BACKGROUND:Craniospinal irradiation (CSI) is a complex treatment requiring precise delineation of extended plan target volumes (PTV) and multiple organs at risk (OARs). The process traditionally demands substantial time and expertise from physicians and medical physicists. While artificial intelligence (AI) has shown promise in auto-contouring and auto-planning in other tumors, its application to CSI remains limited due to challenges in model generalizability and outlier sensitivity. METHODS:We developed an automated workflow integrating deep learning-based auto-contouring and machine learning-enhanced rapidplan. A DPNUNet-based auto-contouring model was trained on 91 CSI patients to delineate the PTV and OARs. A knowledge-based planning (KBP) model was iteratively refined using machine learning algorithms to exclude dosimetric and geometric outliers. The performance was evaluated by comparing 20 manual plans (MP) and rapidplan (RP) using dosimetric indices. RESULTS:The auto-contouring model achieved dice coefficients of 0.93 for PTV and > 0.85 for major OARs, reducing contouring time from 1-2 h to 10 min. RP met clinical targets while demonstrating superior OAR sparing. However, RP showed higher PTV hotspots and required more monitor units. The automated workflow reduced the contouring and planning time by 75%, from 6-8 h to 1 h. CONCLUSION:The proposed framework maintained clinical plan quality while significantly improving efficiency, demonstrating feasibility for standardized CSI planning.

Accurate segmentation of organs at risk in the head and neck is essential for radiation therapy, yet deep learning models often fail on small, complexly shaped organs. While hybrid architectures that combine different models show promise, they typically just concatenate features without exploiting the unique strengths of each component. This results in functional overlap and limited segmentation accuracy. To address these issues, we propose a high uncertainty region-guided multi-architecture collaborative learning (HUR-MACL) model for multi-organ segmentation in the head and neck. This model adaptively identifies high uncertainty regions using a convolutional neural network, and for these regions, Vision Mamba as well as Deformable CNN are utilized to jointly improve their segmentation accuracy. Additionally, a heterogeneous feature distillation loss was proposed to promote collaborative learning between the two architectures in high uncertainty regions to further enhance performance. Our method achieves SOTA results on two public datasets and one private dataset.

Objectives. Assessing the efficacy of radiotherapy in patients with high-grade gliomas (HGGs) is challenging due to the occurrence of pseudo-progression and radionecrosis. This study introduces a directed graph network leveraging MR image features at multiple time points to accurately predict radiotherapy sensitivity. Approach. A total of 120 HGG patients were enrolled and randomly divided into training and internal validation datasets (3:1). 29 cases from multicenter data were included as the external test dataset. Comprehensive clinical records, simulation CT scans, pre-radiotherapy MR images, and mid-treatment imaging for re-planning were collected. Radiosensitivity was classified into sensitive and resistant groups based on recurrence within one year post-radiation. A directed-graph multi-temporal graph convolution network (D-MTGCN) was developed to integrate MR image features across multiple time points during radiotherapy. The D-MTGCN incorporated graph construction schemes accounting for both the radiation target and adjacent regions. Main results. Our D-MTGCN achieved optimal performance, yielding an area under curve of 0.98 and an accuracy (ACC) of 0.95. Beside, this model has a significant higher predictive efficacy compared with clinical model and response assessment in neuro-oncology 2.0 criteria (p < 0.01). Moreover, D-MTGCN outperformed support vector machine and STGCN using initial time point with an ACC of 0.87 and 0.93 respectively. Significance. The MTGCN model demonstrates high ACC in predicting radiotherapy sensitivity and outcomes for HGG using short-term MRI sequences. This tool can assist clinicians in making timely and precise treatment decisions.

