Biomedical Signal Processing and Control最新文献_第4页

EffiCAT: A synergistic approach to skin disease classification through multi-dataset fusion and attention mechanisms EffiCAT：通过多数据集融合和关注机制实现皮肤病分类的协同方法

IF 4.9 2区医学 Q1 ENGINEERING, BIOMEDICAL

Biomedical Signal Processing and Control

Pub Date : 2024-11-11 DOI: 10.1016/j.bspc.2024.107141

A. Sasithradevi , S. Kanimozhi , Parasa Sasidhar , Pavan Kumar Pulipati , Elavarthi Sruthi , P. Prakash

Early and accurate diagnosis of skin diseases is essential for their efficient treatment and effective management. Conventional approaches typically depend on the use of a single dataset, which can introduce biases and limit the generalizability of the models due to dataset-specific idiosyncrasies. This study presents a novel hybrid model, named EffiCAT (EfficientNet Concatenation Attention Technology), for the categorization of skin diseases, specifically focusing on four classes named Actinic Keratosis (ACK), Basal Cell Carcinoma (BCC), Melanoma (MEL), and Melanocytic Nevus (NEV). EffiCAT enhances traditional approaches by integrating features from two different convolutional neural networks, EfficientNet B0 and EfficientNet B4, through feature concatenation. This is followed by applying advanced attention modules, specifically a Dual Channel Attention Layer applied twice and a Convolutional Block Attention Module (CBAM), to refine feature representation and focus on relevant patterns more effectively. Our method is evaluated on a combined dataset composed of HAM10000 and PAD-UFES-20, which enhances the diversity and volume of training samples to improve generalization across various skin types and conditions. The inclusion of multiple datasets helps mitigate the biases associated with single-dataset training and enhances the robustness of the model. EffiCAT attained a test accuracy of 94.48%, with precision, recall, and F1 score all closely aligned at 94.48%. These metrics not only illustrate the efficacy of our method but also underscore its superiority in handling varied and complex skin disease presentations through refined attention-driven feature concatenation. Additionally, external validation was performed on the ISIC 2018 dataset, where the model achieved a test accuracy of 92.08%, with precision of 92.45%, recall of 92.08%, and an F1 score of 92.15%, further confirming its robustness and generalizability. The model’s architecture efficiently leverages concatenated features enriched with attention mechanisms, setting a new standard for image-based diagnostic models.

皮肤病的早期准确诊断对于高效治疗和有效管理至关重要。传统方法通常依赖于使用单一数据集，这可能会引入偏差，并因数据集的特异性而限制模型的普适性。本研究提出了一种名为 EffiCAT（EfficientNet Concatenation Attention Technology）的新型混合模型，用于对皮肤病进行分类，尤其侧重于四类皮肤病，分别是角化性皮肤病（ACK）、基底细胞癌（BCC）、黑色素瘤（MEL）和黑素细胞痣（NEV）。EffiCAT 通过特征串联将来自两个不同卷积神经网络（EfficientNet B0 和 EfficientNet B4）的特征整合在一起，从而改进了传统方法。然后再应用高级注意力模块，特别是应用两次的双通道注意力层和卷积块注意力模块（CBAM），以完善特征表示并更有效地关注相关模式。我们的方法是在由 HAM10000 和 PAD-UFES-20 组成的组合数据集上进行评估的，该数据集增强了训练样本的多样性和数量，从而提高了在各种皮肤类型和条件下的泛化能力。包含多个数据集有助于减轻与单一数据集训练相关的偏差，并增强模型的鲁棒性。EffiCAT 的测试准确率达到 94.48%，精确度、召回率和 F1 分数均接近 94.48%。这些指标不仅说明了我们的方法的有效性，还强调了它在通过精炼的注意力驱动特征串联处理各种复杂的皮肤病表现方面的优越性。此外，我们还在 ISIC 2018 数据集上进行了外部验证，该模型的测试准确率为 92.08%，精确率为 92.45%，召回率为 92.08%，F1 得分为 92.15%，进一步证实了其稳健性和普适性。该模型的架构有效利用了富含注意力机制的串联特征，为基于图像的诊断模型设定了新标准。

{"title":"EffiCAT: A synergistic approach to skin disease classification through multi-dataset fusion and attention mechanisms","authors":"A. Sasithradevi , S. Kanimozhi , Parasa Sasidhar , Pavan Kumar Pulipati , Elavarthi Sruthi , P. Prakash","doi":"10.1016/j.bspc.2024.107141","DOIUrl":"10.1016/j.bspc.2024.107141","url":null,"abstract":"<div><div>Early and accurate diagnosis of skin diseases is essential for their efficient treatment and effective management. Conventional approaches typically depend on the use of a single dataset, which can introduce biases and limit the generalizability of the models due to dataset-specific idiosyncrasies. This study presents a novel hybrid model, named EffiCAT (EfficientNet Concatenation Attention Technology), for the categorization of skin diseases, specifically focusing on four classes named Actinic Keratosis (ACK), Basal Cell Carcinoma (BCC), Melanoma (MEL), and Melanocytic Nevus (NEV). EffiCAT enhances traditional approaches by integrating features from two different convolutional neural networks, EfficientNet B0 and EfficientNet B4, through feature concatenation. This is followed by applying advanced attention modules, specifically a Dual Channel Attention Layer applied twice and a Convolutional Block Attention Module (CBAM), to refine feature representation and focus on relevant patterns more effectively. Our method is evaluated on a combined dataset composed of HAM10000 and PAD-UFES-20, which enhances the diversity and volume of training samples to improve generalization across various skin types and conditions. The inclusion of multiple datasets helps mitigate the biases associated with single-dataset training and enhances the robustness of the model. EffiCAT attained a test accuracy of 94.48%, with precision, recall, and F1 score all closely aligned at 94.48%. These metrics not only illustrate the efficacy of our method but also underscore its superiority in handling varied and complex skin disease presentations through refined attention-driven feature concatenation. Additionally, external validation was performed on the ISIC 2018 dataset, where the model achieved a test accuracy of 92.08%, with precision of 92.45%, recall of 92.08%, and an F1 score of 92.15%, further confirming its robustness and generalizability. The model’s architecture efficiently leverages concatenated features enriched with attention mechanisms, setting a new standard for image-based diagnostic models.</div></div>","PeriodicalId":55362,"journal":{"name":"Biomedical Signal Processing and Control","volume":"100 ","pages":"Article 107141"},"PeriodicalIF":4.9,"publicationDate":"2024-11-11","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142658293","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Enhanced diagnosis of thyroid-associated eye diseases based on deep learning: A novel triplet loss design strategy 基于深度学习的甲状腺相关眼病强化诊断：新型三重损失设计策略

