A Survey on Efficient Vision Transformers: Algorithms, Techniques, and Performance Benchmarking.

Vision Transformer (ViT) architectures are becoming increasingly popular and widely employed to tackle computer vision applications. Their main feature is the capacity to extract global information through the self-attention mechanism, outperforming earlier convolutional neural networks. However, ViT deployment and performance have grown steadily with their size, number of trainable parameters, and operations. Furthermore, self-attention's computational and memory cost quadratically increases with the image resolution. Generally speaking, it is challenging to employ these architectures in real-world applications due to many hardware and environmental restrictions, such as processing and computational capabilities. Therefore, this survey investigates the most efficient methodologies to ensure sub-optimal estimation performances. More in detail, four efficient categories will be analyzed: compact architecture, pruning, knowledge distillation, and quantization strategies. Moreover, a new metric called Efficient Error Rate has been introduced in order to normalize and compare models' features that affect hardware devices at inference time, such as the number of parameters, bits, FLOPs, and model size. Summarizing, this paper firstly mathematically defines the strategies used to make Vision Transformer efficient, describes and discusses state-of-the-art methodologies, and analyzes their performances over different application scenarios. Toward the end of this paper, we also discuss open challenges and promising research directions.

Constructing hierarchical attentive functional brain networks for early AD diagnosis.

Analyzing functional brain networks (FBN) with deep learning has demonstrated great potential for brain disorder diagnosis. The conventional construction of FBN is typically conducted at a single scale with a predefined brain region atlas. However, numerous studies have identified that the structure and function of the brain are hierarchically organized in nature. This urges the need of representing FBN in a hierarchical manner for more effective analysis of the complementary diagnostic insights at different scales. To this end, this paper proposes to build hierarchical FBNs adaptively within the Transformer framework. Specifically, a sparse attention-based node-merging module is designed to work alongside the conventional network feature extraction modules in each layer. The proposed module generates coarser nodes for further FBN construction and analysis by combining fine-grained nodes. By stacking multiple such layers, a hierarchical representation of FBN can be adaptively learned in an end-to-end manner. The hierarchical structure can not only integrate the complementary information from multiscale FBN for joint analysis, but also reduce the model complexity due to decreasing node sizes. Moreover, this paper argues that the nodes defined by the existing atlases are not necessarily the optimal starting level to build FBN hierarchy and exploring finer nodes may further enrich the FBN representation. In this regard, each predefined node in an atlas is split into multiple sub-nodes, overcoming the scale limitation of the existing atlases. Extensive experiments conducted on various data sets consistently demonstrate the superior performance of the proposed method over the competing methods.

Efficacy and safety analysis of hypofractionated and conventional fractionated radiotherapy in postoperative breast cancer patients.

OBJECTIVES: In this meta-analysis, we conducted a comparative analysis of the safety and efficacy of hypofractionated and conventional fractionated radiotherapy in individuals who had undergone surgery for breast cancer. METHODS: This study involved a systematic and independent review of relevant research articles published in reputable databases such as PubMed, Embase, Cochrane Library, and Web of Science. Two investigators conducted the review, which included studies published up to January 3, 2023. The quality of the eligible studies was evaluated and data were extracted using Review Manager software 5.4 (RevMan 5.4) to calculate odds ratios (ORs) and 95% confidence intervals (CIs). RESULTS: The analysis comprised 35 studies and encompassed a collective sample of 18,246 individuals diagnosed with breast cancer. We did not find a statistically significant disparity in efficacy between conventional fractionated (CF) radiotherapy and hypofractionated (HF) radiotherapy regarding local recurrence (LR; OR = 0.91, 95% CI: 0.76-1.09, P = 0.30), disease-free survival (DFS; OR = 1.20, 95% CI: 1.01-1.42, P = 0.03), and overall survival (OS; OR = 1.08, 95% CI: 0.93-1.26, P = 0.28). Concerning safety, there was no significant difference between the HF and CF regimens in terms of breast pain, breast atrophy, lymphedema, pneumonia, pulmonary fibrosis, telangiectasia, and cardiotoxicity. However, the HF regimen resulted in lower skin toxicity (OR = 0.43, 95% CI: 0.33-0.55, P < 0.01) and improved patient fatigue outcomes (OR = 0.73, 95% CI: 0.60 - 0.88, P < 0.01). CONCLUSIONS: Although there is no substantial difference in LR, DFS, OS, or many other side effects between the HF and CF regimens, the HF regimen reduces skin toxicity and relieves patient fatigue. If these two issues need to be addressed in clinical situations, the HF regimen may be a superior alternative to conventional radiotherapy in postoperative breast cancer patients.