ABSTRACT Purpose There is ongoing debate regarding the therapeutic approach and prognosis for IDH‐mutant grade 4 astrocytoma, a newly defined subtype of diffuse glioma in the 2021 WHO classification system for central nervous system tumors (WHO CNS 5). The aim of this study was to explore the clinical outcome and prognosticators for newly diagnosed IDH‐mutant grade 4 astrocytoma based on our single institutional data. Methods This retrospective analysis included 53 consecutive patients with newly diagnosed IDH‐mutant grade 4 astrocytoma, who underwent radiotherapy between September 2021 and December 2023. All patients were administered concurrent and adjuvant temozolomide. Eleven patients received adjuvant tumor‐treating fields (TTFields). Results The median follow‐up was 15.7 months. Twenty patients had tumor relapse; three patients died, all of whom were without TTFields therapy. The median PFS for the entire cohort was 19.3 months, and the median OS was not reached. Univariate analysis indicated patients younger than 40 years (p = 0.11) or without homozygous deletion of CDKN2A/B (p = 0.11) tended to have better PFS. In addition, the TTFields group tended to have longer median PFS than the non‐TTFields group in both analyses before and after propensity score matching (PSM) (24.4 vs. 18.5 months, p = 0.097, before PSM; 24.4 vs. 15.9 months, p = 0.080, after PSM). No significant independent prognostic factor was found in the multivariate analysis. Conclusions The study reveals important insights into clinical practice for IDH‐mutant grade 4 astrocytoma. Younger age and tumor without deleted CDKN2A/B might be predictive of better outcomes. The addition of TTFields trended towards improved PFS, necessitating prospective clinical trials for further investigation.

Purpose To compare the dosimetry and biological risk of volumetric modulated arc therapy (VMAT), helical tomotherapy (HT) and cyberKnife (CK) in the treatment of lung oligometastases. Methods and materials This retrospective study included a cohort of 21 lung oligometastasis patients, each with 2 or 3 lesions, who had previously undergone stereotactic body radiation therapy (SBRT). VMAT, HT and CK plans were made for each patient. The dose distribution of planning target volume (PTV) and organs at risk (OARs) were evaluated. Three biological risks were evaluated, namely radiation pneumonitis (RP), coronary artery disease (CAD) and congestive heart failure (CHF). Monitor Units (MUs) and beam-on-time were also recorded. Results All techniques were able to produce clinically deliverable plans. The expected biological risks for VMAT plans, CK plans, and HT plans were 6.69%, 5.05%, 5.88% for RP, 1.20%, 1.15%, and 1.17% for CAD, 1.26%, 1.19%, and 1.22% for CHF. The expected risks of RP were slightly lower in CK plans compared to VMAT and HT plans (p < 0.001), with VMAT plans showing the highest expected risks. For central lung cancer, the expected CAD risks of CK and HT plans were lower than those of VMAT plans (p < 0.05). The delivery efficiency of VMAT plans was significantly higher than that of CK plans and HT plans. Conclusions All three techniques, VMAT, HT, and CK, meet the therapeutic requirements for target coverage and dose constraints for OARs. Although there are statistical differences, the difference between the expected risk values of RP and CAD is very small, so the clinical manifestations may not show differences.

Background and Purpose: This study aimed to compare the dosimetric attributes of two multi-leaf collimator based techniques, HyperArc and Incise CyberKnife, in the treatment of brain metastases. Material and Methods: 17 cases of brain metastases were selected including 6 patients of single lesion and 11 patients of multiple lesions. Treatment plans of HyperArc and CyberKnife were designed in Eclipse 15.5 and Precision 1.0, respectively, and transferred to Velocity 3.2 for comparison. Results: HyperArc plans provided superior Conformity Index (0.91 +/- 0.06 vs. 0.77 +/- 0.07, p < 0.01) with reduced dose distribution in organs at risk (D-max, p < 0.05) and lower normal tissue exposure (V4Gy-V20Gy, p < 0.05) in contrast to CyberKnife plans, although the Gradient Indexes were similar. CyberKnife plans showed higher Homogeneity Index (1.54 +/- 0.17 vs. 1.39 +/- 0.09, p < 0.05) and increased D-2% and D-50% in the target (p < 0.05). Additionally, HyperArc plans had significantly fewer Monitor Units (MUs) and beam-on time (p < 0.01). Conclusion: HyperArc plans demonstrated superior performance compared with MLC-based CyberKnife plans in terms of conformity and the sparing of critical organs and normal tissues, although no significant difference in GI outcomes was noted. Conversely, CyberKnife plans achieved a higher target dose and HI. The study suggests that HyperArc is more efficient and particularly suitable for treating larger lesions in brain metastases.