IF 4.9 2区医学 Q1 ENGINEERING, BIOMEDICAL

Biomedical Signal Processing and Control

Pub Date : 2024-11-11 DOI: 10.1016/j.bspc.2024.107161

Zhenyong Qian , Ke Li , Miaomiao Kong , Tianli Qin , Wentao Yan , Zixuan Xi , Tao Wu , Hongliang Zhong , Wencan Wu , Jianzhang Wu , Wulan Li

Thyroid-associated ophthalmopathy (TAO) is an orbital disease that significantly impacts patients’ quality of life. The early diagnosis and treatment of TAO are faced with many difficulties, so some studies have attempted to identify and diagnose TAO at an early stage. However, the diagnostic classification in relevant studies is all based on traditional cross-entropy loss, and the accuracy will decrease under complex conditions with high similarity of eye images. To enhance the precision of TAO diagnosis, this study introduces a data metric method called IP-Triplet, based on triplet loss. Given the data characteristics, we select the DenseNet backbone network for optimization to better extract features from eye images. However, merely modifying the network structure is insufficient. Therefore, inspired by C-Triplet, we use ‘class proxy’ concept to replace the positive and negative samples in the triplet and adjust the distance between the triplets using an enhancement mapper to improve training effectiveness. Finally, this approach is combined with the cross-entropy loss function for mixed training. Our experimental results show that the proposed IP-Triplet loss significantly enhances TAO diagnostic accuracy, achieving a classification accuracy of 95.97 %±0.09, an F1 score of 95.98 %±0.09, and a quadratic weighted kappa score of 96.96 %±0.07. Our model outperforms existing studies on two public datasets, OCT-2017 and OCT-C8, with an accuracy of 99.80 % and 98.18 %, a recall of 99.80 % and 98.18 %, and a precision of 99.80 % and 98.20 %, respectively. Notably, IP-Triplet can be easily integrated into existing CNN models, providing robust support for TAO diagnosis and treatment. The source code is available at https://github.com/lwlwzmu/IP_Triplet_Classification.

甲状腺相关眼病（TAO）是一种严重影响患者生活质量的眼眶疾病。TAO的早期诊断和治疗面临诸多困难，因此一些研究试图对TAO进行早期识别和诊断。然而，相关研究中的诊断分类都是基于传统的交叉熵损失，在眼部图像相似度较高的复杂条件下，准确率会有所下降。为了提高 TAO 诊断的精确度，本研究引入了一种基于三重损失的数据度量方法--IP-Triplet。根据数据特征，我们选择 DenseNet 骨干网络进行优化，以更好地提取眼部图像的特征。然而，仅仅修改网络结构是不够的。因此，受 C-Triplet 的启发，我们使用 "类代理 "概念来替换三元组中的正负样本，并使用增强映射器调整三元组之间的距离，以提高训练效果。最后，这种方法与交叉熵损失函数相结合，用于混合训练。我们的实验结果表明，所提出的 IP-Triplet 损失能显著提高 TAO 诊断的准确性，分类准确率达到 95.97 %±0.09，F1 分数达到 95.98 %±0.09，二次加权卡帕分数达到 96.96 %±0.07。我们的模型在两个公共数据集 OCT-2017 和 OCT-C8 上的表现优于现有研究，准确率分别为 99.80 % 和 98.18 %，召回率分别为 99.80 % 和 98.18 %，精确率分别为 99.80 % 和 98.20 %。值得注意的是，IP-Triplet 可以轻松集成到现有的 CNN 模型中，为 TAO 诊断和治疗提供强大的支持。源代码见 https://github.com/lwlwzmu/IP_Triplet_Classification。

{"title":"Enhanced diagnosis of thyroid-associated eye diseases based on deep learning: A novel triplet loss design strategy","authors":"Zhenyong Qian , Ke Li , Miaomiao Kong , Tianli Qin , Wentao Yan , Zixuan Xi , Tao Wu , Hongliang Zhong , Wencan Wu , Jianzhang Wu , Wulan Li","doi":"10.1016/j.bspc.2024.107161","DOIUrl":"10.1016/j.bspc.2024.107161","url":null,"abstract":"<div><div>Thyroid-associated ophthalmopathy (TAO) is an orbital disease that significantly impacts patients’ quality of life. The early diagnosis and treatment of TAO are faced with many difficulties, so some studies have attempted to identify and diagnose TAO at an early stage. However, the diagnostic classification in relevant studies is all based on traditional cross-entropy loss, and the accuracy will decrease under complex conditions with high similarity of eye images. To enhance the precision of TAO diagnosis, this study introduces a data metric method called IP-Triplet, based on triplet loss. Given the data characteristics, we select the DenseNet backbone network for optimization to better extract features from eye images. However, merely modifying the network structure is insufficient. Therefore, inspired by C-Triplet, we use ‘class proxy’ concept to replace the positive and negative samples in the triplet and adjust the distance between the triplets using an enhancement mapper to improve training effectiveness. Finally, this approach is combined with the cross-entropy loss function for mixed training. Our experimental results show that the proposed IP-Triplet loss significantly enhances TAO diagnostic accuracy, achieving a classification accuracy of 95.97 %±0.09, an F1 score of 95.98 %±0.09, and a quadratic weighted kappa score of 96.96 %±0.07. Our model outperforms existing studies on two public datasets, OCT-2017 and OCT-C8, with an accuracy of 99.80 % and 98.18 %, a recall of 99.80 % and 98.18 %, and a precision of 99.80 % and 98.20 %, respectively. Notably, IP-Triplet can be easily integrated into existing CNN models, providing robust support for TAO diagnosis and treatment. The source code is available at <span><span>https://github.com/lwlwzmu/IP_Triplet_Classification</span><svg><path></path></svg></span>.</div></div>","PeriodicalId":55362,"journal":{"name":"Biomedical Signal Processing and Control","volume":"100 ","pages":"Article 107161"},"PeriodicalIF":4.9,"publicationDate":"2024-11-11","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142657559","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

DeepBrainTumorNet: An effective framework of heuristic-aided brain Tumour detection and classification system using residual Attention-Multiscale Dilated inception network 深度脑肿瘤网络：使用残差注意-多尺度稀释感知网络的启发式辅助脑肿瘤检测和分类系统的有效框架

IF 4.9 2区医学 Q1 ENGINEERING, BIOMEDICAL

Biomedical Signal Processing and Control

Pub Date : 2024-11-11 DOI: 10.1016/j.bspc.2024.107180

A. Vinisha , Ravi Boda

A tumor is formed by controlled and rapid cell growth in the brain. If it is not treated in the early stages, it leads to death. Despite many important efforts and promising solutions, accurate segmentation and classification remain a challenge. Traditional automated models have complex architectures, high computing systems, and large amounts of data. Also, most of the existing models still rely on manual intervention. To address all the limitations, a new deep-learning approach is proposed. Initially, brain images are collected from standard sources and passed to the pre-processing stage. In this stage, the input images are pre-processed using scaling, contrast enhancement, and Anisotropic Diffusion Filtering (ADF). Later, the resultant images are provided to the image segmentation stage. Here, image segmentation is performed using adaptive transunet3+, and also their parameters are optimized by using the Hybrid Beluga Whale Glowworm Swarm Optimization (HBWGSO). Further, brain tumor classifications are performed with the hybridization of Residual Attention Network (RAN) and Multiscale Dilated Inception Network (MDIN) termed RA-MDIN, and the model parameters are optimally selected by using the designed HBWGSO approach. Through the experimentation results, the proposed model tends to provide effective classification results. Thus, the recommended system guarantees to yield relatively satisfactory outcomes over conventional mechanisms.

