Pub Date : 2025-03-06DOI: 10.1016/j.bspc.2025.107747
Mennatullah Mahmoud , Mohammad Mansour , Hisham M. Elrefai , Amira J. Hamed , Essam A. Rashed
The intricate network of retinal blood vessels serves as a sensitive window into systemic health, offering valuable insights into diseases like diabetic retinopathy. However, unraveling these insights poses challenges due to limitations of traditional visible light fundus photography. Infrared imaging emerges as a transformative tool, enabling deeper tissue penetration and enhanced visualization of the retinal vasculature. Yet, unlocking its full potential hinges on accurate and reliable segmentation of retinal arteries and veins within IR images. This study explores different ways to improve the accurate mapping of blood vessels in the eye using deep learning architectures. We used a special dataset captured with advanced technology to train and test three different models. This study amplifies the dataset’s adaptability, facilitating the training of U-Net, Residual U-Net, and Attention U-Net models. Among these models, the Attention Residual U-Net demonstrated superior segmentation performance, achieving an accuracy of 96.03%, a dice coefficient of 0.882, and a recall of 0.895 after post-processing. This research opens up possibilities for further improvements in eye-related healthcare.
{"title":"Enhanced retinal arteries and veins segmentation through deep learning with conditional random fields","authors":"Mennatullah Mahmoud , Mohammad Mansour , Hisham M. Elrefai , Amira J. Hamed , Essam A. Rashed","doi":"10.1016/j.bspc.2025.107747","DOIUrl":"10.1016/j.bspc.2025.107747","url":null,"abstract":"<div><div>The intricate network of retinal blood vessels serves as a sensitive window into systemic health, offering valuable insights into diseases like diabetic retinopathy. However, unraveling these insights poses challenges due to limitations of traditional visible light fundus photography. Infrared imaging emerges as a transformative tool, enabling deeper tissue penetration and enhanced visualization of the retinal vasculature. Yet, unlocking its full potential hinges on accurate and reliable segmentation of retinal arteries and veins within IR images. This study explores different ways to improve the accurate mapping of blood vessels in the eye using deep learning architectures. We used a special dataset captured with advanced technology to train and test three different models. This study amplifies the dataset’s adaptability, facilitating the training of U-Net, Residual U-Net, and Attention U-Net models. Among these models, the Attention Residual U-Net demonstrated superior segmentation performance, achieving an accuracy of 96.03%, a dice coefficient of 0.882, and a recall of 0.895 after post-processing. This research opens up possibilities for further improvements in eye-related healthcare.</div></div>","PeriodicalId":55362,"journal":{"name":"Biomedical Signal Processing and Control","volume":"106 ","pages":"Article 107747"},"PeriodicalIF":4.9,"publicationDate":"2025-03-06","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143550778","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}
Pub Date : 2025-03-06DOI: 10.1016/j.bspc.2025.107820
Naiwen Zhang , Jiale Chen , Jinting Ma , Xiaolong Guo , Jing Guo , Guo Dan
Given that blood pressure is a vital indicator of cardiovascular health, the domain of non-invasive continuous blood pressure monitoring has emerged as a hot area of interest in current research. However, existing studies in this field are often constrained by their limited capacity for clinical physiological interpretation and for reflecting cardiovascular and hemodynamic information. This gap hinders their effectiveness in elucidating the influence of cardiovascular system changes on blood pressure. This study aims to address these issues by using impedance cardiogram signal and the Windkessel (WK) model. First, we extracted features representing hemodynamic parameters from impedance cardiogram signal. Then, these features were utilized alongside the XGBoost algorithm to estimate parameters within the WK model. Finally, this model was used to model the subject’s cardiovascular system, thereby precisely simulating and estimating blood pressure changes. This methodology was validated using a public dataset, with results indicating that in resting scenario, the mean absolute error for systolic blood pressure and diastolic blood pressure were 4.72 mmHg and 3.72 mmHg, respectively. Furthermore, our findings identified a positive correlation between the WK model’s resistance parameter and blood pressure, and a negative correlation between its compliance parameter and blood pressure. These insights are instrumental in pioneering new avenues for continuous blood pressure estimation and in deepening our understanding of the physiological mechanisms of blood pressure changes.