Akkermansia muciniphila MucT attenuates sodium valproate-induced hepatotoxicity and upregulation of Akkermansia muciniphila in rats.

In the previous study, we found that the oral sodium valproate (SVP) increased the relative abundance of Akkermansia muciniphila (A. muciniphila) in rats, and plasma aspartate transaminase (AST) and alanine aminotransferase (ALT) activities were positively correlated with A. muciniphila levels. This study aimed to further investigate the role of A. muciniphila in SVP-induced hepatotoxicity by orally supplementing rats with the representative strain of A. muciniphila, A. muciniphila MucT. Additionally, the fresh faeces were incubated anaerobically with SVP to investigate the effect of SVP on faecal A. muciniphila in the absence of host influence. Results showed that A. muciniphila MucT ameliorated the hepatotoxicity and upregulation of A. muciniphila induced by SVP. SVP also induced a noteworthy elevation of A. muciniphila level in vitro, supporting the observation in vivo. Therefore, we speculate that A. muciniphila MucT may be a potential therapeutic strategy for SVP-induced hepatotoxicity. In addition, the increased A. muciniphila induced by SVP may differ from A. muciniphila MucT, but further evidence is needed. These findings provide new insights into the relationships between A. muciniphila and SVP-induced hepatotoxicity, highlighting the potential for different A. muciniphila strains to have distinct or even opposing effects on SVP-induced hepatotoxicity.

Comparative effectiveness of nonpharmacological interventions in reducing psychological symptoms among patients with chronic low back pain.

OBJECTIVES: Chronic low back pain (CLBP) can seriously impair the quality of life of patients and has a remarkable comorbidity with psychological symptoms, which, in turn, can further exacerbate the symptoms of CLBP. Psychological treatments are critical and nonnegligent for the management of CLBP, and thus, should attract sufficient attention. However, current evidence does not suggest the superiority and effectiveness of nonpharmacological interventions in reducing psychological symptoms among patients with CLBP.Thus, this study was designed to compare the effectiveness of nonpharmacological interventions for depression, anxiety, and mental health among patients with CLBP and to recommend preferred strategies for attenuating psychological symptoms in clinical practice. METHODS: In this systematic review and network meta-analysis (NMA), PubMed, Embase Database, Web of Science, and Cochrane Library were searched from database inception until March 2022. Randomized clinical trials (RCTs) that compare different nonpharmacological interventions for depression, anxiety, and mental health among patients with CLBP were eligible. The Preferred Reporting Items for Systematic Reviews and Meta-analyses statement was used. Four reviewers in pairs and divided into two groups independently performed literature selection, data extraction, and risk of bias, and certainty of evidence assessments. This NMA was conducted with a random effects model under a frequentist framework. The major outcomes were depression, anxiety, and mental health presented as the standardized mean difference (SMD) with the corresponding 95% CI. RESULTS: A total of 66 RCTs that randomized 4806 patients with CLBP met the inclusion criteria. The quality of evidence was typically low or some risks of bias (47 out of 66 trials, 71.3%), and the precision of summary estimates for effectiveness varied substantially. In addition, 7 categories of interventions with 26 specific treatments were evaluated. For depression, mind body therapy (pooled SMD = -1.20, 95% CI: -1.63 to -0.78), biopsychosocial approach (pooled SMD = -0.41, 95% CI: -0.70 to -0.12), and physical therapy (pooled SMD = -0.26, 95% CI: -0.50 to -0.02) exhibited remarkable effectiveness in reducing depression compared with the control group. For managing anxiety, mind body therapy (pooled SMD = -1.35, 95% CI: -1.90 to -0.80), multicomponent intervention (pooled SMD = -0.47, 95% CI: -0.88 to -0.06), and a biopsychosocial approach (pooled SMD = -0.46, 95% CI: -0.79 to -0.14) were substantially superior to the control group. For improving mental health, multicomponent intervention (pooled SMD = 0.77, 95% CI: 0.14 to 1.39), exercise (pooled SMD = 0.60, 95% CI: 0.08 to 1.11), and physical therapy (pooled SMD = 0.47, 95% CI: 0.02-0.92) demonstrated statistically substantial effectiveness compared with the control group. The rank probability indicated that mind body therapy achieved the highest effectiveness in reducing depression and anxiety among patients with CLBP. Besides, the combined results should be interpreted cautiously based on the results of analyses evaluating the inconsistency and certainty of the evidence. CONCLUSION: This systemic review and NMA suggested that nonpharmacological interventions show promise for reducing psychological symptoms among patients with CLBP. In particular, mind body therapy and a biopsychosocial approach show considerable promise, and mind body therapy can be considered a priority choice in reducing depression and anxiety. These findings can aid clinicians in assessing the potential risks and benefits of available treatments for CLBP comorbidity with psychological symptoms and provide evidence for selecting interventions in clinical practice. More RCTs involving different interventions with rigorous methodology and an adequate sample size should be conducted in future research.

