BACKGROUND:Predicting tumor regression grade (TRG) after neoadjuvant chemoradiotherapy (NCRT) in patients with locally advanced rectal cancer (LARC) preoperatively accurately is crucial for providing individualized treatment plans. This study aims to develop transrectal contrast-enhanced ultrasound-based (TR-CEUS) radiomics models for predicting TRG. METHODS:A total of 190 LARC patients undergoing NCRT and subsequent total mesorectal excision were categorized into good and poor response groups based on pathological TRG. TR-CEUS examinations were conducted before and after NCRT. Machine learning (ML) models for predicting TRG were developed by employing pre- and post-NCRT TR-CEUS image series, based on seven classifiers, including random forest (RF), multi-layer perceptron (MLP) and so on. The predictive performance of models was evaluated using receiver operating characteristic curve analysis and Delong test. RESULTS:A total of 1525 TR-CEUS images were included for analysis, and 3360 ML models were constructed using image series before and after NCRT, respectively. The optimal pre-NCRT ML model, constructed from imaging series before NCRT, was RF; whereas the optimal post-NCRT model, derived from imaging series after NCRT, was MLP. The areas under the curve for the optimal RF and MLP models demonstrated values of 0.609 and 0.857, respectively, in the cross-validation cohort, with corresponding values of 0.659 and 0.841 observed in the independent test cohort. Delong tests showed that the predictive efficacy of the post-NCRT model was statistically higher than that of the pre-NCRT model (p < 0.05). CONCLUSIONS:Radiomics model developed by TR-CEUS images after NCRT demonstrated high predictive performance for TRG, thereby facilitating precise evaluation of therapeutic response to NCRT in LARC patients.

Image dehazing is crucial for robust visual perception in railway systems. However, current datasets primarily target general scenes, lacking the vital, specific structures and elements found in railway tracks. This data gap significantly hinders the development of railway visual perception algorithms, preventing models from achieving ideal performance. To bridge this, this research introduce RailSem19-H, the first large-scale foggy image dataset specifically designed for railway scenes, comprising 30,000 training images and 1,000 pairs of hazy test images. RailSem19-H construction strictly adheres to the physical model of atmospheric scattering. This research uses original clear images from RailSem19 as the scene radiance parameter $J(x)$, generating scene depth maps $d(x)$ via the Midas depth estimation algorithm. Then, by systematically adjusting core parameters like atmospheric light $A$ and scattering coefficient $\beta$, this research create ten multi-gradient haze levels (moderate to dense fog) for each original image. DehazeFormer, trained on RailSem19-H, achieves exceptionally high results across various metrics and visual effects. RailSem19-H fills a critical void for dedicated datasets in railway dehazing and establishes a highperformance baseline model. This paves the way for algorithmic innovation and practical application breakthroughs in this field under complex meteorological conditions.

The upcoming 6G technology, with its high speed and low latency, is poised to become a foundational technology for intelligent transportation systems. To handle the massive data generated by connected vehicles in 6G environments, federated learning methods are essential. However, traditional centralized federated learning approaches still face challenges related to data and device heterogeneity, which significantly affects training efficiency. To address these challenges, we propose FedCPC, a context-based adaptive pruning clustered federated learning method. Based on the positive correlation between similar data distributions and model representations, we use centralized kernel alignment (CKA) to group clients with similar data distributions, thus reducing the impact of data heterogeneity. Furthermore, we introduce a context-aware random forest multi-armed bandit method to determine appropriate pruning rates based on device capabilities and historical performance which addresses device heterogeneity concerns. Experimental results on open-source datasets demonstrate that FedCPC outperforms traditional FL methods in both learning efficiency and communication effectiveness.