肿瘤是由脑内细胞受控快速生长形成的。如果在早期得不到治疗，就会导致死亡。尽管有许多重要的努力和有前景的解决方案，但准确的分割和分类仍然是一项挑战。传统的自动模型具有复杂的架构、高计算系统和大量数据。此外，大多数现有模型仍然依赖人工干预。为了解决所有这些限制，我们提出了一种新的深度学习方法。起初，大脑图像是从标准来源收集的，并进入预处理阶段。在这一阶段，输入图像通过缩放、对比度增强和各向异性扩散滤波（ADF）进行预处理。之后，生成的图像将提供给图像分割阶段。在此，使用自适应 Transunet3+ 进行图像分割，并使用混合白鲸萤火虫群优化（HBWGSO）对其参数进行优化。此外，还利用残留注意力网络（RAN）和多尺度稀释感知网络（MDIN）的混合方法（称为 RA-MDIN）进行脑肿瘤分类，并利用所设计的 HBWGSO 方法对模型参数进行优化选择。通过实验结果，建议的模型往往能提供有效的分类结果。因此，与传统机制相比，所推荐的系统能保证产生相对令人满意的结果。

{"title":"DeepBrainTumorNet: An effective framework of heuristic-aided brain Tumour detection and classification system using residual Attention-Multiscale Dilated inception network","authors":"A. Vinisha , Ravi Boda","doi":"10.1016/j.bspc.2024.107180","DOIUrl":"10.1016/j.bspc.2024.107180","url":null,"abstract":"<div><div>A tumor is formed by controlled and rapid cell growth in the brain. If it is not treated in the early stages, it leads to death. Despite many important efforts and promising solutions, accurate segmentation and classification remain a challenge. Traditional automated models have complex architectures, high computing systems, and large amounts of data. Also, most of the existing models still rely on manual intervention. To address all the limitations, a new deep-learning approach is proposed. Initially, brain images are collected from standard sources and passed to the pre-processing stage. In this stage, the input images are pre-processed using scaling, contrast enhancement, and Anisotropic Diffusion Filtering (ADF). Later, the resultant images are provided to the image segmentation stage. Here, image segmentation is performed using adaptive transunet3+, and also their parameters are optimized by using the Hybrid Beluga Whale Glowworm Swarm Optimization (HBWGSO). Further, brain tumor classifications are performed with the hybridization of Residual Attention Network (RAN) and Multiscale Dilated Inception Network (MDIN) termed RA-MDIN, and the model parameters are optimally selected by using the designed HBWGSO approach. Through the experimentation results, the proposed model tends to provide effective classification results. Thus, the recommended system guarantees to yield relatively satisfactory outcomes over conventional mechanisms.</div></div>","PeriodicalId":55362,"journal":{"name":"Biomedical Signal Processing and Control","volume":"100 ","pages":"Article 107180"},"PeriodicalIF":4.9,"publicationDate":"2024-11-11","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142658297","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

An adaptive approach for securing patient data in intellectual disability care with 8D Hyperchaotic DNA encryption and IWT 利用 8D Hyperchaotic DNA 加密和 IWT 保护智障护理中患者数据安全的自适应方法

IF 4.9 2区医学 Q1 ENGINEERING, BIOMEDICAL

Biomedical Signal Processing and Control

Pub Date : 2024-11-11 DOI: 10.1016/j.bspc.2024.107174

J.R. Anisha, Y.P. Arul Teen

Intellectual disability (ID) is a neurodevelopmental disease characterized by significant intellectual and adaptive functioning impairments. Continuous monitoring and data collection are crucial for managing this condition. To collect patient’s skin temperature data, doctors and caretakers often use thermal IR imaging, and they must securely share this information. To address privacy concerns and ensure robust encryption during data transmission, this study introduces an adaptive encryption algorithm that operates in both frequency and spatial domains. The algorithm counteracts various security threats, including plain text attacks, differential attacks, brute force attacks, and cropping attacks. The core of this research is the 8D Hyperchaotic DNA Encryption Algorithm, which enhances security by combining block scrambling with DNA coding. Security is further optimized through integrating the Integer Wavelet Transform (IWT) in the frequency domain and a DNA sequence in the spatial domain. The encryption process aims to balance security and computational efficiency, ensuring reliable performance. Extensive testing and validation against multiple types of attacks show the algorithm’s effectiveness and reliability in practical applications. Experimental results show that the proposed system significantly enhances security against various attacks, thus setting a higher standard for encryption in biomedical image and signal processing. This algorithm shows potential for broader applications beyond intellectual disability care, including secure data transmission in medical imaging, financial transactions, and confidential communications, because of its robust encryption capabilities and adaptability to different data types.

智力障碍（ID）是一种神经发育疾病，其特点是智力和适应功能明显受损。持续监测和数据收集对于管理这种疾病至关重要。为了收集患者的皮肤温度数据，医生和护理人员通常会使用红外热成像技术，而且他们必须安全地共享这些信息。为了解决隐私问题并确保数据传输过程中的稳健加密，本研究引入了一种在频率和空间域均可运行的自适应加密算法。该算法可抵御各种安全威胁，包括明文攻击、差分攻击、暴力攻击和裁剪攻击。这项研究的核心是 8D 超混沌 DNA 加密算法，它通过将块加扰与 DNA 编码相结合来增强安全性。通过整合频域的整数小波变换（IWT）和空间域的 DNA 序列，进一步优化了安全性。加密过程旨在兼顾安全性和计算效率，确保性能可靠。针对多种类型攻击的广泛测试和验证表明，该算法在实际应用中非常有效和可靠。实验结果表明，所提出的系统大大增强了抵御各种攻击的安全性，从而为生物医学图像和信号处理中的加密设定了更高的标准。该算法因其强大的加密能力和对不同数据类型的适应性，在智障护理以外的更广泛应用中显示出潜力，包括医疗成像、金融交易和保密通信中的安全数据传输。