3D multi-modality Transformer-GAN for high-quality PET reconstruction.

Positron emission tomography (PET) scans can reveal abnormal metabolic activities of cells and provide favorable information for clinical patient diagnosis. Generally, standard-dose PET (SPET) images contain more diagnostic information than low-dose PET (LPET) images but higher-dose scans can also bring higher potential radiation risks. To reduce the radiation risk while acquiring high-quality PET images, in this paper, we propose a 3D multi-modality edge-aware Transformer-GAN for high-quality SPET reconstruction using the corresponding LPET images and T1 acquisitions from magnetic resonance imaging (T1-MRI). Specifically, to fully excavate the metabolic distributions in LPET and anatomical structural information in T1-MRI, we first use two separate CNN-based encoders to extract local spatial features from the two modalities, respectively, and design a multimodal feature integration module to effectively integrate the two kinds of features given the diverse contributions of features at different locations. Then, as CNNs can describe local spatial information well but have difficulty in modeling long-range dependencies in images, we further apply a Transformer-based encoder to extract global semantic information in the input images and use a CNN decoder to transform the encoded features into SPET images. Finally, a patch-based discriminator is applied to ensure the similarity of patch-wise data distribution between the reconstructed and real images. Considering the importance of edge information in anatomical structures for clinical disease diagnosis, besides voxel-level estimation error and adversarial loss, we also introduce an edge-aware loss to retain more edge detail information in the reconstructed SPET images. Experiments on the phantom dataset and clinical dataset validate that our proposed method can effectively reconstruct high-quality SPET images and outperform current state-of-the-art methods in terms of qualitative and quantitative metrics.

Construction of a Collagen-like Protein Based on Elastin-like Polypeptide Fusion and Evaluation of Its Performance in Promoting Wound Healing.

In the healing of wounds, human-like collagen (hCol) is essential. However, collagen-based composite dressings have poor stability in vivo, which severely limits their current therapeutic potential. Based on the above, we have developed a recombinant fusion protein named hCol-ELP, which consists of hCol and an elastin-like peptide (ELP). Then, we examined the physicochemical and biological properties of hCol-ELP. The results indicated that the stability of the hCol-ELP fusion protein exhibited a more compact and homogeneous lamellar microstructure along with collagen properties, it was found to be significantly superior to the stability of free hCol. The compound hCol-ELP demonstrated a remarkable capacity to induce the proliferation and migration of mouse embryo fibroblast cells (NIH/3T3), as well as enhance collagen synthesis in human skin fibroblasts (HSF) when tested in vitro. In vivo, hCol-ELP demonstrated significant enhancements in healing rate and a reduction in the time required for scab removal, thereby exhibiting a scar-free healing effect. The findings provide a crucial theoretical foundation for the implementation of an hCol-ELP protein dressing in fields associated with the healing of traumatic injuries.