Ultrasound imaging is widely used in medical diagnostics due to its non-invasive and real-time capabilities. However, existing methods often overlook the benefits of fractional-order filters for denoising and dehazing. Thus, this work introduces an efficient multi-scale wavelet method for dehazing and denoising ultrasound images using a fractional-order filter, which integrates a guided filter, directional filter, fractional-order filter, and haze removal to the different resolution images generated by a multi-scale wavelet. In the directional filter stage, an eigen-analysis of each pixel is conducted to extract structural features, which are then classified into edges for targeted filtering. The guided filter subsequently reduces speckle noise in homogeneous anatomical regions. The fractional-order filter allows the algorithm to effectively denoise while improving edge definition, irrespective of the edge size. Haze removal can effectively eliminate the haze caused by attenuation. Our method achieved significant improvements, with PSNR reaching 31.25 and SSIM 0.905 on our ultrasound dataset, outperforming other methods. Additionally, on external datasets like McMaster and Kodak24, it achieved the highest PSNR (29.68, 28.62) and SSIM (0.858, 0.803). Clinical evaluations by four radiologists confirmed its superiority in liver and carotid artery images. Overall, our approach outperforms existing speckle reduction and structural preservation techniques, making it highly suitable for clinical ultrasound imaging.

ObjectiveThis study aimed to develop a deep learning system to identify and differentiate the metastatic cervical lymph nodes (CLNs) of thyroid cancer.MethodsFrom January 2014 to December 2020, 3059 consecutive patients with suspected with metastatic CLNs of thyroid cancer were retrospectively enrolled in this study. All CLNs were confirmed by fine needle aspiration. The patients were randomly divided into the training (1228 benign and 1284 metastatic CLNs) and test (307 benign and 240 metastatic CLNs) groups. Grayscale ultrasonic images were used to develop and test the performance of the Y-Net deep learning model. We used the Y-Net network model to segment and differentiate the lymph nodes. The Dice coefficient was used to evaluate the segmentation efficiency. Sensitivity, specificity, accuracy, positive predictive value (PPV), and negative predictive value (NPV) were used to evaluate the classification efficiency.ResultsIn the test set, the median Dice coefficient was 0.832. The sensitivity, specificity, accuracy, PPV, and NPV were 57.25%, 87.08%, 72.03%, 81.87%, and 66.67%, respectively. We also used the Y-Net classified branch to evaluate the classification efficiency of the LNs ultrasonic images. The classification branch model had sensitivity, specificity, accuracy, PPV, and NPV of 84.78%, 80.23%, 82.45%, 79.35%, and 85.61%, respectively. For the original ultrasonic reports, the sensitivity, specificity, accuracy, PPV, and NPV were 95.14%, 34.3%, 64.66%, 59.02%, 87.71%, respectively. The Y-Net model yielded better accuracy than the original ultrasonic reports.ConclusionThe Y-Net model can be useful in assisting sonographers to improve the accuracy of the classification of ultrasound images of metastatic CLNs.

Speckle noise is a granular interference that degrades image quality in coherent imaging systems, including underwater sonar, Synthetic Aperture Radar (SAR), and medical ultrasound. This study aims to enhance speckle noise reduction through advanced deep learning techniques. We introduce the Deep Gradient-Guidance Network (DGGNet), which features an architecture comprising one encoder and two decoders—one dedicated to image recovery and the other to gradient preservation. Our approach integrates a gradient map and fractional-order total variation into the loss function to guide training. The gradient map provides structural guidance for edge preservation and directs the denoising branch to focus on sharp regions, thereby preventing over-smoothing. The fractional-order total variation mitigates detail ambiguity and excessive smoothing, ensuring rich textures and detailed information are retained. Extensive experiments yield an average Peak Signal-to-Noise Ratio (PSNR) of 31.52 dB and a Structural Similarity Index (SSIM) of 0.863 across various benchmark datasets, including McMaster, Kodak24, BSD68, Set12, and Urban100. DGGNet outperforms existing methods, such as RIDNet, which achieved a PSNR of 31.42 dB and an SSIM of 0.853, thereby establishing new benchmarks in speckle noise reduction.

Recently, multimodal survival analysis that integrates histology images and genomic data has become a hot topic. Existing multimodal survival analysis methods have evolved from direct fusion strategies to cross-modal attention mechanisms to incorporate multimodal features. However, these methods ignore the redundancy and noise in the fusion features. To solve this problem, we introduced a Cross Modal Interaction with Information Bottleneck (CMIB) framework for multimodal survival analysis, which filters out redundancy and noise while exploring the latent complementary information across modalities. Specifically, CMIB uses the Private Feature Extraction Block (PFEB) and Common Feature Extraction Block (CFEB) to extract the private and the common features of different modalities, respectively. Subsequently, it captures the deep interactions between these features through Co-Attention (CA). Additionally, a Multimodal Information Bottleneck (MIB) is employed to yield a robust representation of the fused features. To verify the effectiveness of CMIB, we conducted extensive experiments on three public TCGA datasets. The results show that CMIB outperforms the current state-of-the-art methods.

