Background: Accurate identification of bone metastases in lung cancer is essential for effective diagnosis and treatment. However, existing methods for detecting bone metastases face significant limitations, particularly in whole-body bone scans, due to low resolution, blurred boundaries, and the variability in lesion shapes and sizes, which challenge traditional convolutional neural networks. Purpose: To accurately isolate the metastasized lesions from whole-body bone scans, we propose a lesion-aware segmentation model using deep learning techniques. Methods: The proposed model integrates lesion boundary-guided strategies, multi-scale learning, and image shape guidance into an encoder-decoder architecture network. This approach significantly improves segmentation performance in low-resolution and blurred boundary conditions while effectively managing lesion shape variability and mitigating interference from the rectangular format of the images. Results: Experimental evaluations conducted on clinical data of 274 whole-body bone scans demonstrate that the proposed model achieves a 7.45% improvement in the Dice Similarity Coefficient and a 11.75% improvement in Recall compared to specialized segmentation models for whole-body bone scans, achieving significant improvements and balanced performance across key metrics. Conclusions: This model offers a more accurate and efficient solution for identifying bone metastases in lung cancer, alleviating the challenges of deep learning-based automated analysis of low-resolution, large-size medical images of whole-body bone scans. The code is available at https://github.com/carorange/segmentation Clinical and Impact: This lesion-aware deep learning model provides a robust, automated solution for identifying bone metastases in low-resolution, large-scale whole-body bone scans, enabling earlier and more accurate clinical decisions and potentially improving patient outcomes in lung cancer care.
{"title":"Integrating Non-Square Filter and Boundary Enhancement Into Encoder–Decoder Network for Lesion-Aware Segmentation of Large-Size Low-Resolution Bone Scintigrams","authors":"Ailing Xie;Qiang Lin;Xianwu Zeng;Yongchun Cao;Zhengxing Man;Caihong Liu;Xiaodi Huang","doi":"10.1109/JTEHM.2025.3605042","DOIUrl":"https://doi.org/10.1109/JTEHM.2025.3605042","url":null,"abstract":"Background: Accurate identification of bone metastases in lung cancer is essential for effective diagnosis and treatment. However, existing methods for detecting bone metastases face significant limitations, particularly in whole-body bone scans, due to low resolution, blurred boundaries, and the variability in lesion shapes and sizes, which challenge traditional convolutional neural networks. Purpose: To accurately isolate the metastasized lesions from whole-body bone scans, we propose a lesion-aware segmentation model using deep learning techniques. Methods: The proposed model integrates lesion boundary-guided strategies, multi-scale learning, and image shape guidance into an encoder-decoder architecture network. This approach significantly improves segmentation performance in low-resolution and blurred boundary conditions while effectively managing lesion shape variability and mitigating interference from the rectangular format of the images. Results: Experimental evaluations conducted on clinical data of 274 whole-body bone scans demonstrate that the proposed model achieves a 7.45% improvement in the Dice Similarity Coefficient and a 11.75% improvement in Recall compared to specialized segmentation models for whole-body bone scans, achieving significant improvements and balanced performance across key metrics. Conclusions: This model offers a more accurate and efficient solution for identifying bone metastases in lung cancer, alleviating the challenges of deep learning-based automated analysis of low-resolution, large-size medical images of whole-body bone scans. The code is available at <uri>https://github.com/carorange/segmentation</uri> Clinical and Impact: This lesion-aware deep learning model provides a robust, automated solution for identifying bone metastases in low-resolution, large-scale whole-body bone scans, enabling earlier and more accurate clinical decisions and potentially improving patient outcomes in lung cancer care.","PeriodicalId":54255,"journal":{"name":"IEEE Journal of Translational Engineering in Health and Medicine-Jtehm","volume":"13 ","pages":"421-436"},"PeriodicalIF":4.4,"publicationDate":"2025-09-02","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=11146776","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145011316","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-08-22DOI: 10.1109/JTEHM.2025.3601988
Ayman Anwar;Yassin Khalifa;Amanda S. Mahoney;Mehdy Dousty;James L. Coyle;Ervin Sejdic