Abstract OBJECTIVES There is ongoing debate regarding the efficacy of Tumor Treating Fields (TTFields) for treating IDH-mutant grade 4 astrocytoma (IDHmA4). In fact, the evidence supporting TTFields in treating IDHmA4 is largely extrapolated from EF-14, of which the results stem form heterogeneous entities of glioblastoma lack of IDH status. This study was conducted by reviewing the database in our center following the release of WHO CNS 5, aiming to preliminarily explore the efficacy of TTFields and prognosticators of newly-diagnosed IDHmA4. METHODS This retrospective analysis included 53 consecutive patients with newly-diagnosed IDHmA4, who underwent radiotherapy at Huashan Hospital, Fudan University between September 2021 and December 2023. All patients were administered with concurrent and adjuvant temozolomide. Eleven patients received adjuvant TTFields after surgery. Information regarding patient baseline characteristics, treatment approaches and molecular markers were collected with the purpose of analyzing prognostic factors. RESULTS The median follow-up was 15.7 months. At the time of data cutoff of this analysis, 20 patients had tumor relapse; 3 patients died, all of them were without TTFields therapy. The most common tumor failure pattern was local recurrence (80%, 16/20). The median PFS for the entire cohort was 19.3 months, the median OS was not reached. Univariate analysis indicated that patients who received TTFields therapy tended to have longer median PFS than that of patients without TTFields therapy (24.4 months vs. 18.5 months, P=0.097). Additionally, patients younger than 40 years (24.4 vs. 18.5 months, P=0.107) or without homozygous deletion of CDKN2A/B (23.7 vs. 18.0 months, P=0.108) tended to have better median PFS. No significant independent prognostic factor was found in the multivariate analysis. CONCLUSIONS The overall prognosis of IDHmA4 is poor. TTFields may help to improve the outcome of patients with newly-diagnosed IDHmA4. Longer follow-up, larger sample size and prospective clinical trials are warranted for deeper comprehension.

Stereotactic radiotherapy (SRT) and hypo-fractionated radiotherapy are feasible treatment options for single glioblastoma multiforme when combined with conventional radiotherapy or delivered alone. HyperArc (HA), a novel linac-based method with 4 noncoplanar arcs, has been introduced into stereotactic radiosurgery (SRS) for single and multiple metastases. In this study, we compared the dosimetric quality of HyperArc with the well-established CyberKnife (CK) and conventional VMAT methods of SRT for a single, large target. Sixteen patients treated in our center with their clinical CK plans were enrolled, and the linac-based plans were designed in silico. From the aspect of normal tissue protection and treatment efficacy, we compared the conformity index (CI), gradient index (GI), homogeneity index (HI), dose distribution in planning target volume, dose in the normal brain tissue, and mean dose of several organs at risk (OARs). All of the data were evaluated with nonparametric Kruskal‒Wallis tests. We further investigated the relationship of the dose distribution with the tumor volume and its location. The results showed that with a higher CI (0.94 ± 0.03) and lower GI (2.57 ± 0.53), the HA plans generated a lower dose to the OARs and the normal tissue. Meanwhile, the CK plans achieved a higher HI (0.35 ± 0.10) and generated a higher dose inside the tumor. Although manual VMAT showed slight improvement in dose quality and less monitoring units (2083 ± 225), HA can save half of the delivery time of CK (37 minutes) on average. HA plans have higher conformity and spare OARs with lower normal tissue irradiation, while CK plans achieve a higher mean dose in tumors. HA with 4 arcs is sufficient in dosimetric quality for a single tumor with great convenience in planning and treatment processes compared with conventional VMAT. The tumor size and location are factors to be considered when selecting treatment equipment.