Background:Thyroid cancer (TC) prone to cervical lymph node (CLN) metastasis both before and after surgery. Ultrasonography (US) is the first-line imaging method for evaluating the thyroid gland and CLNs. However, this assessment relies mainly on the subjective judgment of the sonographer and is very much dependent on the sonographer's experience. This prospective study was designed to construct a machine learning model based on contrast-enhanced ultrasound (CEUS) videos of CLNs to predict the risk of CLN metastasis in patients with TC. Methods:Patients who were proposed for surgical treatment due to TC from August 2019 to May 2020 were prospectively included. All patients underwent US of CLNs suspected of metastasis, and a 2-minute imaging video was recorded. After target tracking, feature extraction, and feature selection through the lymph node imaging video, three machine learning models, namely, support vector machine, linear discriminant analysis (LDA), and decision tree (DT), were constructed, and the sensitivity, specificity, and accuracy of each model for diagnosing lymph nodes were calculated by leave-one-out cross-validation (LOOCV). Results:A total of 75 lymph nodes were included in the study, with 42 benign cases and 33 malignant cases. Among the machine learning models constructed, the support vector machine had the best diagnostic efficacy, with a sensitivity of 93.0%, a specificity of 93.8%, and an accuracy of 93.3%. Conclusions:The machine learning model based on US video is helpful for the diagnosis of whether metastasis occurs in the CLNs of TC patients.

This paper presents a multi-scale hybrid method for speckle reduction in ultrasound (US) images. Speckle is removed by guided filtering, directional filtering, and fractional order filtering on the coarse to fine resolution images of a wavelet pyramid. For the directional filter, the eigen-analysis of each pixel is firstly carried out to obtain its structural features, and then it is classified into edges for filtering. Speckle noise, corresponding to the homogeneous anatomical regions, is then alleviated by the guided filter. Thereby, the algorithm reduces speckle noise while enhancing edge sharpness regardless of the size of the edges. In the synthetic images, the proposed method showed statistically significant improvements in peak signal-to-noise ratio(PSNR), structural similarity(SSIM), feature similarity index(FSIM) index and Mean Squared Error(MSE) compared with other speckle reduction methods, e.g., the squeeze boxes (SBF) filter, optimal Bayesian NLM (OBNLM) filter, speckle reducing anisotropic diffusion filter (SRAD), nonlocal low-rank framework (NLLRF) and multi-scale attention-guided neural network (MSANN). Similarly, our method outperformed the other methods in terms of mainly metrics. All the clinical images that were denoised using the six speckle reduction methods were reviewed by four radiologists for evaluation based on each radiologist’s diagnostic preferences. All the radiologists showed a significant preference for the liver images and arotid artery images obtained using our methods in terms of effectively suppresses speckle noise while preserving the structural details. For the kidney and thyroid images, our method showed similar improvement over other methods. The experimental results show that this method has better performance than other state-of-the-art medical ultrasonic image speckle removal methods.

Background Predicting tumor regression grade (TRG) after neoadjuvant chemoradiotherapy (NCRT) in patients with locally advanced rectal cancer (LARC) before surgery accurately can help to provide an individualized treatment plan. However, the predictive efficacy of traditional imaging modalities is unsatisfactory. Radiomics based on transrectal contrast-enhanced ultrasound (TR-CEUS) imaging may provide promising prediction results. Methods A total of 190 patients with LARC who underwent NCRT and subsequent total mesorectal excision were included. Based on the pathological TRG, patients were divided into the good response group (TRG 0 and 1, n = 53) and poor response group (TRG 2 and 3, n = 137). TR-CEUS examinations before and after NCRT were performed on these patients. With 10-fold cross validation, machine learning (ML) models for predicting the efficacy of NCRT were trained and established based on seven classifiers, including support vector machine, auto-encoder (AE), linear discriminant analysis, random forest (RF), logistic regression, logistic regression via Lasso or gaussian process. Predictive performances between different models were evaluated by receiver operating characteristic (ROC) curve analyses, including sensitivity, specificity, accuracy, area under the curve (AUC), 1-SE rule and Delong test. Results A total of 1525 TR-CEUS images were included for analysis and 3360 ML models were established before and after NCRT respectively. Based on the AUC and 1-SE rule, two optimal ML models were selected before and after NCRT respectively, including AE and RF. The AUC values of AE and RF models after NCRT were 0.86 and 0.84 in the cross validation set and were 0.83 and 0.84 in the testing set. The differences between RF models (D = -7.00, P < .001 for cross validation set and D = -2.04, P = .042 for testing set) and AE models (D = -7.40 for cross validation set and D = -6.68 for testing set, P < .001 for both) before and after NCRT were both statistically significant. Conclusions Radiomics models based on post-NCRT TR-CEUS images has a high predictive performance for TRG of LARC, which could accurately predict the NCRT efficacy in patients with LARC relatively.

