among ESCC based on the extracted radiomics features from CT and FDG PET images. The best-performing model based on radiomics features of CT and PET images could identify non-pCR to NCRT + pembrolizumab in EC patients.

Dosimetric Benefits of Online Adaptive Radiotherapy in Nasopharyngeal Carcinoma

PURPOSE/OBJECTIVE(S)

Online adaptive radiotherapy (ART) has the advantage of compensating for potential underdosing to targets and overdosing to organs-at-risk (OARs) caused by variations in patient anatomy and tumor geometry. Artificial intelligence (AI)-assisted rapid generation of new plans makes online ART possible. We aimed to evaluate the dosimetric benefits of online ART on tumor coverage and OARs sparing in nasopharyngeal carcinoma (NPC).

MATERIALS/METHODS

Twenty patients diagnosed with NPC (19 with stage III and 1 with stage II according to the 8th edition of the AJCC/UICC staging system) who underwent definitive radiotherapy or concurrent chemoradiotherapy and received online ART on CT-Linac between April 2022 and December 2022 were included in this study, consisting of 14 males and 6 females with a median age of 48 years (range: 29-68 years). The prescription dose was 6996 cGy/33 fractions for primary gross tumor volume (GTVp), 6600-6996 cGy/33 fractions for gross tumor volume of nodes (GTVn), 6006 cGy/33 fractions for high-risk clinical tumor volume (CTV1), 5412 cGy/33 fractions for low-risk clinical tumor volume (CTV2). The majority of the patients (15/20) received online ART during the fourth to fifth week of their radiotherapy treatment The auto-segmented contours and auto-plan generated by AI were manually reviewed and edited by radiotherapists and physicists. The paired samples t-test was used to compare the dose and volumes metrics of targets and OARs between scheduled plan and online ART plan.

RESULTS

The results of this study showed that compared to the scheduled plan, the online ART plan resulted in significant reductions in the volumes of all targets and 8/12 OARs (temporal lobes, optic nerves, lenses, eyes, parotids, submandibulars, mandibles, and thyroid) (P<0.05). The online ART plan also improved target coverage, with D98% for GTVp in the scheduled plan compared to the online ART plan being 7063.4 ± 76.1 cGy and 7096.1 ± 53.9 cGy (P = 0.1), CTV1 being 6266.7 ± 114.9 cGy and 6208.7 ± 54.7 cGy (P<0.05), and CTV2 being 4142.5 ± 1700.9 cGy and 5416.4 ± 23.8 cGy (P<0.01), respectively. The dose to all 12 OARs was reduced with the use of online ART, with 5/12 OARs showing statistical significance. The D0.03cm3 for the spinal cord in the scheduled plan and online ART plan were 3630.9 ± 197.6 and 3454.1 ± 132.0 cGy; for the temporal lobes were 7075.2 ± 303.0 and 6994.2 ± 345.1 cGy; and 4396.0 ± 2575.0 and for the pituitary were 4214.5 ± 2499.2 cGy. Meanwhile the Dmean for the eyes in the scheduled plan and online ART plan was 769.0 ± 232.0 and 714.8 ± 200.1 cGy; and for the mandibles were 3187.7 ± 211.5 and 3066.0 ± 152.1 cGy.

CONCLUSION

Online ART was effective in protecting most of the OARs in NPC patients, while simultaneously indicating a trend towards enhancing target coverage. This study demonstrated the promising potential of online ART for patients with NPC. This approach will be tested in an upcoming phase III trial.

2V-CBCT: Two-Orthogonal-Projection Based CBCT Reconstruction and Dose Calculation from Real CBCT Projection Data

PURPOSE/OBJECTIVE(S)

Not all radiation therapy (RT) treatments/fractions have CBCT acquired, but two orthogonal projections (i.e., KV radiography) are always available. This work demonstrates the feasibility of two-orthogonal-projection-based CBCT (2V-CBCT) reconstruction and dose calculation for RT from real CBCT projection data, which is the first 2V-CBCT feasibility study using real projection data, to the best of our knowledge.

MATERIALS/METHODS

2V-CBCT is a severely ill-posed inverse problem for which we propose a coarse-to-fine learning strategy. First, a 3D deep neural network that can extract and exploit the inter-slice and intra-slice information is adopted to predict the initial 3D volumes. Then, a 2D deep neural network is utilized to fine-tune the initial 3D volumes slice-by-slice. During the fine-tuning stage, a perceptual loss based on multi-frequency features is employed to enhance the image reconstruction. Dose calculation results from both photon and proton RT demonstrate that 2V-CBCT provides comparable accuracy with full-view CBCT based on real projection data.