{"title":"Combining impedance cardiography with Windkessel model for blood pressure estimation","authors":"Naiwen Zhang , Jiale Chen , Jinting Ma , Xiaolong Guo , Jing Guo , Guo Dan","doi":"10.1016/j.bspc.2025.107820","DOIUrl":"10.1016/j.bspc.2025.107820","url":null,"abstract":"<div><div>Given that blood pressure is a vital indicator of cardiovascular health, the domain of non-invasive continuous blood pressure monitoring has emerged as a hot area of interest in current research. However, existing studies in this field are often constrained by their limited capacity for clinical physiological interpretation and for reflecting cardiovascular and hemodynamic information. This gap hinders their effectiveness in elucidating the influence of cardiovascular system changes on blood pressure. This study aims to address these issues by using impedance cardiogram signal and the Windkessel (WK) model. First, we extracted features representing hemodynamic parameters from impedance cardiogram signal. Then, these features were utilized alongside the XGBoost algorithm to estimate parameters within the WK model. Finally, this model was used to model the subject’s cardiovascular system, thereby precisely simulating and estimating blood pressure changes. This methodology was validated using a public dataset, with results indicating that in resting scenario, the mean absolute error for systolic blood pressure and diastolic blood pressure were 4.72 mmHg and 3.72 mmHg, respectively. Furthermore, our findings identified a positive correlation between the WK model’s resistance parameter and blood pressure, and a negative correlation between its compliance parameter and blood pressure. These insights are instrumental in pioneering new avenues for continuous blood pressure estimation and in deepening our understanding of the physiological mechanisms of blood pressure changes.</div></div>","PeriodicalId":55362,"journal":{"name":"Biomedical Signal Processing and Control","volume":"106 ","pages":"Article 107820"},"PeriodicalIF":4.9,"publicationDate":"2025-03-06","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143551362","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}
Pub Date : 2025-03-05DOI: 10.1016/j.bspc.2025.107756
Shuai Guo , Yi Wang , Yuang Liu , Xin Zhang , Baoping Tang
Recent research has increasingly utilized deep learning (DL) to decode electroencephalogram (EEG) signals, enhancing the accuracy of motor imagery (MI) classification. While DL has improved MI decoding performance, challenges persist as distribution variances in MI-EEG data across different sessions. Additionally, the collection of EEG signals for MI tasks presents significant challenges, particularly in terms of time and economic costs. Data collection not only requires professional equipment and controlled environments but also demands the cooperation of a large number of participants to obtain sufficient sample size and diversity. To address these issues, this study proposes two methods to improve the decoding performance of MI-EEG signals based on an improved lightweight network. Firstly, a recombination-based data augmentation method leveraging channel knowledge is proposed to expand the training dataset and enhance model classification generalization, without the need for additional experiments to collect new data. Secondly, an improved domain adaptation network is introduced to align feature distributions between different domains, minimizing domain gaps. The proposed domain adaptation method aligns the target EEG domain with the corresponding class centers using pseudo-labeling. Extensive experiments are conducted using a cross-session training strategy on the BCIC IV 2a and BCIC IV 2b datasets. The results demonstrate that the proposed data augmentation method and improved domain adaptation method effectively enhance classification accuracy, providing a novel perspective for the practical application of MI-EEG.