Breast Fibroglandular Tissue Segmentation for Automated BPE Quantification With Iterative Cycle-Consistent Semi-Supervised Learning.

Background Parenchymal Enhancement (BPE) quantification in Dynamic Contrast-Enhanced Magnetic Resonance Imaging (DCE-MRI) plays a pivotal role in clinical breast cancer diagnosis and prognosis. However, the emerging deep learning-based breast fibroglandular tissue segmentation, a crucial step in automated BPE quantification, often suffers from limited training samples with accurate annotations. To address this challenge, we propose a novel iterative cycle-consistent semi-supervised framework to leverage segmentation performance by using a large amount of paired pre-/post-contrast images without annotations. Specifically, we design the reconstruction network, cascaded with the segmentation network, to learn a mapping from the pre-contrast images and segmentation predictions to the post-contrast images. Thus, we can implicitly use the reconstruction task to explore the inter-relationship between these two-phase images, which in return guides the segmentation task. Moreover, the reconstructed post-contrast images across multiple auto-context modeling-based iterations can be viewed as new augmentations, facilitating cycle-consistent constraints across each segmentation output. Extensive experiments on two datasets with various data distributions show great segmentation and BPE quantification accuracy compared with other state-of-the-art semi-supervised methods. Importantly, our method achieves 11.80 times of quantification accuracy improvement along with 10 times faster, compared with clinical physicians, demonstrating its potential for automated BPE quantification. The code is available at https://github.com/ZhangJD-ong/Iterative-Cycle-consistent-Semi-supervised-Learning-for-fibroglandular-tissue-segmentation.

A robust and efficient AI assistant for breast tumor segmentation from DCE-MRI via a spatial-temporal framework.

Dynamic contrast-enhanced magnetic resonance imaging (DCE-MRI) allows screening, follow up, and diagnosis for breast tumor with high sensitivity. Accurate tumor segmentation from DCE-MRI can provide crucial information of tumor location and shape, which significantly influences the downstream clinical decisions. In this paper, we aim to develop an artificial intelligence (AI) assistant to automatically segment breast tumors by capturing dynamic changes in multi-phase DCE-MRI with a spatial-temporal framework. The main advantages of our AI assistant include (1) robustness, i.e., our model can handle MR data with different phase numbers and imaging intervals, as demonstrated on a large-scale dataset from seven medical centers, and (2) efficiency, i.e., our AI assistant significantly reduces the time required for manual annotation by a factor of 20, while maintaining accuracy comparable to that of physicians. More importantly, as the fundamental step to build an AI-assisted breast cancer diagnosis system, our AI assistant will promote the application of AI in more clinical diagnostic practices regarding breast cancer.

Characteristics and evaluation of C1 posterior arch variation for transpedicular screw placement between patients with and without basilar invagination.