{"title":"An adaptive approach for securing patient data in intellectual disability care with 8D Hyperchaotic DNA encryption and IWT","authors":"J.R. Anisha, Y.P. Arul Teen","doi":"10.1016/j.bspc.2024.107174","DOIUrl":"10.1016/j.bspc.2024.107174","url":null,"abstract":"<div><div>Intellectual disability (ID) is a neurodevelopmental disease characterized by significant intellectual and adaptive functioning impairments. Continuous monitoring and data collection are crucial for managing this condition. To collect patient’s skin temperature data, doctors and caretakers often use thermal IR imaging, and they must securely share this information. To address privacy concerns and ensure robust encryption during data transmission, this study introduces an adaptive encryption algorithm that operates in both frequency and spatial domains. The algorithm counteracts various security threats, including plain text attacks, differential attacks, brute force attacks, and cropping attacks. The core of this research is the 8D Hyperchaotic DNA Encryption Algorithm, which enhances security by combining block scrambling with DNA coding. Security is further optimized through integrating the Integer Wavelet Transform (IWT) in the frequency domain and a DNA sequence in the spatial domain. The encryption process aims to balance security and computational efficiency, ensuring reliable performance. Extensive testing and validation against multiple types of attacks show the algorithm’s effectiveness and reliability in practical applications. Experimental results show that the proposed system significantly enhances security against various attacks, thus setting a higher standard for encryption in biomedical image and signal processing. This algorithm shows potential for broader applications beyond intellectual disability care, including secure data transmission in medical imaging, financial transactions, and confidential communications, because of its robust encryption capabilities and adaptability to different data types.</div></div>","PeriodicalId":55362,"journal":{"name":"Biomedical Signal Processing and Control","volume":"100 ","pages":"Article 107174"},"PeriodicalIF":4.9,"publicationDate":"2024-11-11","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142658267","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

HET-RL: Multiple pulmonary disease diagnosis via hybrid efficient transformers based representation learning model using multi-modality data HET-RL：利用多模态数据，通过基于混合高效变压器的表征学习模型，诊断多种肺部疾病

IF 4.9 2区医学 Q1 ENGINEERING, BIOMEDICAL

Biomedical Signal Processing and Control

Pub Date : 2024-11-11 DOI: 10.1016/j.bspc.2024.107157

A.P. Narmadha, N. Gobalakrishnan

Pulmonary diseases, encompassing conditions such as chronic bronchitis, emphysema, asthma, and pulmonary fibrosis, involve intricate pathophysiological mechanisms affecting the respiratory system, necessitating precise diagnosis and tailored therapeutic approaches. Timely and accurate diagnosis of pulmonary diseases is crucial as it enables early intervention, optimal management, and prevention of complications, thereby improving patient outcomes and quality of life. The scarcity of multi-modality datasets and challenges in accurate diagnosis underscore the complexities faced by deep learning models in achieving comprehensive pulmonary diagnoses, emphasizing the need for enhanced data diversity and algorithmic robustness in addressing diagnostic issues. To overcome these challenges, we have proposed a hybrid efficient transformer based on representation learning named as HET-RL model for accurate various pulmonary disease using multi-modality. Primarily in our work, we utilized multiple data including radiograph, chief complaints and clinical parameters for enhancing the efficiency of model performance. We have performed dual-level pre-processing such as denoising and normalization for amplifying data quality using Steered Filter (SF) and Min-Max normalization, respectively. Then, we proposed HET-RL model which encompasses of Inter-Attention Transformer (IAT) and Text Analyzer Transformer (TAT) for data analyzing. In which, appropriate features are analyzed and extracted from radiograph (CT scan) by IAT and TAT with representation learning (RL) encode the text and extract significant information from both chief complaint clinical parameters. Finally, the extracted features from hybrid transformers are fused by adapting Fusion Network and then pulmonary disease are classified into multiple classes. Incorporating diverse data sources, including results of laboratory test and patient demographic characteristics, our model demonstrated superior performance compared to non-unified multimodal and an image-only model diagnosis model. It exhibited a 12% and 9% improvement in identifying pulmonary disease. Multimodal hybrid transformer-based models hold promise for streamlining patient triaging and enhancing the clinical decision-making process.

肺部疾病包括慢性支气管炎、肺气肿、哮喘和肺纤维化等病症，涉及影响呼吸系统的复杂病理生理机制，需要精确诊断和有针对性的治疗方法。及时、准确地诊断肺部疾病至关重要，因为这有助于早期干预、优化管理和预防并发症，从而改善患者的预后和生活质量。多模态数据集的稀缺和准确诊断的挑战凸显了深度学习模型在实现全面肺部诊断时所面临的复杂性，强调了在解决诊断问题时增强数据多样性和算法鲁棒性的必要性。为了克服这些挑战，我们提出了一种基于表征学习的混合高效变换器，命名为 HET-RL 模型，用于利用多模态准确诊断各种肺部疾病。在我们的工作中，我们主要利用了多种数据，包括 X 光片、主诉和临床参数，以提高模型性能的效率。我们进行了双重预处理，如去噪和归一化，分别使用 Steered Filter（SF）和 Min-Max 归一化来提高数据质量。然后，我们提出了 HET-RL 模型，其中包括用于数据分析的 Inter-Attention Transformer（IAT）和 Text Analyzer Transformer（TAT）。其中，IAT 和 TAT 通过表征学习（RL）对文本进行编码，从放射照片（CT 扫描）中分析并提取适当的特征，并从主诉和临床参数中提取重要信息。最后，通过自适应融合网络将从混合变换器中提取的特征进行融合，然后将肺部疾病分为多个类别。与非统一多模态诊断模型和纯图像诊断模型相比，我们的模型结合了包括实验室检测结果和患者人口特征在内的多种数据源，表现出更优越的性能。它在识别肺部疾病方面分别提高了 12% 和 9%。基于多模态混合变压器的模型有望简化病人分流和加强临床决策过程。