{"title":"Improving cross-session motor imagery decoding performance with data augmentation and domain adaptation","authors":"Shuai Guo , Yi Wang , Yuang Liu , Xin Zhang , Baoping Tang","doi":"10.1016/j.bspc.2025.107756","DOIUrl":"10.1016/j.bspc.2025.107756","url":null,"abstract":"<div><div>Recent research has increasingly utilized deep learning (DL) to decode electroencephalogram (EEG) signals, enhancing the accuracy of motor imagery (MI) classification. While DL has improved MI decoding performance, challenges persist as distribution variances in MI-EEG data across different sessions. Additionally, the collection of EEG signals for MI tasks presents significant challenges, particularly in terms of time and economic costs. Data collection not only requires professional equipment and controlled environments but also demands the cooperation of a large number of participants to obtain sufficient sample size and diversity. To address these issues, this study proposes two methods to improve the decoding performance of MI-EEG signals based on an improved lightweight network. Firstly, a recombination-based data augmentation method leveraging channel knowledge is proposed to expand the training dataset and enhance model classification generalization, without the need for additional experiments to collect new data. Secondly, an improved domain adaptation network is introduced to align feature distributions between different domains, minimizing domain gaps. The proposed domain adaptation method aligns the target EEG domain with the corresponding class centers using pseudo-labeling. Extensive experiments are conducted using a cross-session training strategy on the BCIC IV 2a and BCIC IV 2b datasets. The results demonstrate that the proposed data augmentation method and improved domain adaptation method effectively enhance classification accuracy, providing a novel perspective for the practical application of MI-EEG.</div></div>","PeriodicalId":55362,"journal":{"name":"Biomedical Signal Processing and Control","volume":"106 ","pages":"Article 107756"},"PeriodicalIF":4.9,"publicationDate":"2025-03-05","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143550772","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}
Pub Date : 2025-03-05DOI: 10.1016/j.bspc.2025.107722
Jiapeng Li , Yijia Chen , Shijie Li , Lisheng Xu , Wei Qian , Shuai Tian , Lin Qi
Unsupervised domain adaptation (UDA) in medical image segmentation aims to transfer knowledge from a labeled source domain to an unlabeled target domain, especially when there are significant differences in data distribution across multi-modal medical images. Traditional UDA methods typically involve image translation and segmentation modules. However, during image translation, the anatomical structure of the generated images may vary, resulting in a mismatch of source domain labels and impacting subsequent segmentation. In addition, during image segmentation, although the Transformer architecture is used in UDA tasks due to its superior global context capture ability, it may not effectively facilitate knowledge transfer in UDA tasks due to lacking the adaptability of the self-attention mechanism in Transformers. To address these issues, we propose a generative UDA network with invariance mining, named TIG-UDA, for cross-modality multi-organ medical image segmentation, which includes an image style translation network (ISTN) and an invariance adaptation segmentation network (IASN). In ISTN, we not only introduce a structure preservation mechanism to guide image generation to achieve anatomical structure consistency, but also align the latent semantic features of source and target domain images to enhance the quality of the generated images. In IASN, we propose an invariance adaptation module that can extract the invariability weights of learned features in the attention mechanism of Transformer to compensate for the differences between source and target domains. Experimental results on two public cross-modality datasets (MS-CMR dataset and Abdomen dataset) show the promising segmentation performance of TIG-UDA compared with other state-of-the-art UDA methods.
{"title":"TIG-UDA: Generative unsupervised domain adaptation with transformer-embedded invariance for cross-modality medical image segmentation","authors":"Jiapeng Li , Yijia Chen , Shijie Li , Lisheng Xu , Wei Qian , Shuai Tian , Lin Qi","doi":"10.1016/j.bspc.2025.107722","DOIUrl":"10.1016/j.bspc.2025.107722","url":null,"abstract":"<div><div>Unsupervised domain adaptation (UDA) in medical image segmentation aims to transfer knowledge from a labeled source domain to an unlabeled target domain, especially when there are significant differences in data distribution across multi-modal medical images. Traditional UDA methods typically involve image translation and segmentation modules. However, during image translation, the anatomical structure of the generated images may vary, resulting in a mismatch of source domain labels and impacting subsequent segmentation. In addition, during image segmentation, although the Transformer architecture is used in UDA tasks due to its superior global context capture ability, it may not effectively facilitate knowledge transfer in UDA tasks due to lacking the adaptability of the self-attention mechanism in Transformers. To address these issues, we propose a generative UDA network with invariance mining, named TIG-UDA, for cross-modality multi-organ medical image segmentation, which includes an image style translation network (ISTN) and an invariance adaptation segmentation network (IASN). In ISTN, we not only introduce a structure preservation mechanism to guide image generation to achieve anatomical structure consistency, but also align the latent semantic features of source and target domain images to enhance the quality of the generated images. In IASN, we propose an invariance adaptation module that can extract the invariability weights of learned features in the attention mechanism of Transformer to compensate for the differences between source and target domains. Experimental results on two public cross-modality datasets (MS-CMR dataset and Abdomen dataset) show the promising segmentation performance of TIG-UDA compared with other state-of-the-art UDA methods.</div></div>","PeriodicalId":55362,"journal":{"name":"Biomedical Signal Processing and Control","volume":"106 ","pages":"Article 107722"},"PeriodicalIF":4.9,"publicationDate":"2025-03-05","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143551359","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}
Pub Date : 2025-03-05DOI: 10.1016/j.bspc.2025.107766
Yongcun Zhang , Jiajun Xu , Yina He , Shaozi Li , Zhiming Luo , Huangwei Lei
Tongue diagnosis in Traditional Chinese Medicine (TCM) is a crucial diagnostic method that can reflect an individual’s health status. Traditional methods for identifying tooth-marked tongues are subjective and inconsistent because they rely on practitioner experience. We propose a novel fully automated Weakly Supervised method using Vision transformer and Multiple instance learning (WSVM) for tongue extraction and tooth-marked tongue recognition. Our approach first accurately detects and extracts the tongue region from clinical images, removing any irrelevant background information. Then, we implement an end-to-end weakly supervised object detection method. We utilize Vision Transformer (ViT) to process tongue images in patches and employ multiple instance loss to identify tooth-marked regions with only image-level annotations. WSVM achieves high accuracy in tooth-marked tongue classification and tooth-marked tongue detection. Visualization experiments further demonstrate its effectiveness in pinpointing these regions. This automated approach enhances the objectivity and accuracy of tooth-marked tongue diagnosis. It provides significant clinical value by assisting TCM practitioners in making precise diagnoses and treatment recommendations. Code is available at https://github.com/yc-zh/WSVM.