BACKGROUND: C1 transpedicular screw (C1TS) placement provided satisfactory pullout resistance and 3D stability, but its application might be limited in patients with basilar invagination (BI) due to the high incidences of the atlas anomaly and vertebral artery (VA) variation. However, no study has explored the classifications of C1 posterior arch variations and investigated their indications and ideal insertion trajectories for C1TS in BI. PURPOSE: To investigate the bony and surrounding arterial characteristics of the atlas, classify posterior arch variations, identify indications for C1TS, evaluate ideal insertion trajectories for C1TS in BI patients without atlas occipitalization (AO), and compare them with those without BI and AO as control. METHODS: A total of 130 non-AO patients with and without BI (52 patients and 78 patients, respectively) from two medical centers were included at a 1:1.5 ratio. The posterior arch variations were assessed using a modified C1 morphological classification. Comparisons regarding the bony and surrounding arterial characteristics, morphological classification distributions, and ideal insertion trajectories between BI and control groups were performed. The subgroup analyses based on different morphological classifications were also conducted. In addition, the factors possibly affecting the insertion parameters were investigated using multiple linear regression analyses. RESULTS: The BI group was associated with significantly smaller lateral mass height and width, sagittal length of posterior arch, pedicle height, vertical height of posterior arch, and distance between VA and VA groove (VAG) than control group. Four types of posterior arch variations with indications for different screw placement techniques were classified; Classifications I and II were suitable for C1TS. The BI cohort showed a significantly lower rate of Classification I than the control cohort. In the BI group, the subgroup of Classification I had significantly larger distance between the insertion point (IP) and inferior aspect of the posterior arch. In addition, it had the narrowest width along ideal screw trajectory, but a significantly more lateral ideal mediolateral angle than the subgroup of Classification II. Multiple linear regression indicated that the cephalad angle was significantly associated with the diagnosis of BI (B = 3.708, P < 0.001) and sagittal diameter of C1 (B = 3.417, P = 0.027); the ideal mediolateral angle was significantly associated with BMI (B = 0.264, P = 0.031), sagittal diameter of C1 (B = - 4.559, P = 0.002), and pedicle height (B = - 2.317, P < 0.001); the distance between the IP and inferior aspects of posterior arch was significantly associated with age (B = - 0.002, P = 0.035), BMI (B = - 0.007, P = 0.028), sagittal length of posterior arch (B = - 0.187, P = 0.032), pedicle height (B = - 0.392, P < 0.001), and middle and lower parts of posterior arch (B = 0.862, P < 0.001). CONCLUSION: The incidence of posterior arch variation in BI patients without AO was remarkably higher than that in control patients. The insertion parameters of posterior screws were different between the morphological classification types in BI and control groups. The distance between VA V3 segments and VAG in BI cohort was substantially smaller than that in control cohort. Preoperative individual 3D computed tomography (CT), CT angiography and intraoperative navigation are recommended for BI patients receiving posterior screw placement.

Cardiorespiratory dose comparison among six radiotherapy regimens for patients with left-sided breast cancer.

There is uncertainty regarding the benefits and drawbacks of various radiation protocols for the treatment of left-sided breast cancer. To address this issue, we conducted a Bayesian network analysis. First, we searched several electronic databases for eligible literature. Next, we pooled the data from twelve studies concerning three-dimensional conformal radiation therapy (3D-CRT), intensity modulated radiation therapy (IMRT), and volumetric modulated arc therapy (VMAT), combined with either deep inspiratory breath-holding (DIBH) or free-breathing (FB) modalities. The integrated cardiac and pulmonary dosimetric indexes for all included treatments were compared using Bayesian networks. A direct meta-analysis indicated that for the two methods of 3D-CRT and IMRT, DIBH technology was more effective than FB in reducing the radiation dose to the heart and lungs. Additionally, according to the network results, DIBH was superior to FB in all six treatment options, regardless of whether the plan was 3D-CRT, IMRT, or VMAT. Besides, the combined data indicated that the FB-3D-CRT regimen had the weakest dosimetric advantage of all the treatments. Excluding FB-3D-CRT, each of the other five treatments had its own specific benefits. This is the first Bayesian study of several radiotherapy regimens for breast cancer patients on the left side, and the findings can be used to select appropriate radiotherapy programs for breast cancer patients.

Clinical application of spinal robot in cervical spine surgery: safety and accuracy of posterior pedicle screw placement in comparison with conventional freehand methods.