IntroductionPrecise delineation of glioblastoma in multi-parameter magnetic resonance images is pivotal for neurosurgery and subsequent treatment monitoring. Transformer models have shown promise in brain tumor segmentation, but their efficacy heavily depends on a substantial amount of annotated data. To address the scarcity of annotated data and improve model robustness, self-supervised learning methods using masked autoencoders have been devised. Nevertheless, these methods have not incorporated the anatomical priors of brain structures.MethodsThis study proposed an anatomical prior-informed masking strategy to enhance the pre-training of masked autoencoders, which combines data-driven reconstruction with anatomical knowledge. We investigate the likelihood of tumor presence in various brain structures, and this information is then utilized to guide the masking procedure.ResultsCompared with random masking, our method enables the pre-training to concentrate on regions that are more pertinent to downstream segmentation. Experiments conducted on the BraTS21 dataset demonstrate that our proposed method surpasses the performance of state-of-the-art self-supervised learning techniques. It enhances brain tumor segmentation in terms of both accuracy and data efficiency.DiscussionTailored mechanisms designed to extract valuable information from extensive data could enhance computational efficiency and performance, resulting in increased precision. It's still promising to integrate anatomical priors and vision approaches.

Radiotherapy plays an important role in the treatment of glioma, and predicting radiotherapy sensitivity can help physicians develop more individualized treatment plans. However, few studies have used deep learning for glioma radiotherapy sensitivity. To better explore the impact of the relationship between tumor and neighboring regions on radiotherapy, we applied Graph Convolutional Networks (GCN) to explore predicting radiotherapy sensitivity. Firstly, we use the tumor core and the adjacent region as the nodes, where we use the radiotherapy Planning Target Volume (PTV) as the adjacent region. Secondly, we use the relationship between the tumor core and the PTV as the connected edge relationship. Finally, the Radiomics Features of each region are extracted as the node features. In this way, we construct the graph and use GCN to learn the representation of nodes in the graph to capture the structural information and association relationships among nodes for the prediction of radiotherapy sensitivity. In addition, we experimented with different node construction approaches and modular construction models. We used slice-level data to construct the graph and used a hard voting method to predict the labeling of patients. The experimental results show that our proposed node construction approach and GCN model achieve 93% accuracy after voting, which is an 11% improvement in accuracy compared to the traditional classifier.

PURPOSE:Tumor treating fields (TTFields) with concurrent radiation therapy (RT) might improve the outcome of patients with newly diagnosed glioblastoma. Several trials, including that conducted in our center, have allowed patients to wear TTFields during RT. We aimed to evaluate the setup uncertainty introduced by TTFields and calculate the planning target volume (PTV) margin for clinical reference. METHODS AND MATERIALS:We collected and analyzed 201 cone beam computed tomography images of 22 patients in our center. Patients with or without TTFields were divided into the control and TTFields groups. We evaluated the setup errors in 6 degrees of freedom and 3 degrees of freedom and the magnitudes in the 3-dimensional vectors. An estimated PTV margin for patients requiring nonimaging-guided RT was recommended. RESULTS:A significant difference was observed in the longitudinal axis between the TTFields and control groups (P < .05). These results were consistent with that of the intragroup comparison of the TTFields group. The position error of the longitudinal axis (from head to feet) was -0.51 ± 2.05 mm in the TTFields group. CONCLUSIONS:Wearing TTFields during RT increased the uncertainty, especially in the longitudinal axis, with a system error of 1.40 mm and a random error of 1.28 mm. Daily image guided RT for TTFields patients seems necessary. However, the recommended expansion margin of the PTV is 5 mm for patients requiring nonimage-guided RT to enhance the safety and efficacy of treatment.