RESULTS

The proposed method was evaluated on real HN data acquired from on-board CBCT scanners rather than the low-resolution simulated data or down-sampled data. Both visual assessment and quantitative analysis demonstrate that the proposed coarse-to-fine learning strategy has the potential to produce satisfactory volumetric images from two orthogonal projections. Furthermore, we assessed the utility of 2V-CBCT in RT. The results show that the dose distribution maps, dose-volume histograms, and dose parameters calculated using 2V-CBCT have comparable accuracy with the counterparts calculated using the corresponding full-view CBCT for both photon and proton RT. In the table, the methods under comparison are pCT (planning CT), FV-CBCT (CBCT reconstructed with full-view projection data), and 2V-CBCT (CBCT reconstructed with two orthogonal projections).

CONCLUSION

A new effective 2V-CBCT reconstruction method is proposed and validated using real CBCT projection data, which can potentially provide comparable dose calculation accuracy for both photon and proton RT.

Hsa_circ_0003928 regulates the progression of diabetic nephropathy through miR-136-5p/PAQR3 axis

BACKGROUND

Diabetic nephropathy (DN) is one of the complications of diabetes and has a high mortality, but its specific pathogenesis is not clear. In recent years, researches on the mechanism of circRNAs in DN have been proved a lot, whereas the functional mechanism of circ_0003928 in DN remains open and it must be investigated to value its important role in DN prevention.

METHODS

HK-2 cells were treated with high glucose (HG), normal glucose (NG) or Mannitol. Cell counting kit-8 (CCK8) and 5-ethynyl-2'-deoxyuridine (EdU) assays were performed to detect cell proliferation. Enzyme-linked immunosorbent assay (ELISA) was applied to analyze malondialdehyde (MDA) and superoxide dismutase 1 (SOD) levels. Flow cytometry and western blot were preformed to measure cell apoptosis. Real-time quantitative PCR (RT-qPCR) was used to test the levels of circ_0003928, miR-136-5p and progestin and adipoQ receptor family member 3 (PAQR3) mRNA. Western blot was executed to detect Bcl2 associated X (Bax), B cell leukemia/lymphoma 2 (Bcl2), smooth muscle (αSMA), apolipoprotein (C-IV) and PAQR3 levels. Luciferase reporter assay and RNA pull-down assay were used to analyze the target relationship between miR-136-5p and circ_0003928 or PAQR3.

RESULTS

Circ_0003928 and PAQR3 expression were up-regulated, whereas miR-136-5p was decreased in DN serum and HG-induced HK-2 cells. Circ_0003928 knockdown promoted cell proliferation, and inhibit cell apoptosis, oxidative stress, and fibrosis in HK-2 cells under HG condition. MiR-136-5p silencing overturned the protective effects of si-circ_0003928 on HG-induced HK-2 cells. MiR-136-5p was targeted by circ_0003928 and directly targeted PAQR3. Overexpression of PAQR3 counteracted the inhibitory functions of circ_0003928 knockdown or miR-136-5p overexpression on HG-induced HK-2 cell injury.

CONCLUSION

Circ_0003928 acted as a sponge of miR-136-5p to up-regulating PAQR3 expression, and then regulate the proliferation, oxidative stress, fibrosis and apoptosis in HG-induced HK-2 cells.

Small Samples and Low-Cost Auto-Segmentation Method for Pelvic Organ-at-Risk Segmentation in Magnetic Resonance Images Using Deep-Learning

PURPOSE/OBJECTIVE(S)

In radiotherapy, magnetic resonance (MR) imaging has higher contrast of soft tissue, and no radiation compared with computed tomography (CT) scanning. Due to the high-cost of manual annotation, the deep-learning based automatic organ-at-risk (OAR) and target delineation algorithms are in high-demand, but the collecting of large amounts of high-quality annotated datasets remains difficulty. In this paper, we proposed a low-cost OAR segmentation method with semi-supervised annotation using small annotation samples of pelvic MR images.

MATERIALS/METHODS

This study consisted of 94 patients diagnosed with rectal cancer from April 2018 to March 2021 at Peking University People's Hospital. We used 17 slices of MR images with annotation and 78 slices without annotation to train a deep-learning based segmentation model. The bladder, femoral heads, rectum and small intestine were selected as OAR. Semi-supervised method and ensemble learning were used for generating training set using small sample with annotation. Post-processing algorithm was used to correct the self-annotation data. Two of 14 annotation samples were set as test set. As for un-labeled images, 40 of them were set as semi-supervised annotation train set, the rest were test set. Besides, both 2D and 3D auto-segmentation networks were evaluated.