Nowadays, using machine learning for image classification is very common. However, due to the increasing demand for data processing and fast computing, the idea of enhancing machine learning with quantum computing has been proposed, known as quantum machine learning (QML). Quantum machine learning has the advantages of higher efficiency and accuracy. Quantum computing uses quantum bits (qubits) for data storage and computing, where a qubit can represent quantum states |0〉and |1〉simultaneously, enabling the processing of information for two states simultaneously, which is unparalleled in classical computing. Moreover, quantum machine learning can handle more complex data and process data faster. In classical machine learning, the processing of large-scale data and complex problems often faces problems of high computational complexity and low algorithm efficiency. Quantum computing can handle multiple computing tasks simultaneously, achieving faster computing. Therefore, in some scenarios that require efficient computing, quantum machine learning may be the best choice. In this study, we simulated quantum circuits using Qiskit and built a hybrid quantum-classical neural network model using VQNet to classify MNIST handwritten digits and CIFAR-10 datasets. The experiments showed that quantum machine learning has the advantages of efficiency, accuracy, and security over classical machine learning, which may be an improvement over classical machine learning. This research proposes a machine learning algorithm based on quantum computing, which promotes the development of quantum computing and quantum technology. At the same time, it provides a new solution and idea for image classification, enabling people to pursue faster and more accurate quantum machine learning instead of being limited to classical machine learning. [1]

Background:Inhomogeneity within tumors can reflect tumor angiogenesis. Existing research into the quantization of angiogenesis mainly focuses on time-intensity curve parameters but has produced inconsistent results. In clinical work, it is difficult to achieve standardization and consistency for manual judgement of the inhomogeneity of contrast-enhanced images, while the artificial intelligence technology may be helpful. The aim of this study was to assess whether computers can assist in the artificial classification of tumor inhomogeneity in contrast-enhanced ultrasound (CEUS) images of rectal cancer.Methods:A total of 500 contrast-enhanced ultrasonograms were retrospectively collected, which was verified of rectal cancer pathologically from 2016 to 2018 as training set. All images are from 18-80 years old patients with rectal cancer in our hospital. These tumors are usually located in the middle and lower segment of the rectum, which can be completely observed on ultrasound. The images were divided into 3 categories according to the inhomogeneous distribution of contrast agents inside the tumors. Computing methods were used to simulate manual classification. Computer processing steps included segmentation, gray level quantization, dimension reduction, and classification. The results of 6 different gray level quantization, 2 dimensionality reduction methods, and 3 classifiers were compared, from which the optimal parameters were selected in each step. The performance of computer classification was evaluated using manual classification results as the reference. Ninety-seven ultrasonograms of contrast-enhanced rectal tumors were collected as validation set from 2018.1 to 2018.6.Results:The optimal gray level was set at 32. Principal component analysis (PCA) was the first choice for dimensionality reduction. The best classifier was support vector machines (SVM). The accuracy of computer classification was 87.80% (439/500). The accuracy of computer classification in the validation cohort was 60.82%. The area under the curve (AUC) of class 1, 2, and 3 were 0.76, 0.41, and 0.48, respectively.Conclusions:Results showed that the computer methods are competent for classifying inhomogeneity of contrast-enhanced rectal cancers inside ultrasonograms.