{"title":"HET-RL: Multiple pulmonary disease diagnosis via hybrid efficient transformers based representation learning model using multi-modality data","authors":"A.P. Narmadha, N. Gobalakrishnan","doi":"10.1016/j.bspc.2024.107157","DOIUrl":"10.1016/j.bspc.2024.107157","url":null,"abstract":"<div><div>Pulmonary diseases, encompassing conditions such as chronic bronchitis, emphysema, asthma, and pulmonary fibrosis, involve intricate pathophysiological mechanisms affecting the respiratory system, necessitating precise diagnosis and tailored therapeutic approaches. Timely and accurate diagnosis of pulmonary diseases is crucial as it enables early intervention, optimal management, and prevention of complications, thereby improving patient outcomes and quality of life. The scarcity of multi-modality datasets and challenges in accurate diagnosis underscore the complexities faced by deep learning models in achieving comprehensive pulmonary diagnoses, emphasizing the need for enhanced data diversity and algorithmic robustness in addressing diagnostic issues. To overcome these challenges, we have proposed a hybrid efficient transformer based on representation learning named as HET-RL model for accurate various pulmonary disease using multi-modality. Primarily in our work, we utilized multiple data including radiograph, chief complaints and clinical parameters for enhancing the efficiency of model performance. We have performed dual-level pre-processing such as denoising and normalization for amplifying data quality using Steered Filter (SF) and Min-Max normalization, respectively. Then, we proposed HET-RL model which encompasses of Inter-Attention Transformer (IAT) and Text Analyzer Transformer (TAT) for data analyzing. In which, appropriate features are analyzed and extracted from radiograph (CT scan) by IAT and TAT with representation learning (RL) encode the text and extract significant information from both chief complaint clinical parameters. Finally, the extracted features from hybrid transformers are fused by adapting Fusion Network and then pulmonary disease are classified into multiple classes. Incorporating diverse data sources, including results of laboratory test and patient demographic characteristics, our model demonstrated superior performance compared to non-unified multimodal and an image-only model diagnosis model. It exhibited a 12% and 9% improvement in identifying pulmonary disease. Multimodal hybrid transformer-based models hold promise for streamlining patient triaging and enhancing the clinical decision-making process.</div></div>","PeriodicalId":55362,"journal":{"name":"Biomedical Signal Processing and Control","volume":"100 ","pages":"Article 107157"},"PeriodicalIF":4.9,"publicationDate":"2024-11-11","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142652845","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

YOLO-HV: A fast YOLOv8-based method for measuring hemorrhage volumes YOLO-HV：基于 YOLOv8 的出血量快速测量方法

IF 4.9 2区医学 Q1 ENGINEERING, BIOMEDICAL

Biomedical Signal Processing and Control

Pub Date : 2024-11-11 DOI: 10.1016/j.bspc.2024.107131

Haoran Wang, Guohui Wang, Yongliang Li, Kairong Zhang

Measuring the volume of a cerebral hemorrhage is crucial for clinical diagnosis and treatment. It helps doctors assess the severity of the bleeding, guide treatment decisions, and improve patient survival rates and quality of life. However, due to the irregularity and fluid nature of the hemorrhages, existing methods struggle to segment and measure different hemorrhage instances. This paper introduces an efficient cerebral hemorrhage segmentation network, YOLO-HV, based on YOLOv8n-seg, designed for volumetric measurement of cerebral hemorrhages. To enhance the extraction of spatial feature information from irregular hemorrhagic areas, A CoordAttention mechanism is integrated into the backbone of the network. Addressing the limitations of lightweight models in training with large-scale data, a GDConv (Ghost Dynamic Convolution) module is introduced in the Neck component to replace the original C2f module. The original detection head is replaced with LGND (Lightweight Group Normalized Detection Head), enhancing positioning and classification performance of the network while additionally reducing computational costs. A Union-Find is used on a spatial level to match cross-layer instances of the same hemorrhages. Experimental results demonstrate that the YOLO-HV network achieved a F1 (F1_score) of 93.0 % and a MIoU (Mean Intersection over Union) of 87.1 %. Meanwhile, the model size has been reduced to 4.2 MB, surpassing other mainstream segmentation networks. Furthermore, the precision of volume measurement reached 93.7 %.

测量脑出血的体积对临床诊断和治疗至关重要。它可以帮助医生评估出血的严重程度，指导治疗决策，提高患者的存活率和生活质量。然而，由于出血的不规则性和流动性，现有的方法很难分割和测量不同的出血情况。本文介绍了基于 YOLOv8n-seg 的高效脑出血分割网络 YOLO-HV，该网络专为脑出血的体积测量而设计。为了加强对不规则出血区域空间特征信息的提取，该网络的骨干网中集成了协调注意（CoordAttention）机制。针对轻量级模型在大规模数据训练中的局限性，在颈部组件中引入了 GDConv（幽灵动态卷积）模块，以取代原有的 C2f 模块。原来的检测头被 LGND（轻量级组归一化检测头）取代，从而提高了网络的定位和分类性能，同时还降低了计算成本。在空间层面使用了联合查找功能，以匹配相同出血的跨层实例。实验结果表明，YOLO-HV 网络的 F1（F1_score）达到了 93.0%，MIoU（Mean Intersection over Union）达到了 87.1%。同时，模型大小减少到 4.2 MB，超过了其他主流分割网络。此外，体积测量的精确度达到了 93.7%。

{"title":"YOLO-HV: A fast YOLOv8-based method for measuring hemorrhage volumes","authors":"Haoran Wang, Guohui Wang, Yongliang Li, Kairong Zhang","doi":"10.1016/j.bspc.2024.107131","DOIUrl":"10.1016/j.bspc.2024.107131","url":null,"abstract":"<div><div>Measuring the volume of a cerebral hemorrhage is crucial for clinical diagnosis and treatment. It helps doctors assess the severity of the bleeding, guide treatment decisions, and improve patient survival rates and quality of life. However, due to the irregularity and fluid nature of the hemorrhages, existing methods struggle to segment and measure different hemorrhage instances. This paper introduces an efficient cerebral hemorrhage segmentation network, YOLO-HV, based on YOLOv8n-seg, designed for volumetric measurement of cerebral hemorrhages. To enhance the extraction of spatial feature information from irregular hemorrhagic areas, A CoordAttention mechanism is integrated into the backbone of the network. Addressing the limitations of lightweight models in training with large-scale data, a GDConv (Ghost Dynamic Convolution) module is introduced in the Neck component to replace the original C2f module. The original detection head is replaced with LGND (Lightweight Group Normalized Detection Head), enhancing positioning and classification performance of the network while additionally reducing computational costs. A Union-Find is used on a spatial level to match cross-layer instances of the same hemorrhages. Experimental results demonstrate that the YOLO-HV network achieved a F1 (F1_score) of 93.0 % and a MIoU (Mean Intersection over Union) of 87.1 %. Meanwhile, the model size has been reduced to 4.2 MB, surpassing other mainstream segmentation networks. Furthermore, the precision of volume measurement reached 93.7 %.</div></div>","PeriodicalId":55362,"journal":{"name":"Biomedical Signal Processing and Control","volume":"100 ","pages":"Article 107131"},"PeriodicalIF":4.9,"publicationDate":"2024-11-11","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142652846","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Enhancing plantar pressure distribution reconstruction with conditional generative adversarial networks from multi-region foot pressure sensing 利用多区域足底压力传感条件生成对抗网络加强足底压力分布重建

IF 4.9 2区医学 Q1 ENGINEERING, BIOMEDICAL

Biomedical Signal Processing and Control

Pub Date : 2024-11-09 DOI: 10.1016/j.bspc.2024.107187

Hsiao-Lung Chan , Jing-Rong Liang , Ya-Ju Chang , Rou-Shayn Chen , Cheng-Chung Kuo , Wen-Yen Hsu , Meng-Tsan Tsai