The novel robot-assisted (RA) technique has been utilized increasingly to improve the accuracy of cervical pedicle screw placement. Although the clinical application of the RA technique has been investigated in several case series and comparative studies, the superiority and safety of RA over conventional freehand (FH) methods remain controversial. Meanwhile, the intra-pedicular accuracy of the two methods has not been compared for patients with cervical traumatic conditions. This study aimed to compare the rate and risk factors of intra-pedicular accuracy of RA versus the conventional FH approach for posterior pedicle screw placement in cervical traumatic diseases. A total of 52 patients with cervical traumatic diseases who received cervical screw placement using RA (26 patients) and FH (26 patients) techniques were retrospectively included. The primary outcome was the intra-pedicular accuracy of cervical pedicle screw placement according to the Gertzbin-Robbins scale. Secondary outcome parameters included surgical time, intraoperative blood loss, postoperative drainage, postoperative hospital stay, and complications. Moreover, the risk factors that possibly affected intra-pedicular accuracy were assessed using univariate analyses. Out of 52 screws inserted using the RA method, 43 screws (82.7%) were classified as grade A, with the remaining 7 (13.5%) and 2 (3.8%) screws classified as grades B and C. In the FH cohort, 60.8% of the 79 screws were graded A, with the remaining screws graded B (21, 26.6%), C (8, 10.1%), and D (2, 2.5%). The RA technique showed a significantly higher rate of optimal intra-pedicular accuracy than the FH method (P = 0.008), but there was no significant difference between the two groups in terms of clinically acceptable accuracy (P = 0.161). Besides, the RA technique showed remarkably longer surgery time, less postoperative drainage, shorter postoperative hospital stay, and equivalent intraoperative blood loss and complications than the FH technique. Furthermore, the univariate analyses showed that severe obliquity of the lateral atlantoaxial joint in the coronal plane (P = 0.003) and shorter width of the lateral mass at the inferior margin of the posterior arch (P = 0.014) were risk factors related to the inaccuracy of C1 screw placement. The diagnosis of HRVA (P < 0.001), severe obliquity of the lateral atlantoaxial joint in the coronal plane (P < 0.001), short pedicle width (P < 0.001), and short pedicle height (P < 0.001) were risk factors related to the inaccuracy of C2 screw placement. RA cervical pedicle screw placement was associated with a higher rate of optimal intra-pedicular accuracy to the FH technique for patients with cervical traumatic conditions. The severe obliquity of the lateral atlantoaxial joint in the coronal plane independently contributed to high rates of the inaccuracy of C1 and C2 screw placements. RA pedicle screw placement is safe and useful for cervical traumatic surgery.

The SHDRA syndrome-associated gene TMEM260 encodes a protein-specific O-mannosyltransferase.

Mutations in the TMEM260 gene cause structural heart defects and renal anomalies syndrome, but the function of the encoded protein remains unknown. We previously reported wide occurrence of O-mannose glycans on extracellular immunoglobulin, plexin, transcription factor (IPT) domains found in the hepatocyte growth factor receptor (cMET), macrophage-stimulating protein receptor (RON), and plexin receptors, and further demonstrated that two known protein O-mannosylation systems orchestrated by the POMT1/2 and transmembrane and tetratricopeptide repeat-containing proteins 1-4 gene families were not required for glycosylation of these IPT domains. Here, we report that the TMEM260 gene encodes an ER-located protein O-mannosyltransferase that selectively glycosylates IPT domains. We demonstrate that disease-causing TMEM260 mutations impair O-mannosylation of IPT domains and that TMEM260 knockout in cells results in receptor maturation defects and abnormal growth of 3D cell models. Thus, our study identifies the third protein-specific O-mannosylation pathway in mammals and demonstrates that O-mannosylation of IPT domains serves critical functions during epithelial morphogenesis. Our findings add a new glycosylation pathway and gene to a growing group of congenital disorders of glycosylation.

Single-View 3D Mesh Reconstruction for Seen and Unseen Categories.

Single-view 3D object reconstruction is a fundamental and challenging computer vision task that aims at recovering 3D shapes from single-view RGB images. Most existing deep learning based reconstruction methods are trained and evaluated on the same categories, and they cannot work well when handling objects from novel categories that are not seen during training. Focusing on this issue, this paper tackles Single-view 3D Mesh Reconstruction, to study the model generalization on unseen categories and encourage models to reconstruct objects literally. Specifically, we propose an end-to-end two-stage network, GenMesh, to break the category boundaries in reconstruction. Firstly, we factorize the complicated image-to-mesh mapping into two simpler mappings, i.e., image-to-point mapping and point-to-mesh mapping, while the latter is mainly a geometric problem and less dependent on object categories. Secondly, we devise a local feature sampling strategy in 2D and 3D feature spaces to capture the local geometry shared across objects to enhance model generalization. Thirdly, apart from the traditional point-to-point supervision, we introduce a multi-view silhouette loss to supervise the surface generation process, which provides additional regularization and further relieves the overfitting problem. The experimental results show that our method significantly outperforms the existing works on the ShapeNet and Pix3D under different scenarios and various metrics, especially for novel objects.