对于颈部淋巴结的超声造影视频病例,可利用其时间强度曲线提取灌注特征进行病情诊断.现有的研究方法对感兴趣区域进行像素级的分析可以更准确地描述灌注特征,然而很少有深入研究灌注流向可视化的方法.本文利用了像素级的时间强度曲线TIC分析,采取双重筛选方式对TIC曲线进行筛选,进而针对TIC曲线提取二维灌注参数对灌注流向进行可视化.特征提取后生成的流线图像能够一定程度反映血管的分布,对医生病情诊断有一定的辅助价值,也可以对微血管重构有一定的启发价值.

随着科学技术的不断发展,医学诊断技术也在不断的进步之中,超声技术作为一种医学诊断手段已广泛地应用于各个医疗领域,并且由于对人体的无害性以及能够动态且清晰地展现人体组织和器官的健康状态从而普遍得到了医生和患者的认可.在超声技术的不断发展中,人们对超声实时成像质量上的要求显著提高,由于超声探头的材质例如陶瓷换能器制造的局限性以及在降低成本及帧速率等原因而采用的低通道扫描的折中方案所造成的噪点和伪影会遮挡人体组织和器官的有用信息从而严重影响医生的辅助诊断,在超声领域如何进行图像及视频的增强和伪影的抑制成为一个重要的挑战.本文首先描述了几种空间域抑制伪影的滤波算法及其局限性,并提出了一种基于频率域的伪影抑制算法,该算法能够良好的抑制在超声实时成像中的周期性伪影,本文先通过正弦波模拟周期性伪影实验以突显其在频率域上的特性,然后将超声图像进行二维傅立叶变换到频率域来对这些伪影进行抑制,由于这些伪影具有周期性,所以在频率域上具有明显的特征,本文通过滑动窗口扫描结合阈值的算法模型找出频率域上对应这些伪影的集合,然后根据频域的动态范围及给定的阈值来对集合中的这些疑似伪影的点进行压低处理,再通过反傅立叶变换将超声图像变换到空间域上来从而得到处理后的图像.通过这种方法,能够提高超声图像对周期性伪影抑制且保留有用的信息,能够提高医生对人体器官状况的判断结果的准确性.

AbstractPurposeAlthough homocysteine (Hcy) has been proven to be associated with the incidence of white matter hyperintensities (WMH) in patients with stroke, this association remains unclear in participants with asymptomatic intracranial arterial stenosis (aICAS). This study aimed to investigate the association of Hcy with WMH in participants with aICAS.Materials and methodsThis was a cross‐sectional study based on the Kongcun Town Study. Participants diagnosed with aICAS by magnetic resonance angiography in the Kongcun Town Study were enrolled in this study. Data on demographics, lifestyle, medical histories, and Hcy levels were collected via interviews, clinical examinations, and laboratory tests. The volume of WMH was calculated using the lesion segmentation tool system for the Statistical Parametric Mapping package based on magnetic resonance imaging. The association between Hcy and WMH volume was analyzed using linear and logistic regression analyses.ResultsA total of 137 aICAS participants were enrolled in the present study. Hcy was associated with the incidence of severe WMH (4th quartile, ≥4.20 ml) after adjustment for certain covariates [Hcy as a continuous variable, odds ratio (95% confidence interval) (OR (95% CI)): 1.09 (1.00, 1.19), p = .047; as a categorical variable (Hcy ≥15 μmol/L), OR (95% CI): 3.74 (1.37, 10.19), p = .010)]. After stratification according to the degree of aICAS, this relationship remained significant only in the moderate‐to‐severe stenosis group (stenosis ≥50%). (Hcy as continuous variable, OR (95% CI): 1.14 (1.02, 1.27), p = .025; as a categorical variable (Hcy ≥15 μmol/L), OR (95% CI): 5.59 (1.40, 15.25), p = .015).ConclusionSerum Hcy concentration may be positively associated with the volume of WMH in rural‐dwelling Chinese people with moderate‐to‐severe (stenosis ≥50%) aICAS.