Estimating foot pressure distribution and the center of pressure (COP) using a sparse sensor topology offers cost-effective benefits. While deep learning neural networks improve the prediction of information in areas with incomplete sensing, there are still gaps in foot pressure recordings due to limited sensor coverage in certain plantar regions. To address this, we used eleven larger sensors to increase coverage across critical foot areas, including the big toe, little toe, medial, middle, and lateral metatarsus, as well as the medial and lateral arches, foreheels, and heels. These regions are commonly used to study the effects of muscle fatigue during walking and jogging, as well as to predict ground reaction forces during walking. We employed a conditional generative adversarial network (GAN) to reconstruct high-resolution foot pressure distributions from the data collected by these sensors. This method operates on individual samples, eliminating the need for gait cycle segmentation and normalization. Compared to ground truth data from a 99-sensor array, the GAN approach significantly improved COP estimation over direct computation from the eleven sensors. The highest accuracy was achieved during level walking, with reduced performance during jogging and stair walking. In conclusion, the conditional GAN effectively reconstructed foot pressure distributions, and future research should explore reallocating sensor topology to improve resolution and coverage while balancing simplified instrumentation with improved plantar pressure distribution reconstruction.

利用稀疏传感器拓扑结构估算脚压分布和压力中心（COP）具有成本效益。虽然深度学习神经网络改善了不完全传感区域的信息预测，但由于某些足底区域的传感器覆盖范围有限，足底压力记录仍存在缺口。为了解决这个问题，我们使用了 11 个更大的传感器来增加关键足部区域的覆盖范围，包括大脚趾、小脚趾、跖骨内侧、中段和外侧，以及足弓内侧和外侧、前足跟和后足跟。这些区域通常用于研究行走和慢跑时肌肉疲劳的影响，以及预测行走时的地面反作用力。我们采用条件生成式对抗网络 (GAN) 从这些传感器收集的数据中重建高分辨率脚压分布。这种方法在单个样本上运行，无需步态周期分割和归一化。与来自 99 个传感器阵列的地面实况数据相比，GAN 方法显著提高了 COP 估值，而不是直接计算 11 个传感器的结果。平地行走时的准确率最高，而慢跑和阶梯行走时的准确率较低。总之，条件 GAN 有效地重建了足底压力分布，未来的研究应探索重新分配传感器拓扑结构，以提高分辨率和覆盖范围，同时在简化仪器和改进足底压力分布重建之间取得平衡。

{"title":"Enhancing plantar pressure distribution reconstruction with conditional generative adversarial networks from multi-region foot pressure sensing","authors":"Hsiao-Lung Chan , Jing-Rong Liang , Ya-Ju Chang , Rou-Shayn Chen , Cheng-Chung Kuo , Wen-Yen Hsu , Meng-Tsan Tsai","doi":"10.1016/j.bspc.2024.107187","DOIUrl":"10.1016/j.bspc.2024.107187","url":null,"abstract":"<div><div>Estimating foot pressure distribution and the center of pressure (COP) using a sparse sensor topology offers cost-effective benefits. While deep learning neural networks improve the prediction of information in areas with incomplete sensing, there are still gaps in foot pressure recordings due to limited sensor coverage in certain plantar regions. To address this, we used eleven larger sensors to increase coverage across critical foot areas, including the big toe, little toe, medial, middle, and lateral metatarsus, as well as the medial and lateral arches, foreheels, and heels. These regions are commonly used to study the effects of muscle fatigue during walking and jogging, as well as to predict ground reaction forces during walking. We employed a conditional generative adversarial network (GAN) to reconstruct high-resolution foot pressure distributions from the data collected by these sensors. This method operates on individual samples, eliminating the need for gait cycle segmentation and normalization. Compared to ground truth data from a 99-sensor array, the GAN approach significantly improved COP estimation over direct computation from the eleven sensors. The highest accuracy was achieved during level walking, with reduced performance during jogging and stair walking. In conclusion, the conditional GAN effectively reconstructed foot pressure distributions, and future research should explore reallocating sensor topology to improve resolution and coverage while balancing simplified instrumentation with improved plantar pressure distribution reconstruction.</div></div>","PeriodicalId":55362,"journal":{"name":"Biomedical Signal Processing and Control","volume":"100 ","pages":"Article 107187"},"PeriodicalIF":4.9,"publicationDate":"2024-11-09","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142658177","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

An efficient IoT enabled heart disease prediction model using Finch hunt optimization modified BiLSTM classifier 利用芬奇狩猎优化改进的 BiLSTM 分类器建立高效的物联网心脏病预测模型

IF 4.9 2区医学 Q1 ENGINEERING, BIOMEDICAL

Biomedical Signal Processing and Control

Pub Date : 2024-11-09 DOI: 10.1016/j.bspc.2024.107170

Yogesh Suresh Chichani , Smita L. Kasar

Prediction of Cardiovascular disease (CVD) with a more accurate and timely diagnosis is crucial to ensure accurate classification, which assists medical professionals in providing appropriate treatment to the patient. Recently, healthcare organizations have begun utilizing Internet of Things (IoT) technology to gather sensor information for the purpose of diagnosing and forecasting heart disease. Cloud computing solutions have been utilized to manage the vast amount of data created by IoT devices in the medical profession, which amounts to an enormous number. Heart disease prediction is a challenging undertaking that demands both sophisticated knowledge and expertise. Although a lot of study has been done on the diagnosis of heart disease, the results are not very accurate. Further protecting the data from numerous general privacy concerns is a complex process. To address these limitations, this research utilizes the Finch hunt optimization modified BiLSTM classifier (FHO modified BiLSTM) to develop an IoT enabled Heart disease prediction model. Further, the incorporation of the smart IoT-based framework assists in monitoring heart disease patients and provides effective, timely, and quality healthcare services. Additionally, to improve mobility, privacy, security, low latency, and bandwidth, the biomedical data are stored in a cloud server that is equipped with a decentralized blockchain. The proposed approach exploits the Bi-LSTM model to improve the prediction abilities and extract intricate temporal patterns from patient data by combining predictive modeling. Specifically, the FHO integrates the characteristics of honey badger and sparrow to find the optimal solution for tuning the hyperparameters in the modified BiLSTM which in turn enhances the prediction accuracy. For analyzing the performance of the proposed method the CACHET-CADB dataset with 1602 samples is utilized. The experimental results demonstrates that the proposed FHO-modified Bi-LSTM attains the values of 95.17%, 96.52%, 93.86%, and 97.24% for F1-score, precision, recall, and accuracy respectively at 80% of training which exceeded the other existing techniques.