Automatic segmentation of brain tumors is still a challenging task. To improve the segmentation performance and better ensemble all the candidate models with different architectures, we proposed a three-stage model with the quality-aware model ensemble. The first stage locates the tumor with coarse segmentation, while the second stage refines the coarse segmentation in the region of interest. The last stage performs the quality-aware model ensemble with a quality score prediction net to fuse the results from the multiple outputs of sub-networks. Besides, we warp a standard SRI24 brain template to the subject image, which is a strong prior of the brain structure and symmetry. Our method shows competitive performance on the BraTS 2021 online validation dataset, obtaining an average dice similarity coefficient (DSC) of 0.911, 0.850, 0.816, and average $$95_{th}$$ percentile of Hausdorff distance (HD95) of 4.58, 8.959, 10.400, for whole tumor, tumor core, and enhancing tumor, respectively.

Background The most frequently diagnosed primary brain tumor is glioblastoma (GBM). Nearly all patients experience tumor recurrence and up to 90% of which is local recurrence. Thus, increasing the therapeutic ratio of radiotherapy using hypofractionated stereotactic radiotherapy (HSRT) can reduce treatment time and may increase tumor control and improve survival. To evaluate the efficacy and toxicity of the combination of HSRT and intensity-modulated radiotherapy (IMRT) with temozolomide after surgery in GBM patients and provide evidence for further randomized controlled trials. Methods/design HSCK-010 is an open-label, single-arm phase II trial (NCT04547621) which includes newly diagnosed GBM patients who underwent gross total resection. Patients will receive the combination of 30 Gy/5fx HSRT, and 20 Gy/10fx IMRT adjuvant therapy with concurrent temozolomide and adjuvant chemotherapy. The primary endpoint is overall survival (OS). Secondary outcomes include progression-free survival (PFS) rate, objective-response rate (ORR), quality of life (Qol) before and after the treatment, cognitive function before and after the treatment, and rate of treatment-related adverse events (AE). The combination of HSRT and IMRT with temozolomide can benefit the patients after surgery with good survival, acceptable toxicity, and reduced treatment time. Trial registration NCT04547621 . Registered on 14 September 2020.

目的 探索低温联合奥拉西坦对重型颅脑损伤患者髓鞘碱性蛋白(MBP)及胶质纤维酸性蛋白(GFAP)表达的影响.方法 选取2017年5月至2019年6月我院收治的106例重型颅脑损伤患者,根据入院先后编号,随机将患者分成观察组和对照组,各53例.对照组患者静脉滴注奥拉西坦,观察组患者在对照组的基础上给予低温疗法.对比两组患者颅内压、临床疗效、血清MBP和GFAP及不良反应等指标.结果 治疗后,两组患者颅内压值均明显降低,且观察组降低程度更高(P<0.05);两组患者昏迷程度评分明显提高,神经功能评分明显降低,且观察组患者变化程度更高(P<0.05);两组患者血清MBP和GFAP水平明显降低,且观察组患者降低程度更高(P<0.05).治疗后6个月,观察组患者临床总有效率为71.70％,明显高于对照组的52.83％(χ2=4.015,P=0.045).治疗期间,两组患者均出现皮疹、腹泻、头晕及发热等不良反应事件,但两组差异无统计学意义(P>0.05).结论 低温联合奥拉西坦可有效改善患者昏迷程度和神经功能,提高临床疗效,下调血清MBP和GFAP,安全有效,值得临床进一步研究并推广.

针对目前显著性目标检测算法中存在的特征融合不充分、模型较为冗余等问题,提出了一种基于全局引导渐进特征融合的轻量级显著性目标检测算法.首先,使用轻量特征提取网络MobileNetV3对图像提取不同层次的多尺度特征;然后对MobileNetV3提取的高层语义特征使用轻量级多尺度感受野增强模块以进一步增强其全局特征的表征力;最后设计渐进特征融合模块对多层多尺度特征自顶而下逐步融合,并采用常用的交叉熵损失函数在多个阶段对这些融合特征进行优化,得到由粗到细的显著图.整个网络模型是无需预处理和后处理的端到端结构.在6个基准数据集上进行了大量实验,并采用PR_Curve,F-measure,S-measure和MAE指标来衡量性能.结果表明,所提方法明显优于10种先进的对比方法,并且算法模型大小仅约为10 MB,在GTX2080Ti显卡上处理大小为400×300像素的图像的速度可以达到46帧/秒.