{"title":"Weakly supervised object detection for automatic tooth-marked tongue recognition","authors":"Yongcun Zhang , Jiajun Xu , Yina He , Shaozi Li , Zhiming Luo , Huangwei Lei","doi":"10.1016/j.bspc.2025.107766","DOIUrl":"10.1016/j.bspc.2025.107766","url":null,"abstract":"<div><div>Tongue diagnosis in Traditional Chinese Medicine (TCM) is a crucial diagnostic method that can reflect an individual’s health status. Traditional methods for identifying tooth-marked tongues are subjective and inconsistent because they rely on practitioner experience. We propose a novel fully automated <strong>W</strong>eakly <strong>S</strong>upervised method using <strong>V</strong>ision transformer and <strong>M</strong>ultiple instance learning (<strong>WSVM</strong>) for tongue extraction and tooth-marked tongue recognition. Our approach first accurately detects and extracts the tongue region from clinical images, removing any irrelevant background information. Then, we implement an end-to-end weakly supervised object detection method. We utilize Vision Transformer (ViT) to process tongue images in patches and employ multiple instance loss to identify tooth-marked regions with only image-level annotations. WSVM achieves high accuracy in tooth-marked tongue classification and tooth-marked tongue detection. Visualization experiments further demonstrate its effectiveness in pinpointing these regions. This automated approach enhances the objectivity and accuracy of tooth-marked tongue diagnosis. It provides significant clinical value by assisting TCM practitioners in making precise diagnoses and treatment recommendations. Code is available at <span><span>https://github.com/yc-zh/WSVM</span><svg><path></path></svg></span>.</div></div>","PeriodicalId":55362,"journal":{"name":"Biomedical Signal Processing and Control","volume":"106 ","pages":"Article 107766"},"PeriodicalIF":4.9,"publicationDate":"2025-03-05","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143551373","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}
Pub Date : 2025-03-05DOI: 10.1016/j.bspc.2025.107735
Xiaoyan Kui , Bo Liu , Zanbo Sun , Qinsong Li , Min Zhang , Wei Liang , Beiji Zou
Learning-based methods for medical image translation have proven effective in addressing the challenge of obtaining complete multimodal medical images in clinical practice, particularly when patients are allergic to contrast agents or critical illnesses. Recently, diffusion models have exhibited superior performance in various image-generation tasks and are expected to replace generative adversarial networks (GANs) for medical image translation. However, existing methods suffer from unintuitive training objectives and complex network structures that curtail their efficacy in this domain. To address this gap, we propose a novel medical latent variational diffusion model (Med-LVDM) for efficient medical image translation. Firstly, we introduce a new parametric representation based on the variational diffusion model (VDM) and optimize the training objective to the weighted mean square error between the synthetic and target images, which is intuitive and has fewer model parameters. Then, we map the diffusion training and sampling process to the latent space, significantly reducing computational complexity to enhance the feasibility of clinical applications. Finally, to capture global information without focusing solely on local features, we utilize U-ViT as the backbone for Med-LVDM to effectively adapt to the latent space representing abstract information rather than concrete pixel-level information. Extensive qualitative and quantitative results in multi-contrast MRI and cross-modality MRI-CT datasets demonstrate our superiority in translation quality compared to state-of-the-art methods. In particular, Med-LVDM achieved its highest SSIM and PSNR of 92.37% and 26.23 dB on the BraTS2018 dataset, 90.18% and 24.55 dB on the IXI dataset, 91.61% and 25.52 dB on the MRI-CT dataset.