RESULTS

The dice of bladder, femoral head left and right, rectum and small intestine between segmentation results and reference masks is 0.947, 0.983, 0.981, 0.900, 0.845 only using self-annotation and post-processing method of 2D segmentation model. And the dice of corresponding OAR is 0.871, 0.975, 0.975, 0.783, 0.724 using 3D segmentation network, 0.885,0.982, 0.982, 0.882, 0,814 using 2D segmentation network with supervised method (nnUNet). The 2D model outperformed 3D model with better segmentation performance, shorter inference time and fewer parameters.

CONCLUSION

The results proved that we can train a multi-OAR segmentation model only using small annotation samples and other unlabeled samples. Ensemble learning and post-processing methods are necessary for semi-supervised data annotation. For anisotropy data, 2D model shows better performance than 3D models.

Unsupervised Domain Adaptation of Auto-Segmentation on Multi-Source MRIs

PURPOSE/OBJECTIVE(S)

Deep learning has achieved great success in medical image segmentation. Most existing deep learning (DL) approaches make no adjustments to the model prior to inference. These models can perform well on the data of the same distribution, but their performance usually degrades when applied to the images from different source, i.e., different scanners. To tackle the problem caused by domain shift, we proposed an unsupervised domain adaptation (UDA) method based on entropy minimization and physical consistency constraints.

MATERIALS/METHODS

The proposed method combines feature-level and instance-level domain adaptation techniques to transfer knowledge from the source to the target domain. Specifically, the feature-level adaptation technique uses a graph-based entropy minimization to reduce the discrepancy between the source and target domains. The instance-level adaptation technique employs a novel consistency loss to regularize the physical consistency of the same object, such as volume, length, and centroid, thus improving the segmentation accuracy of the target domain. A collection of 93 abdominal MR images, comprising 45 cases from a 0.35T MRI scanner (TRUFI) and 48 cases from a 1.5T MRI scanner (T2), was utilized to evaluate the effectiveness of the proposed method. The contours of 6 organs-at-risk were delineated by a senior radiation oncologist, serving as the ground truth. Three models, the source model (SRC) trained on the source domain, the target model (TGT) trained on the target domain, and the UDA model adapted from the source domain to the target domain, were compared on the target domain using the Dice Similarity Coefficient (DSC).

RESULTS

In the experiment of 0.35T-to-1.5T, the proposed UDA method outperformed the source model, achieving an average DSC score of 0.82 ± 0.11, compared to 0.58 ± 0.23 (SRC) and 0.85 ± 0.09 (TGT), respectively. In the inverse experiment 1.5T-to-0.35T, the UDA model achieved an average DSC score of 0.79±0.13, compared to DSCs of 0.52 ± 0.25 and 0.81 ± 0.11 for the SRC and TGT respectively. The UDA method yielded a significant improvement of 46%, compared to the SRC. Particularly, OARs (organ at risk) with higher deformability such as the stomach and duodenum achieved a 58% and 63% improvement in performance, respectively.

CONCLUSION

This work presents a compelling approach of UDA for auto-segmentation on multi-source MRIs. Experimental results demonstrate that the UDA effectively improve the segmentation performance of the source model in the target domain, resulting in a more robust segmentation model.

Vertex Position Optimization for LATTICE Therapy

PURPOSE/OBJECTIVE(S)

LATTICE radiation therapy (RT) aims to deliver 3D heterogenous dose of high peak-to-valley dose ratio (PVDR) to the tumor target, with peak dose at lattice vertices inside the target and valley dose for the rest of the target. In current clinical practice the lattice vertex positions are constant during treatment planning. This work proposes a new LATTICE plan optimization method that can optimize lattice vertex positions as plan variables, which is the first lattice vertex position optimization study to the best of our knowledge.

MATERIALS/METHODS

The new LATTICE treatment planning method optimizes lattice vertex positions as well as other plan variables (e.g., photon fluences or proton spot weights), with optimization objectives for target PVDR and organs-at-risk (OAR) sparing. To satisfy mathematical differentiability, the lattice vertices are approximated in sigmoid functions. For geometric feasibility, proper geometry constraints are enforced onto the lattice vertex positions. The lattice vertex position optimization problem is solved by iterative convex relaxation method, where lattice vertex positions and photon/proton plan variables are jointly updated via the Quasi-Newton method.