Background: The automated segmentation of brain gliomas regions in magnetic resonance (MR) images plays an important role in the early diagnosis, intraoperative navigation, radiotherapy planning and prognosis of brain tumors. It is very challenging to segment gliomas and intratumoral structures since the location, size, shape, edema range and boundary of gliomas are heterogeneous, and multimodal brain gliomas images (such as T1, T2, fluid-attenuated inversion recovery (FLAIR), and T1c images) are collected from multiple radiation centers. Methods: This paper presents a multimodal, multi-scale, double-pathway, 3D residual convolution neural network (CNN) for automatic gliomas segmentation. First, a robust gray-level normalization method is proposed to solve the multicenter problem, such as very different intensity ranges due to different imaging protocols. Second, a multi-scale, double-pathway network based on DeepMedic toolkit is trained with different combinations of multimodal MR images for gliomas segmentation. Finally, a fully connected conditional random field (CRF) is used as a post-processing strategy to optimize the segmentation results for addressing the isolated segmentations and holes. Results: Experiments on the Multimodal Brain Tumor Segmentation (BraTS) 2017 and 2019 challenge data show that our methods achieve a good performance in delineating the whole tumor with a Dice coefficient, a sensitivity and a positive predictive value (PPV) of 0.88, 0.89 and 0.88, respectively. Regarding the segmentation of the tumor core and the enhancing area, the sensitivity reached 0.80. Conclusions: Experiments show that our method can accurately segment gliomas and intratumoral structures from multimodal MR images, and it is of great significance to clinical neurosurgery.

Recently, benefiting from the fast development of deep convolutional neural networks, salient object detection (SOD) has achieved gratifying performance in a variety of challenging scenarios. Among them, how to learn more discriminative features plays a key role. In this paper, we propose a novel network architecture that progressively fuses the rich multi-level contextual features from top to bottom to learn a more effective feature presentation for robust SOD. Concretely, we first design a multi-receptive field block (MRFB) to capture multi-scale contextual information. Then, we develop a feature fusion block that progressively fuses different outputs of MRFBs from top to bottom, which can effectively filter out the non-complementary parts of the high-level and low-level features. Afterwards, we leverage a refinement residual block to refine the results further. Finally, we leverage an edge-aware loss as an aid to guide the network to learn more sharpen details of the salient objects. The whole network is trained end-to-end without any pre-processing and post-processing. Exhaustive evaluations on six benchmark datasets demonstrate superiority of the proposed method against state-of-the-arts in terms of all metrics.

目的:分析高血压基底节区脑出血患者经术中实时超声引导行小骨窗经侧裂显微手术治疗的效果.方法:回顾性分析本院神经外科2015年12月至2018年12月收治的100例高血压基底节区脑出血患者临床资料,根据手术方式将57例行超声引导下小骨窗经侧裂显微手术治疗患者作为研究组,43例行单纯小骨窗经侧裂显微手术治疗患者作为对照组,观察两组患者手术相关指标变化,术后1d血肿清除率,术后1月GOS评分,术后6个月ADL分级及术后并发症发生情况.结果:研究组患者手术时间、手术出血量、住院时间均明显优于对照组(P<0.05);研究组患者血肿清除率>85％高于对照组、清除率50％～85％低于对照组(P<0.05);与对照组比较,研究组GOS评分4分患者比例显著增加(P<0.05),3分患者比例显著降低(P<0.05);研究组Ⅱ级患者比例明显高于对照组(P<0.05);研究组患者术后并发症总发生率为5.26％,与对照组的13.95％比较差异无统计学意义(P>0.05).结论:超声引导下小骨窗经侧裂显微手术治疗高血压基底节区脑出血可缩短手术时间、减少手术出血量、缩短住院时间,且血肿清除率高、并发症较少,有利于术后恢复.