{"title":"Med-LVDM: Medical latent variational diffusion model for medical image translation","authors":"Xiaoyan Kui , Bo Liu , Zanbo Sun , Qinsong Li , Min Zhang , Wei Liang , Beiji Zou","doi":"10.1016/j.bspc.2025.107735","DOIUrl":"10.1016/j.bspc.2025.107735","url":null,"abstract":"<div><div>Learning-based methods for medical image translation have proven effective in addressing the challenge of obtaining complete multimodal medical images in clinical practice, particularly when patients are allergic to contrast agents or critical illnesses. Recently, diffusion models have exhibited superior performance in various image-generation tasks and are expected to replace generative adversarial networks (GANs) for medical image translation. However, existing methods suffer from unintuitive training objectives and complex network structures that curtail their efficacy in this domain. To address this gap, we propose a novel medical latent variational diffusion model (Med-LVDM) for efficient medical image translation. Firstly, we introduce a new parametric representation based on the variational diffusion model (VDM) and optimize the training objective to the weighted mean square error between the synthetic and target images, which is intuitive and has fewer model parameters. Then, we map the diffusion training and sampling process to the latent space, significantly reducing computational complexity to enhance the feasibility of clinical applications. Finally, to capture global information without focusing solely on local features, we utilize U-ViT as the backbone for Med-LVDM to effectively adapt to the latent space representing abstract information rather than concrete pixel-level information. Extensive qualitative and quantitative results in multi-contrast MRI and cross-modality MRI-CT datasets demonstrate our superiority in translation quality compared to state-of-the-art methods. In particular, Med-LVDM achieved its highest SSIM and PSNR of 92.37% and 26.23 dB on the BraTS2018 dataset, 90.18% and 24.55 dB on the IXI dataset, 91.61% and 25.52 dB on the MRI-CT dataset.</div></div>","PeriodicalId":55362,"journal":{"name":"Biomedical Signal Processing and Control","volume":"106 ","pages":"Article 107735"},"PeriodicalIF":4.9,"publicationDate":"2025-03-05","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143550770","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}
Pub Date : 2025-03-05DOI: 10.1016/j.bspc.2025.107739
Simin Lin , Longxin Deng , Ziwei Hu , Chengda Lin , Yongxin Mao , Yuntao Liu , Wei Li , Yue Yang , Rui Zhou , Yancheng Lai , Huang He , Tao Tan , Xinlin Zhang , Tong Tong , Na Ta , Rui Chen
Prostate cancer stands as the foremost cause of cancer-related mortality among men globally, with its incidence and mortality rates increasing alongside the aging population. The FOXA1 gene assumes a pivotal role in prostate cancer pathology, which is potential as a prognostic indicator and a potent therapeutic target across various stages of prostate cancer. Mutations in FOXA1 have been shown to amplify, supplant, and reconfigure Androgen Receptor function, thereby fostering prostate cancer proliferation. FOXA1 is the most common molecular mutation type in Asian prostate cancer patients, with a mutation rate reaching an astonishing 41 in China. It is also an important molecular subtype in Western populations. Currently, targeted therapy for FOXA1 is rapidly developing. Therefore, effective identification of FOXA1 mutations is of great clinical significance. Gene mutation detection is usually carried out by molecular biological methods, which is expensive and has a long-time cycle. To address this problem, we proposed a multi-modal deep learning network. This network can predict the FOXA1 gene mutation status using only Hematoxylin–Eosin (H&E) stained pathological images and clinical data. Following five-fold cross-validation, our model achieved an optimal Area Under the receiver operating characteristic Curve (AUC) of 0.808, with an average predicted AUC of 0.74, surpassing other comparative models. Furthermore, we observed a discernible correlation between FOXA1 mutations and ISUP grade.