In order to effectively segment the visceral adipose tissue and help the doctors to rapidly diagnose the potential risks of metabolic syndrome, here we developed a deep learning-based method which is based on the U-net architecture for segmenting and measuring the visceral adipose tissue(VAT). And even no matter which orientation that the operator takes, the model can segment the visceral fat area and then use the appropriate outputs to compute the max thickness of VAT. One hundred and fourteen healthy volunteers were enrolled in this study. Ultrasound(US) was performed, and then the visceral adipose tissue was segmented and measured by the model that we use. We regard the distance behind the linea alba in the xiphoid process as the thickest visceral adipose tissue(VAT max). The dice score and accuracy are 3.46%, 96.44% respectively. In addition, compared with the manually outlined segmentation, the pearson correlation coefficient and the mean relative error (MRE) are R=0.9231 (P<0.001) and 10.12% in the measurement of the VAT max between original and output images. The auto-segmentation and measurement of visceral adipose tissue on ultrasound method demonstrate the accuracy of deep learning in segmentation and measurement of visceral adipose tissue.

WFOV (wide field of view) imaging mode is used to provide an ultrasound image much larger than the normal probe field of view, which is achieved by compounding successive frames in a sequence as the probe moves and scans. The eraser feature in this paper is designed to erase the undesirable part of panorama image with reverse probe movement. To implement the eraser function we propose a method which is a three step process composed of probe movement detection, pixel erasing and smooth connection of added field of view image. Here we compute the position of real-time image relative to the static prior image based on image registration and spatial transformation for detecting probe motion. The fine-tuning in Y-direction and line fusion based on persistence method are used for smooth connection between prior WFOV image and added field-of view image.

Pulsed Wave Doppler (PW) is a traditional ultrasound technique used for the diagnosis of vascular diseases. The conventional diagnostic method is mainly based on hemodynamic parameters obtained from the PW spectrum. However, it relies on clinical observation and medical experience through lots of patient data investigation and analysis. The collected patient data are varied by different ultrasound equipment, detection regions and operation techniques, resulting in different image styles, which decreases the application and generality of the conventional method. And this method also has a strong dependence on patients’ data, especially on negative samples. Thus this paper proposes a rapid disease screening method, named as PauTa Criterion, which is based on statistical distribution characteristics for screening out anomalous targets. The proposed rapid screening method is based on multiple hemodynamic parameters to detect the outliers that are different from healthy samples. Compared with the conventional methods, the proposed method does not rely on a fixed or single ultrasound system and has low sensitivity to system noise. The experimental results show that the proposed method reaches a high accuracy of 93.14%, which is at least 20% higher than existing clustering methods, K-Means and Support Vector Machine (SVM). Accordingly, high accuracy and fast convergence makes the prospect of the proposed method to be used for rapid disease screening possible.

It is a significant challenge to obtain accurate boundary of tumors due to much speckle noises. In this paper, we proposed some meaningful modules to address the problem. Firstly, a large amount of semantic information is needed to determine the boundary on account of fuzziness around the edge of breast tumor, we proposed residual multi-scale(RMS) block to collect larger receptive filed. Secondly, we redesigned the skip connections and combined Squeeze-and-Excitation(SE) block to incorporate information from different layers. Finally we introduced deep supervision and hybrid loss function to accelerate the convergence of network. Dice similarity coefficient and intersection over union(IoU) were used to evaluate segmentation results which were 94.69 % and 90.01 % respectively on test set. It is shown that our method is effective in this kind of problem.

Liver fiber is an intermediate and reversible link in the process of cirrhosis, early detection and intervention of liver fibrosis is of great significance for the development and prognosis. Ultrasound as a one of the common diagnostic methods for liver fibers has many advantages like convenience, great accuracy and robustness. However, it is difficult to get subjective and uniform diagnoses because of the ultrasound images can be inevitably affected by the device characteristics, the interactions between ultrasound and body tissues, operation approaches and other uncontrollable factors. So we proposed a new liver fibrosis detecting algorithm based on the ultrasound echo amplitude analysis and a deep learning to classify normal and fibrosis tissue in computer simulation data. In order to study the relationship between scatterer density and hepatic fibrosis, we simulated various scatterer density liver fibrosis ultrasound image by creating random scatterer field and convolving with point spread function. Compared with the detection of traditional statistical analysis and parameter imaging, we use the data of echo amplitude distribution and image gray histogram distribution to classify the category of the window by CNN. The result shows that CNN can provide a better performance in classification and prediction than parameter imaging.