Objective: Accurate tracking of anatomical landmarks during swallowing is critical for early diagnosis and treatment of dysphagia. Hyoid bone displacement plays a pivotal role in upper esophageal sphincter opening and airway protection, traditionally assessed via a videofluoroscopic swallow study (VFSS). However, VFSSs are subjective, expose patients to radiation, and are not universally accessible. High-resolution cervical auscultation (HRCA) offers a noninvasive alternative, utilizing acoustic and vibratory signals. Prior studies have validated HRCA’s efficacy in analyzing swallowing kinematics and correlating with hyoid bone displacement, typically employing transform domain characteristics and recurrent neural networks to achieve 50% overlap in predicted displacementsMethods: We introduce a transformer-based architecture for tracking hyoid bone displacement directly from raw HRCA signals, leveraging advanced temporal and spatial feature extraction methods using attention mechanism. The proposed pipeline preprocesses HRCA signals, segments individual swallows, and tracks the hyoid bone.Results: Our approach significantly improves upon existing methods, achieving over 70% relative overlap in predicted hyoid bone displacements across validation folds, surpassing state-of-the-art baseline models by a margin of at least 20%. Comprehensive statistical analysis confirms the robustness and accuracy of our predictions, demonstrating strong generalization capabilities on an independent dataset.Conclusion: This novel approach underscores the potential of transformer models in promoting noninvasive dysphagia assessment, offering a precise tracking of hyoid bone without VFSS images, and providing clinicians with insights about its movement trends, potentially aiding in clinical decision-making and bringing us one step closer to automated noninvasive swallowing assessment protocols. Clinical Impact– This study highlights the potential of automated hyoid bone tracking using HRCA signals to enhance dysphagia assessment by providing objective, noninvasive measurements that potentially support earlier detection and monitoring of swallowing impairments in both clinical and home healthcare settings, ultimately improving patient management and treatment outcomes.
{"title":"Videographic-Free Tracking of Hyoid Bone Displacement During Swallowing Using Accelerometer Signals and Transformers","authors":"Ayman Anwar;Yassin Khalifa;Amanda S. Mahoney;Mehdy Dousty;James L. Coyle;Ervin Sejdic","doi":"10.1109/JTEHM.2025.3601988","DOIUrl":"https://doi.org/10.1109/JTEHM.2025.3601988","url":null,"abstract":"Objective: Accurate tracking of anatomical landmarks during swallowing is critical for early diagnosis and treatment of dysphagia. Hyoid bone displacement plays a pivotal role in upper esophageal sphincter opening and airway protection, traditionally assessed via a videofluoroscopic swallow study (VFSS). However, VFSSs are subjective, expose patients to radiation, and are not universally accessible. High-resolution cervical auscultation (HRCA) offers a noninvasive alternative, utilizing acoustic and vibratory signals. Prior studies have validated HRCA’s efficacy in analyzing swallowing kinematics and correlating with hyoid bone displacement, typically employing transform domain characteristics and recurrent neural networks to achieve 50% overlap in predicted displacementsMethods: We introduce a transformer-based architecture for tracking hyoid bone displacement directly from raw HRCA signals, leveraging advanced temporal and spatial feature extraction methods using attention mechanism. The proposed pipeline preprocesses HRCA signals, segments individual swallows, and tracks the hyoid bone.Results: Our approach significantly improves upon existing methods, achieving over 70% relative overlap in predicted hyoid bone displacements across validation folds, surpassing state-of-the-art baseline models by a margin of at least 20%. Comprehensive statistical analysis confirms the robustness and accuracy of our predictions, demonstrating strong generalization capabilities on an independent dataset.Conclusion: This novel approach underscores the potential of transformer models in promoting noninvasive dysphagia assessment, offering a precise tracking of hyoid bone without VFSS images, and providing clinicians with insights about its movement trends, potentially aiding in clinical decision-making and bringing us one step closer to automated noninvasive swallowing assessment protocols. Clinical Impact– This study highlights the potential of automated hyoid bone tracking using HRCA signals to enhance dysphagia assessment by providing objective, noninvasive measurements that potentially support earlier detection and monitoring of swallowing impairments in both clinical and home healthcare settings, ultimately improving patient management and treatment outcomes.","PeriodicalId":54255,"journal":{"name":"IEEE Journal of Translational Engineering in Health and Medicine-Jtehm","volume":"13 ","pages":"402-412"},"PeriodicalIF":4.4,"publicationDate":"2025-08-22","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=11134417","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144918163","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-08-18DOI: 10.1109/JTEHM.2025.3600110
Mohsen Nabian;Louis Atallah