RESULTS

Both photon and proton LATTICE RT were considered, and the optimal lattice vertex positions in terms of plan objectives were found by solving all possible combinations on given discrete positions via heuristic searching based on standard IMRT/IMPT, which served as the ground truth for validating the new LATTICE method ("NEW"). That is, the plan with the smallest optimization objective ("BEST"), the plan with the median optimization objective ("MID"), and the plan with the largest optimization objective ("WORST") were selected as the reference plans to be compared with NEW. The table was for an abdomen case with the large bowel as the OAR, where the parameters are total optimization objective f, the mean valley dose of target Dvalley, the mean peak dose of target Dpeak, PVDR = Dpeak/Dvalley, and the mean dose of large bowel Dbowel. The unit of doses is Gy. The results in the table show that the new method indeed provided the optimal lattice vertex positions with the smallest optimization objective, the largest target PVDR, and the best OAR sparing.

CONCLUSION

A new LATTICE treatment planning method is proposed and validated that can optimize lattice vertex positions as well as other photon or proton plan variables for improving target PVDR and OAR sparing.

The Pathogenesis of COVID-19-Related Taste Disorder and Treatments

COVID-19, mainly manifested as acute respiratory distress syndrome, has afflicted millions of people worldwide since 2019. Taste dysfunction is a common early-stage symptom of COVID-19 infection that burdens patients for weeks or even permanently in some cases. Owing to its subjectivity and complexity, the mechanism of taste disorder is poorly studied. Previous studies have reported that the COVID-19 entry receptors are highly expressed in taste buds, thereby intensifying the cytocidal effect. Taste receptor cells are vulnerable to inflammation, and the COVID-19-induced cytokine storm causes secondary damage to taste function. Interferon and various proinflammatory cytokines can trigger cell apoptosis and disrupt the renewal of taste bud stem cells. This immune response can be further enhanced by the accumulation of Angiotensin II (Ang II) caused by an unbalanced local renin-angiotensin system (RAS) system. In addition, severe acute respiratory syndrome coronavirus type 2 (SARS-CoV-2) is neurotropic and can invade the brain through the olfactory bulb, affecting the nervous system. Other factors, such as host zinc deficiency, genetic susceptibility, sialic acid, and some neurotransmitters, also contribute to the pathogenesis process. Although several medical interventions have displayed effectiveness, only a few strategies exist for the treatment of postinfectious dysgeusia. Stem cell-based taste regeneration offers promise for long-term taste disorders. Clinical studies have demonstrated that stem cells can treat long COVID-19 through immune regulation. In dysgeusia, the differentiation of taste bud stem cells can be stimulated through exogenous epithelial-derived and neural-derived factors to regenerate taste buds. Tongue organoids are also emerging as functional taste buds, offering new insights into the study of taste regeneration. This review presents the current evidence of the pathogenesis of COVID-19-related dysgeusia, summarizes currently available treatments, and suggests future directions of taste regeneration therapy.

[Recommendations for the diagnosis and treatment of IgG4-related disease in China]

IgG4-related disease (IgG4-RD) is an immune-mediated fibroinflammatory condition characterized by tumefactive lesions in multi-organs. It is a novel entity presented by variable manifestations. In recent years, there has been progress toward recognizing IgG4-RD. However, the diagnosis and treatment of IgG4-RD still present challenges due to insufficient experience. To address this, the Chinese Rheumatology Association has developed standardized guidelines for the diagnosis and treatment of IgG4-RD based on domestic and international experience. These guidelines aim to enhance the understanding and management of IgG4-RD, ultimately improving the prognosis for patients with IgG4-RD.

Deep Learning-Based Synthesis of Contrast-Enhanced MRI for Automated Delineation of Primary Gross Tumor Volume in Radiotherapy of Nasopharyngeal Carcinoma

PURPOSE/OBJECTIVE(S)

Contrast-enhanced MRIs are necessary to delineate the primary gross tumor volume (GTVp) in radiotherapy of nasopharyngeal carcinoma (NPC). However, using contrast agents to scan contrast-enhanced MRIs is not applicable to some patients due to metal implants or their allergy, and it increases the treatment cost of patients. To address these problems, this work aims at synthesizing contrast-enhance MRIs from unenhanced MRIs by implementing generative adversarial network (GAN).