{"title":"Predicting FOXA1 gene mutation status in prostate cancer through multi-modal deep learning","authors":"Simin Lin , Longxin Deng , Ziwei Hu , Chengda Lin , Yongxin Mao , Yuntao Liu , Wei Li , Yue Yang , Rui Zhou , Yancheng Lai , Huang He , Tao Tan , Xinlin Zhang , Tong Tong , Na Ta , Rui Chen","doi":"10.1016/j.bspc.2025.107739","DOIUrl":"10.1016/j.bspc.2025.107739","url":null,"abstract":"<div><div>Prostate cancer stands as the foremost cause of cancer-related mortality among men globally, with its incidence and mortality rates increasing alongside the aging population. The FOXA1 gene assumes a pivotal role in prostate cancer pathology, which is potential as a prognostic indicator and a potent therapeutic target across various stages of prostate cancer. Mutations in FOXA1 have been shown to amplify, supplant, and reconfigure Androgen Receptor function, thereby fostering prostate cancer proliferation. FOXA1 is the most common molecular mutation type in Asian prostate cancer patients, with a mutation rate reaching an astonishing 41<span><math><mtext>%</mtext></math></span> in China. It is also an important molecular subtype in Western populations. Currently, targeted therapy for FOXA1 is rapidly developing. Therefore, effective identification of FOXA1 mutations is of great clinical significance. Gene mutation detection is usually carried out by molecular biological methods, which is expensive and has a long-time cycle. To address this problem, we proposed a multi-modal deep learning network. This network can predict the FOXA1 gene mutation status using only Hematoxylin–Eosin (H&E) stained pathological images and clinical data. Following five-fold cross-validation, our model achieved an optimal Area Under the receiver operating characteristic Curve (AUC) of 0.808, with an average predicted AUC of 0.74, surpassing other comparative models. Furthermore, we observed a discernible correlation between FOXA1 mutations and ISUP grade.</div></div>","PeriodicalId":55362,"journal":{"name":"Biomedical Signal Processing and Control","volume":"106 ","pages":"Article 107739"},"PeriodicalIF":4.9,"publicationDate":"2025-03-05","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143551363","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}
Pub Date : 2025-03-05DOI: 10.1016/j.bspc.2025.107770
Walid Al-Haidri , Aynur Akhatov , Indira Usmanova , Farkhad Salimov , Mohammed Al-Habeeb , Kamil A. Il’yasov , Ekaterina A. Brui
Detecting and accurately quantifying vertebral body dimensions is essential for the assessment of vertebral deformities and fractures. Traditional manual measurement techniques are time-consuming and labor-intensive. This research presents a deep learning-assisted framework for the automatic segmentation, measurement, and deformity detection of lumbar vertebrae in MR images.
The segmentation procedure was implemented using a Mask-RCNN deep convolutional neural network. The framework also included post-processing of segmented vertebral body masks, detection of anatomical landmarks, and calculation of vertebral dimensions. The proposed landmark detection algorithm identified six key vertebral body landmarks (upper and lower anterior, posterior, and middle points) by analyzing pixel distributions in segmented vertebra masks. Using the detected landmarks, the anterior, posterior, and middle heights of each vertebra were calculated through predefined mathematical formulae. These dimensions were used as critical parameters to assess vertebral deformities (wedge and biconcave) by calculating specific height ratios, such as the anterior-posterior ratio and middle-posterior ratio. The dataset included T2-weighted MR images from 200 subjects, divided into subsets of 120 subjects for training, 50 for testing, and 30 for validation.
Mask-RCNN provided a high median Dice coefficient of 0.95 for vertebral body segmentation on a test subset. Median absolute errors of anterior, middle, and posterior heights measurements were 0.783, 0.856, and 0.785 mm, respectively. The algorithm allowed measuring wedge and biconcave vertebral deformities with median errors of 3.3 % and 4.4 %. The proposed framework is simpler and more universal than the existing methods and significantly automates the precise assessment of lumbar vertebrae deformities.