Bridging Synthetic and Real Images: A Transferable and Multiple Consistency Aided Fundus Image Enhancement Framework.

Deep learning based image enhancement models have largely improved the readability of fundus images in order to decrease the uncertainty of clinical observations and the risk of misdiagnosis. However, due to the difficulty of acquiring paired real fundus images at different qualities, most existing methods have to adopt synthetic image pairs as training data. The domain shift between the synthetic and the real images inevitably hinders the generalization of such models on clinical data. In this work, we propose an end-to-end optimized teacher-student framework to simultaneously conduct image enhancement and domain adaptation. The student network uses synthetic pairs for supervised enhancement, and regularizes the enhancement model to reduce domain-shift by enforcing teacher-student prediction consistency on the real fundus images without relying on enhanced ground-truth. Moreover, we also propose a novel multi-stage multi-attention guided enhancement network (MAGE-Net) as the backbones of our teacher and student network. Our MAGE-Net utilizes multi-stage enhancement module and retinal structure preservation module to progressively integrate the multi-scale features and simultaneously preserve the retinal structures for better fundus image quality enhancement. Comprehensive experiments on both real and synthetic datasets demonstrate that our framework outperforms the baseline approaches. Moreover, our method also benefits the downstream clinical tasks.

Robotics in Cervical Spine Surgery: Feasibility and Safety of Posterior Screw Placement.

OBJECTIVE: Robot-assisted (RA) techniques have been widely investigated in thoracolumbar spine surgery. However, the application of RA methods on cervical spine surgery is rare due to the complex morphology of cervical vertebrae and catastrophic complications. Thus, the feasibility and safety of RA cervical screw placement remain controversial. This study aims to evaluate the feasibility and safety of RA screw placement on cervical spine surgery. METHODS: A comprehensive search on PubMed, Cochrane Library, Embase Database, Web of Science, Chinese National Knowledge Databases, and Wanfang Database was performed to select potential eligible studies. Randomized controlled trials (RCTs), comparative cohort studies, and case series reporting the accuracy of cervical screw placement were included. The Cochrane risk of bias criteria and Newcastle-Ottawa Scale criteria were utilized to rate the risk of bias of the included literatures. The primary outcome was the rate of cervical screw placement accuracy with robotic guidance; subgroup analyses based on the screw type and insertion segments were also performed. RESULTS: One RCT, 3 comparative cohort studies, and 3 case series consisting of 160 patients and 719 cervical screws were included in this meta-analysis. The combined outcomes indicated that the rates of optimal and clinically acceptable cervical screw placement accuracy under robotic guidance were 88.0% (95% confidence interval [CI], 84.1%-91.4%; p = 0.073; I2 = 47.941%) and 98.4% (95% CI, 96.8%-99.5%; p = 0.167; I2 = 35.954%). The subgroup analyses showed that the rate of optimal pedicle screw placement accuracy was 88.2% (95% CI, 83.1%-92.6%; p = 0.057; I2 = 53.305%); the rates of optimal screw placement accuracy on C1, C2, and subaxial segments were 96.2% (95% CI, 80.5%-100.0%; p = 0.167; I2 = 44.134X%), 89.7% (95% CI, 80.6%-96.6%; p = 0.370; I2 = 0.000X%), and 82.6% (95% CI, 70.9%-91.9%; p = 0.057; I2 = 65.127X%;), respectively. CONCLUSION: RA techniques were associated with high rates of optimal and clinically acceptable screw positions. RA cervical screw placement is accurate, safe, and feasible in cervical spine surgery with promising clinical potential.