MATERIALS/METHODS

In this work, 324 MRI datasets of patients with NPC were retrospectively collected between September 2016 and September 2017 from a single institute. MRI examinations were performed with un-enhanced T1-weighted (T1) and T2-weighted (T2) sequences, and contrast-enhanced T1-weighted (T1C) and fat-suppressed T1-weighted (T1FSC) sequences. We designed and developed a modified pix2pix network to synthesize T1C (sT1C) and T1FSC (sT1FSC) from real T1. The end of the generator in this network was assembled with multiple heads (the classification head and gradient head) to learn more representation information and features from real images, the discriminator in this network distinguished whether the synthesized image is real and fake and supervised that the generator outputs more realistic synthesized image. We verified the performance of the synthesized images for automated delineation of GTVp. In an independent testing set of 11 patients, the synthesized sT1C and sT1FSC were inputted into the segmentation deep learning network along with their corresponding T1 and T2 sequences to generate GTVp contours. Delineation performance of the synthesized images and real images for automated delineation were evaluated by dice similarity coefficient (DSC), and average surface distance (ASD), using human expert contours as the ground truth.

RESULTS

In automated contouring of GTVp for NPC, the segmentation deep learning network using one or two synthesized MRIs showed equivalent performance when compared with the automated contours which generated from four real MRI sequences. Mean DSCs between automated contours by sT1C-replaced or sT1C and sT1FSC-replaced network and ground truth contours were 0.726 ± 0.143 and 0.711 ± 0.157, respectively, slightly inferior to that of contours generated from four real MRI sequences (0.740 ± 0.154, both P >0.05). In terms of mean ASD, there was also no significant difference between automated contours generated from synthesized images and real images (3.056 ± 4.216 mm and 3.537 ± 4.793 mm vs. 3.124 ± 4.637 mm; both P > 0.05).

CONCLUSION

We proposed an MRI-synthesis method based on GAN and the synthesized contrast-enhanced MRIs performed equivalent as the real contrast-enhanced MRIs in the automated delineation of gross tumor volume for radiotherapy of NPC.

Clinical effectiveness of semaglutide on weight loss, body composition, and muscle strength in Chinese adults

OBJECTIVE

The aim of this study was to investigate the clinical effectiveness of semaglutide on weight loss, body composition and muscle strength in the Chinese population with obesity.

PATIENTS AND METHODS

Data were retrospectively analyzed for participants prescribed semaglutide in 2021 and 2022 from a Chinese weight management clinic. Changes in weight, body composition, biochemical indicators, calf circumference and handgrip strength were collected. Body fat and skeletal muscle were also measured using the bioelectrical impedance analysis. Paired t-test was used to compare the values after 6 months of treatment with the baseline values.

RESULTS

A total of 53 obese patients received 24 weeks of lifestyle intervention plus semaglutide treatment. 10 patients who failed to adhere to the follow-up were excluded, and 43 patients were studied. The average baseline body mass index (BMI) was 33.0 kg/m2, and the average body weight was 90.0 kg. After 6 months of treatment, the patient's weight was significantly reduced by 9.9 ± 3.9 kg (p < 0.001), and the weight loss percentage was 11.2 ± 4.5% (p< 0.001). The proportion of patients with weight loss ≥ 5% and ≥ 10% was 93% and 54%, respectively. Fasting blood glucose, fasting insulin, Homeostasis Model Assessment of Insulin Resistance (HOMA-IR) index, blood uric acid and blood lipid levels also decreased after treatment. Body composition analysis showed that the loss of skeletal muscle mass was 1.4 ± 1.3 kg (p < 0.001), which was significantly less than the loss of fat mass of 5.6 ± 3.7 kg (p < 0.001). By percentage, the fat mass loss was 15.6 ± 10.1%, and the muscle mass loss was 4.8 ± 4.4% (p < 0.001). The visceral fat area was significantly reduced by 24.4 ± 17.7 cm (p < 0.001). There was no significant change in skeletal muscle index (8.1 ± 1.0 kg/m2 at baseline and 7.9 ± 1.0 kg/m2 at 24 weeks). The calf circumference (42.6 ± 3.6 cm at baseline, 41.2 ± 3.8 cm at 24 weeks) and grip strength (33.3 ± 9.5 kg at baseline, 32.3 ± 9.0 kg at 24 weeks) did not decrease significantly. The main adverse reactions were mild gastrointestinal dysfunction (nausea, diarrhea and vomiting), without ketoacidosis.

CONCLUSIONS

In a real-world setting, semaglutide can reduce the weight and fat of obese patients while effectively maintaining muscle mass and muscle strength.