{"title":"Deep learning-assisted framework for automation of lumbar vertebral body segmentation, measurement, and deformity detection in MR images","authors":"Walid Al-Haidri , Aynur Akhatov , Indira Usmanova , Farkhad Salimov , Mohammed Al-Habeeb , Kamil A. Il’yasov , Ekaterina A. Brui","doi":"10.1016/j.bspc.2025.107770","DOIUrl":"10.1016/j.bspc.2025.107770","url":null,"abstract":"<div><div>Detecting and accurately quantifying vertebral body dimensions is essential for the assessment of vertebral deformities and fractures. Traditional manual measurement techniques are time-consuming and labor-intensive. This research presents a deep learning-assisted framework for the automatic segmentation, measurement, and deformity detection of lumbar vertebrae in MR images.</div><div>The segmentation procedure was implemented using a Mask-RCNN deep convolutional neural network. The framework also included post-processing of segmented vertebral body masks, detection of anatomical landmarks, and calculation of vertebral dimensions. The proposed landmark detection algorithm identified six key vertebral body landmarks (upper and lower anterior, posterior, and middle points) by analyzing pixel distributions in segmented vertebra masks. Using the detected landmarks, the anterior, posterior, and middle heights of each vertebra were calculated through predefined mathematical formulae. These dimensions were used as critical parameters to assess vertebral deformities (wedge and biconcave) by calculating specific height ratios, such as the anterior-posterior ratio and middle-posterior ratio. The dataset included T2-weighted MR images from 200 subjects, divided into subsets of 120 subjects for training, 50 for testing, and 30 for validation.</div><div>Mask-RCNN provided a high median Dice coefficient of 0.95 for vertebral body segmentation on a test subset. Median absolute errors of anterior, middle, and posterior heights measurements were 0.783, 0.856, and 0.785 mm, respectively. The algorithm allowed measuring wedge and biconcave vertebral deformities with median errors of 3.3 % and 4.4 %. The proposed framework is simpler and more universal than the existing methods and significantly automates the precise assessment of lumbar vertebrae deformities.</div></div>","PeriodicalId":55362,"journal":{"name":"Biomedical Signal Processing and Control","volume":"106 ","pages":"Article 107770"},"PeriodicalIF":4.9,"publicationDate":"2025-03-05","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143551374","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}
Owing to a large amount of multi-modal data in modern medical systems, such as medical images and reports, Medical Vision–Language Pre-training (Med-VLP) has demonstrated incredible achievements in coarse-grained downstream tasks (i.e., medical classification, retrieval, and visual question answering). However, the problem of transferring knowledge learned from Med-VLP to fine-grained multi-organ segmentation tasks has barely been investigated. Multi-organ segmentation is challenging mainly due to the lack of large-scale fully annotated datasets and the wide variation in the shape and size of the same organ between individuals with different diseases. In this paper, we propose a novel pre-training & fine-tuning framework for Multi-Organ Segmentation by harnessing Medical repOrt Supervision (MOSMOS). Specifically, we first introduce global contrastive learning to maximally align the medical image-report pairs in the pre-training stage. To remedy the granularity discrepancy, we further leverage multi-label recognition to implicitly learn the semantic correspondence between image pixels and organ tags. More importantly, our pre-trained models can be transferred to any segmentation model by introducing the pixel-tag attention maps. Different network settings, i.e., 2D U-Net and 3D UNETR, are utilized to validate the generalization. We have extensively evaluated our approach using different diseases and modalities on BTCV, AMOS, MMWHS, and BRATS datasets. Experimental results in various settings demonstrate the effectiveness of our framework. This framework can serve as the foundation to facilitate future research on automatic annotation tasks under the supervision of medical reports. Codes for our proposed method are available at https://github.com/paradisetww/MOSMOS.