Objective: In the complex landscape of ICU operations, accurate discharge decisions are crucial yet challenging, as premature discharge risks readmission and mortality while prolonged stays consume resources and heighten infection risk. The objective of this work is to develop a deep learning-based Discharge Readiness Score (DRS) model using minimal clinical features to predict ICU discharge readiness, and to highlight its application in estimating excess ICU stays for resource optimization. Methods and procedures: We utilized nearly 1.8 million ICU patient-stays from 2007–2023 across 300 US hospitals in the Philips eICU database. Six readily available features (age, mean arterial pressure, systolic pressure, heart rate, respiratory rate, and Glasgow Coma Scale) were used as inputs. A 5-layer neural network predicted patient mortality within 48 hours post-ICU discharge as a proxy for discharge readiness. The model was trained on 80% of data, validated on 10%, and tested on 10% (approximately 180,000 patients). We applied the model hourly to estimate excess ICU stays, defining excess stay as the time patients remained at low risk but continued in ICU. Results: The model achieved an AUC of 0.93 on the test set. Performance was consistent across years, ethnicities, ICU types, and admission groups. Using the model, we found that about 22% of patients had excess ICU time, with a median of 16 hours. The analysis highlighted trends over time and across ICU types, providing insights into resource utilization. Conclusion: The DRS model effectively predicts ICU discharge readiness using minimal features and can estimate excess ICU stays, aiding resource optimization. Clinical Impact— The model offers a practical tool for ICU discharge planning and resource utilization analysis, potentially improving patient outcomes and ICU operations
{"title":"A Deep Learning Model for Predicting ICU Discharge Readiness and Estimating Excess ICU Stay Duration","authors":"Mohsen Nabian;Louis Atallah","doi":"10.1109/JTEHM.2025.3600110","DOIUrl":"https://doi.org/10.1109/JTEHM.2025.3600110","url":null,"abstract":"Objective: In the complex landscape of ICU operations, accurate discharge decisions are crucial yet challenging, as premature discharge risks readmission and mortality while prolonged stays consume resources and heighten infection risk. The objective of this work is to develop a deep learning-based Discharge Readiness Score (DRS) model using minimal clinical features to predict ICU discharge readiness, and to highlight its application in estimating excess ICU stays for resource optimization. Methods and procedures: We utilized nearly 1.8 million ICU patient-stays from 2007–2023 across 300 US hospitals in the Philips eICU database. Six readily available features (age, mean arterial pressure, systolic pressure, heart rate, respiratory rate, and Glasgow Coma Scale) were used as inputs. A 5-layer neural network predicted patient mortality within 48 hours post-ICU discharge as a proxy for discharge readiness. The model was trained on 80% of data, validated on 10%, and tested on 10% (approximately 180,000 patients). We applied the model hourly to estimate excess ICU stays, defining excess stay as the time patients remained at low risk but continued in ICU. Results: The model achieved an AUC of 0.93 on the test set. Performance was consistent across years, ethnicities, ICU types, and admission groups. Using the model, we found that about 22% of patients had excess ICU time, with a median of 16 hours. The analysis highlighted trends over time and across ICU types, providing insights into resource utilization. Conclusion: The DRS model effectively predicts ICU discharge readiness using minimal features and can estimate excess ICU stays, aiding resource optimization. Clinical Impact— The model offers a practical tool for ICU discharge planning and resource utilization analysis, potentially improving patient outcomes and ICU operations","PeriodicalId":54255,"journal":{"name":"IEEE Journal of Translational Engineering in Health and Medicine-Jtehm","volume":"13 ","pages":"413-420"},"PeriodicalIF":4.4,"publicationDate":"2025-08-18","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=11129058","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144998065","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-08-08DOI: 10.1109/JTEHM.2025.3597088
Rumaisa Abu Hasan;Tong Boon Tang;Muhamad Saiful Bahri Yusoff;Syed Saad Azhar Ali