对心血管疾病（CVD）进行更准确、更及时的诊断预测对于确保准确分类至关重要，这有助于医疗专业人员为患者提供适当的治疗。最近，医疗机构开始利用物联网（IoT）技术收集传感器信息，用于诊断和预测心脏病。云计算解决方案已被用于管理医疗行业物联网设备产生的大量数据，这些数据数量庞大。心脏病预测是一项极具挑战性的工作，需要复杂的知识和专业技能。虽然对心脏病的诊断进行了大量研究，但结果并不十分准确。此外，保护数据免受众多隐私问题的影响也是一个复杂的过程。为了解决这些局限性，本研究利用芬奇狩猎优化改进型 BiLSTM 分类器（FHO 改进型 BiLSTM）开发了一个支持物联网的心脏病预测模型。此外，基于物联网的智能框架有助于监测心脏病患者，并提供有效、及时和优质的医疗保健服务。此外，为了提高移动性、隐私性、安全性、低延迟和带宽，生物医学数据被存储在配备了分散式区块链的云服务器中。所提出的方法利用 Bi-LSTM 模型提高预测能力，并通过结合预测建模从患者数据中提取复杂的时间模式。具体来说，FHO 综合了蜜獾和麻雀的特点，为调整修正后的 BiLSTM 中的超参数找到了最优解，从而提高了预测准确性。为了分析拟议方法的性能，我们使用了包含 1602 个样本的 CACHET-CADB 数据集。实验结果表明，在 80% 的训练时间内，所提出的 FHO 修正 Bi-LSTM 的 F1 分数、精确度、召回率和准确率分别达到了 95.17%、96.52%、93.86% 和 97.24%，超过了其他现有技术。

{"title":"An efficient IoT enabled heart disease prediction model using Finch hunt optimization modified BiLSTM classifier","authors":"Yogesh Suresh Chichani , Smita L. Kasar","doi":"10.1016/j.bspc.2024.107170","DOIUrl":"10.1016/j.bspc.2024.107170","url":null,"abstract":"<div><div>Prediction of Cardiovascular disease (CVD) with a more accurate and timely diagnosis is crucial to ensure accurate classification, which assists medical professionals in providing appropriate treatment to the patient. Recently, healthcare organizations have begun utilizing Internet of Things (IoT) technology to gather sensor information for the purpose of diagnosing and forecasting heart disease. Cloud computing solutions have been utilized to manage the vast amount of data created by IoT devices in the medical profession, which amounts to an enormous number. Heart disease prediction is a challenging undertaking that demands both sophisticated knowledge and expertise. Although a lot of study has been done on the diagnosis of heart disease, the results are not very accurate.Further protecting the data from numerous general privacy concerns is a complex process. To address these limitations, this research utilizes the Finch hunt optimization modified BiLSTM classifier (FHO modified BiLSTM) to develop an IoT enabled Heart disease prediction model. Further, the incorporation of the smart IoT-based framework assists in monitoring heart disease patients and provides effective, timely, and quality healthcare services. Additionally, to improve mobility, privacy, security, low latency, and bandwidth, the biomedical data are stored in a cloud server that is equipped with adecentralizedblockchain. The proposed approach exploits the Bi-LSTM model to improve the prediction abilities and extract intricate temporal patterns from patient data by combining predictive modeling. Specifically, the FHO integrates the characteristics of honey badger and sparrow to find the optimal solution for tuning the hyperparameters in the modified BiLSTM which in turn enhances the prediction accuracy. For analyzing the performance of the proposed method the CACHET-CADB dataset with 1602 samples is utilized. The experimental results demonstrates that the proposed FHO-modified Bi-LSTM attains the values of 95.17%, 96.52%, 93.86%, and 97.24% for F1-score, precision, recall, and accuracy respectively at 80% of training which exceeded the other existing techniques.</div></div>","PeriodicalId":55362,"journal":{"name":"Biomedical Signal Processing and Control","volume":"100 ","pages":"Article 107170"},"PeriodicalIF":4.9,"publicationDate":"2024-11-09","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142658295","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

UNeSt: A fast segmentation network for colorectal polyps based on MLP and deep separable convolution UNeSt：基于 MLP 和深度可分离卷积的结直肠息肉快速分割网络

IF 4.9 2区医学 Q1 ENGINEERING, BIOMEDICAL

Biomedical Signal Processing and Control

Pub Date : 2024-11-09 DOI: 10.1016/j.bspc.2024.107165

Jian Li , Peng Ding , Fengwu Lin , Zhaomin Chen , Ali Asghar Heidari , Huiling Chen

In medical image segmentation, conventional methodologies based on the UNet method often focus on improving the network performance but overlook the parameters and computational complexity. Due to the limitation of computing resources, these methods can hardly be applied in the landscape of point-of-care (PoC) applications. This study presents UNeSt, a rapid segmentation network tailored for colorectal polyps. The architectural foundation of UNeSt hinges upon the synergistic integration of the depth separable convolutional layer (DSC) and multilayer perceptron (MLP). UNeSt achieves an innovative fusion of these components, resulting in a substantial reduction in model parameters and computational complexity, which is concomitant with a remarkable enhancement in inference speed. Specifically, UNeSt incorporates the convolutional block attention module (CBAM) within the convolutional encoder to extract channel and spatial information proficiently. Furthermore, we introduce an attention mechanism to address the positional information discrepancies introduced in the MLP stage. This comprehensive approach contributes significantly to the augmentation of accuracy in colorectal polyp segmentation. Finally, UNeSt employs skip connections between various levels of encoders and decoders, thereby mitigating information loss problems. In the context of this investigation, UNeSt underwent rigorous evaluation using a demanding polyp segmentation dataset. Relative to UNeXt, a widely employed and exceedingly lightweight network model, the proposed model in this study exhibits a noteworthy reduction with 1.6x fewer parameters, a 2.5x decrease in computational complexity (measured in GFLOPs), and a 1.9x acceleration in inference speed.

在医学影像分割中，基于 UNet 方法的传统方法通常侧重于提高网络性能，但忽略了参数和计算复杂性。由于计算资源的限制，这些方法很难应用到医疗点（PoC）的应用中。本研究介绍了专为大肠息肉定制的快速分割网络 UNeSt。UNeSt 的架构基础取决于深度可分离卷积层（DSC）和多层感知器（MLP）的协同整合。UNeSt 实现了这些组件的创新性融合，从而大幅降低了模型参数和计算复杂度，同时显著提高了推理速度。具体来说，UNeSt 在卷积编码器中加入了卷积块注意模块（CBAM），以熟练提取信道和空间信息。此外，我们还引入了注意力机制，以解决 MLP 阶段引入的位置信息差异问题。这种综合方法大大提高了大肠息肉分割的准确性。最后，UNeSt 在各级编码器和解码器之间采用了跳转连接，从而减轻了信息丢失问题。在本次调查中，UNeSt 使用要求严格的息肉分割数据集进行了严格评估。与 UNeXt（一种广泛使用的超轻量级网络模型）相比，本研究中提出的模型参数减少了 1.6 倍，计算复杂度降低了 2.5 倍（以 GFLOPs 为单位），推理速度加快了 1.9 倍，这些都是值得注意的。