{"title":"MOSMOS: Multi-organ segmentation facilitated by medical report supervision","authors":"Weiwei Tian , Xinyu Huang , Junlin Hou , Caiyue Ren , Longquan Jiang , Rui-Wei Zhao , Gang Jin , Yuejie Zhang , Daoying Geng","doi":"10.1016/j.bspc.2025.107743","DOIUrl":"10.1016/j.bspc.2025.107743","url":null,"abstract":"<div><div>Owing to a large amount of multi-modal data in modern medical systems, such as medical images and reports, Medical Vision–Language Pre-training (Med-VLP) has demonstrated incredible achievements in coarse-grained downstream tasks (i.e., medical classification, retrieval, and visual question answering). However, the problem of transferring knowledge learned from Med-VLP to fine-grained multi-organ segmentation tasks has barely been investigated. Multi-organ segmentation is challenging mainly due to the lack of large-scale fully annotated datasets and the wide variation in the shape and size of the same organ between individuals with different diseases. In this paper, we propose a novel pre-training & fine-tuning framework for Multi-Organ Segmentation by harnessing Medical repOrt Supervision (MOSMOS). Specifically, we first introduce global contrastive learning to maximally align the medical image-report pairs in the pre-training stage. To remedy the granularity discrepancy, we further leverage multi-label recognition to implicitly learn the semantic correspondence between image pixels and organ tags. More importantly, our pre-trained models can be transferred to any segmentation model by introducing the pixel-tag attention maps. Different network settings, i.e., 2D U-Net and 3D UNETR, are utilized to validate the generalization. We have extensively evaluated our approach using different diseases and modalities on BTCV, AMOS, MMWHS, and BRATS datasets. Experimental results in various settings demonstrate the effectiveness of our framework. This framework can serve as the foundation to facilitate future research on automatic annotation tasks under the supervision of medical reports. Codes for our proposed method are available at <span><span>https://github.com/paradisetww/MOSMOS</span><svg><path></path></svg></span>.</div></div>","PeriodicalId":55362,"journal":{"name":"Biomedical Signal Processing and Control","volume":"106 ","pages":"Article 107743"},"PeriodicalIF":4.9,"publicationDate":"2025-03-05","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143550777","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}
Pub Date : 2025-03-05DOI: 10.1016/j.bspc.2025.107734
Prithwish Sen, Pinki Roy
A wound is a rupture in the skin due to physical injury, trauma or surgery. A chronic wound is called an ulcer when it fails to heal properly. Such wounds may cause severe complication like necrosis if left untreated. An important factor in determining the severity stage of chronic wounds is the tissue state. Medical practitioners visually inspect wound tissues and determine the severity. Continuous monitoring of wounds is essential for effective treatment and prevention of further deterioration. However, due to a lack of skilled medical practitioners and limited access to medical facilities for remote patients, the risk of necrosis increases. In recent times, computer-aided diagnosis set up a new benchmark for assessing health-related conditions. Aiming for computer-aided remote wound analysis using computer vision and telemedicine, this paper proposes a lightweight end-to-end system for wound tissue analysis. The proposed work has two modules- (a) wound localization and patch extraction, and (b) wound tissue classification using the extracted patches. Three datasets are used in this work- AZH, Medetec and DFUC2020. The localization phase uses YOLOv8n architecture to detect and extract wound patches for classification. The tissue classification phase uses an ELU-enabled convolutional tokenizer embedded with compact convolutional transformers. The proposed localization algorithm achieves a mean average precision (mAP) of 95.2% and the tissue classifier achieves an accuracy of 97.14%. The end system performs better than traditional algorithms w.r.t. accuracy and computation cost in terms of FLOPS (8.10002 B).
{"title":"YOLO meets CCViT- A lightweight end-to-end system for wound tissue analysis","authors":"Prithwish Sen, Pinki Roy","doi":"10.1016/j.bspc.2025.107734","DOIUrl":"10.1016/j.bspc.2025.107734","url":null,"abstract":"<div><div>A wound is a rupture in the skin due to physical injury, trauma or surgery. A chronic wound is called an ulcer when it fails to heal properly. Such wounds may cause severe complication like necrosis if left untreated. An important factor in determining the severity stage of chronic wounds is the tissue state. Medical practitioners visually inspect wound tissues and determine the severity. Continuous monitoring of wounds is essential for effective treatment and prevention of further deterioration. However, due to a lack of skilled medical practitioners and limited access to medical facilities for remote patients, the risk of necrosis increases. In recent times, computer-aided diagnosis set up a new benchmark for assessing health-related conditions. Aiming for computer-aided remote wound analysis using computer vision and telemedicine, this paper proposes a lightweight end-to-end system for wound tissue analysis. The proposed work has two modules- (a) wound localization and patch extraction, and (b) wound tissue classification using the extracted patches. Three datasets are used in this work- AZH, Medetec and DFUC2020. The localization phase uses YOLOv8n architecture to detect and extract wound patches for classification. The tissue classification phase uses an ELU-enabled convolutional tokenizer embedded with compact convolutional transformers. The proposed localization algorithm achieves a mean average precision (mAP) of 95.2% and the tissue classifier achieves an accuracy of 97.14%. The end system performs better than traditional algorithms w.r.t. accuracy and computation cost in terms of FLOPS (8.10002 B).</div></div>","PeriodicalId":55362,"journal":{"name":"Biomedical Signal Processing and Control","volume":"106 ","pages":"Article 107734"},"PeriodicalIF":4.9,"publicationDate":"2025-03-05","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143550771","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}
扫码关注我们
求助内容：
应助结果提醒方式：
京公网安备 11010802042870号