{"title":"UNeSt: A fast segmentation network for colorectal polyps based on MLP and deep separable convolution","authors":"Jian Li , Peng Ding , Fengwu Lin , Zhaomin Chen , Ali Asghar Heidari , Huiling Chen","doi":"10.1016/j.bspc.2024.107165","DOIUrl":"10.1016/j.bspc.2024.107165","url":null,"abstract":"<div><div>In medical image segmentation, conventional methodologies based on the UNet method often focus on improving the network performance but overlook the parameters and computational complexity. Due to the limitation of computing resources, these methods can hardly be applied in the landscape of point-of-care (PoC) applications. This study presents UNeSt, a rapid segmentation network tailored for colorectal polyps. The architectural foundation of UNeSt hinges upon the synergistic integration of the depth separable convolutional layer (DSC) and multilayer perceptron (MLP). UNeSt achieves an innovative fusion of these components, resulting in a substantial reduction in model parameters and computational complexity, which is concomitant with a remarkable enhancement in inference speed. Specifically, UNeSt incorporates the convolutional block attention module (CBAM) within the convolutional encoder to extract channel and spatial information proficiently. Furthermore, we introduce an attention mechanism to address the positional information discrepancies introduced in the MLP stage. This comprehensive approach contributes significantly to the augmentation of accuracy in colorectal polyp segmentation. Finally, UNeSt employs skip connections between various levels of encoders and decoders, thereby mitigating information loss problems. In the context of this investigation, UNeSt underwent rigorous evaluation using a demanding polyp segmentation dataset. Relative to UNeXt, a widely employed and exceedingly lightweight network model, the proposed model in this study exhibits a noteworthy reduction with 1.6x fewer parameters, a 2.5x decrease in computational complexity (measured in GFLOPs), and a 1.9x acceleration in inference speed.</div></div>","PeriodicalId":55362,"journal":{"name":"Biomedical Signal Processing and Control","volume":"100 ","pages":"Article 107165"},"PeriodicalIF":4.9,"publicationDate":"2024-11-09","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142658176","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

PU-CDM: A pyramid UNet based conditional diffusion model for sparse-view reconstruction in EPRI PU-CDM：用于 EPRI 稀疏视图重建的基于金字塔 UNet 的条件扩散模型

IF 4.9 2区医学 Q1 ENGINEERING, BIOMEDICAL

Biomedical Signal Processing and Control

Pub Date : 2024-11-09 DOI: 10.1016/j.bspc.2024.107182

Peng Liu , Yanjun Zhang , Yarui Xi , Chenyun Fang , Zhiwei Qiao

Sparse-view reconstruction in electron paramagnetic resonance imaging (EPRI) aims to reduce scanning times, which is critical for tumor oxygen imaging, yet is often plagued by streak artifacts in filtered back-projection (FBP) reconstructions. To address this, we propose a pyramid UNet based conditional diffusion model (PU-CDM) to suppress these streak artifacts in EPRI images. PU-CDM uniquely introduces pyramid pooling and aggregation into the UNet architecture of the conditional diffusion model, while incorporating two advanced mechanisms—dense convolutions and self-attention—into the input module. By significantly improving the accuracy of the Gaussian noise prediction network in the conditional diffusion model, PU-CDM achieves superior performance in sparse-view reconstruction, generating high-quality images with only 5 sampling steps. Experimental results, both qualitative and quantitative, show that the images reconstructed by PU-CDM outperform those reconstructed by some existing representative deep learning models in terms of artifact removal and structural fidelity. PU-CDM can achieve accurate sparse-view reconstruction in EPRI, thus promoting EPRI towards fast scanning. In addition, PU-CDM can also be used for fast magnetic resonance imaging (MRI), low-dose computed tomography (LDCT) reconstruction, and natural image processing.

电子顺磁共振成像（EPRI）中的稀疏视图重建旨在缩短扫描时间，这对肿瘤氧成像至关重要，但在滤波后投影（FBP）重建中经常会出现条纹伪影。为此，我们提出了一种基于金字塔 UNet 的条件扩散模型（PU-CDM），以抑制 EPRI 图像中的条纹伪影。PU-CDM 在条件扩散模型的 UNet 架构中独特地引入了金字塔池化和聚合，同时在输入模块中加入了两种先进的机制--密集卷积和自我注意。通过大幅提高条件扩散模型中高斯噪声预测网络的准确性，PU-CDM 在稀疏视图重建方面取得了卓越的性能，只需 5 个采样步骤就能生成高质量的图像。定性和定量实验结果表明，PU-CDM 重建的图像在去除伪影和结构保真度方面优于现有的一些代表性深度学习模型重建的图像。PU-CDM 可以在 EPRI 中实现精确的稀疏视图重建，从而促进 EPRI 走向快速扫描。此外，PU-CDM 还可用于快速磁共振成像（MRI）、低剂量计算机断层扫描（LDCT）重建和自然图像处理。

{"title":"PU-CDM: A pyramid UNet based conditional diffusion model for sparse-view reconstruction in EPRI","authors":"Peng Liu , Yanjun Zhang , Yarui Xi , Chenyun Fang , Zhiwei Qiao","doi":"10.1016/j.bspc.2024.107182","DOIUrl":"10.1016/j.bspc.2024.107182","url":null,"abstract":"<div><div>Sparse-view reconstruction in electron paramagnetic resonance imaging (EPRI) aims to reduce scanning times, which is critical for tumor oxygen imaging, yet is often plagued by streak artifacts in filtered back-projection (FBP) reconstructions. To address this, we propose a pyramid UNet based conditional diffusion model (PU-CDM) to suppress these streak artifacts in EPRI images. PU-CDM uniquely introduces pyramid pooling and aggregation into the UNet architecture of the conditional diffusion model, while incorporating two advanced mechanisms—dense convolutions and self-attention—into the input module. By significantly improving the accuracy of the Gaussian noise prediction network in the conditional diffusion model, PU-CDM achieves superior performance in sparse-view reconstruction, generating high-quality images with only 5 sampling steps. Experimental results, both qualitative and quantitative, show that the images reconstructed by PU-CDM outperform those reconstructed by some existing representative deep learning models in terms of artifact removal and structural fidelity. PU-CDM can achieve accurate sparse-view reconstruction in EPRI, thus promoting EPRI towards fast scanning. In addition, PU-CDM can also be used for fast magnetic resonance imaging (MRI), low-dose computed tomography (LDCT) reconstruction, and natural image processing.</div></div>","PeriodicalId":55362,"journal":{"name":"Biomedical Signal Processing and Control","volume":"100 ","pages":"Article 107182"},"PeriodicalIF":4.9,"publicationDate":"2024-11-09","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142657558","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0