Background: Mental resilience is an important indicator of our defence mechanism against mental illness. The assessment of mental resilience is conventionally done using psychological questionnaires but more recently, has been investigated using neuroimaging modalities such as the Magnetic Resonance Imaging and Positron Emission Tomography. While having high spatial resolution, these modalities might not be cost-effective and accessible to serve larger populations. This pilot trial investigates the performance of electroencephalography (EEG) based system to assess mental resilience under different mental conditions.Methods: A total of sixty-eight healthy adults took part in this trial. Three types of EEG features, namely spectra, functional connectivity (FC) and effective connectivity (EC) were extracted, and their correlation with a standard resilience assessment instrument – the Connor-Davidson Resilience Scale were evaluated at resting and task conditions using stepwise regression. The features with the best goodness of fit model were then used to classify individuals into a low and high mental resilience class.Results: The EC features using phase slope index achieved the highest adjusted $R^{2}$ and the lowest root mean square error, compared to the spectral and FC features. The SVM classifiers trained with the EC features were able to recognize low mental resilience with accuracy at least 66% depending on the mental condition. Fusion of SVM scores from the eyes-closed, eyes-open and task conditions improved the classification accuracy to more than 85%.Conclusion: The pilot trial reveals the EC as the most promising EEG feature type in assessing mental resilience due to its measure of causality in brain activity, and demonstrates that the fusion of decisions among different mental conditions can help improve the recognition of low mental resilience. Findings from this trial contribute to maturing an EEG-based resilience assessment system development for workplace settings. Clinical Impact—Direct assessment using brain imaging modalities such as EEG provides a cost-effective means to assess mental resilience. To our knowledge, this is the first effort for healthy subjects. With the identified neuromarkers, the proposed solution demonstrates the potential to fuse EEG features from different mental conditions to provide accurate mental resilience assessment in workplace settings.
{"title":"Electroencephalography-Based Recognition of Low Mental Resilience Using Multi-Condition Decision-Level Fusion Approach","authors":"Rumaisa Abu Hasan;Tong Boon Tang;Muhamad Saiful Bahri Yusoff;Syed Saad Azhar Ali","doi":"10.1109/JTEHM.2025.3597088","DOIUrl":"https://doi.org/10.1109/JTEHM.2025.3597088","url":null,"abstract":"Background: Mental resilience is an important indicator of our defence mechanism against mental illness. The assessment of mental resilience is conventionally done using psychological questionnaires but more recently, has been investigated using neuroimaging modalities such as the Magnetic Resonance Imaging and Positron Emission Tomography. While having high spatial resolution, these modalities might not be cost-effective and accessible to serve larger populations. This pilot trial investigates the performance of electroencephalography (EEG) based system to assess mental resilience under different mental conditions.Methods: A total of sixty-eight healthy adults took part in this trial. Three types of EEG features, namely spectra, functional connectivity (FC) and effective connectivity (EC) were extracted, and their correlation with a standard resilience assessment instrument – the Connor-Davidson Resilience Scale were evaluated at resting and task conditions using stepwise regression. The features with the best goodness of fit model were then used to classify individuals into a low and high mental resilience class.Results: The EC features using phase slope index achieved the highest adjusted <inline-formula> <tex-math>$R^{2}$ </tex-math></inline-formula> and the lowest root mean square error, compared to the spectral and FC features. The SVM classifiers trained with the EC features were able to recognize low mental resilience with accuracy at least 66% depending on the mental condition. Fusion of SVM scores from the eyes-closed, eyes-open and task conditions improved the classification accuracy to more than 85%.Conclusion: The pilot trial reveals the EC as the most promising EEG feature type in assessing mental resilience due to its measure of causality in brain activity, and demonstrates that the fusion of decisions among different mental conditions can help improve the recognition of low mental resilience. Findings from this trial contribute to maturing an EEG-based resilience assessment system development for workplace settings. Clinical Impact—Direct assessment using brain imaging modalities such as EEG provides a cost-effective means to assess mental resilience. To our knowledge, this is the first effort for healthy subjects. With the identified neuromarkers, the proposed solution demonstrates the potential to fuse EEG features from different mental conditions to provide accurate mental resilience assessment in workplace settings.","PeriodicalId":54255,"journal":{"name":"IEEE Journal of Translational Engineering in Health and Medicine-Jtehm","volume":"13 ","pages":"387-401"},"PeriodicalIF":4.4,"publicationDate":"2025-08-08","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=11121398","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144831642","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-08-07DOI: 10.1109/JTEHM.2025.3596561
Armin W. Pomeroy;Alexander Upfill-Brown;Brandon T. Peterson;Dean Chen;Joel Weisenburger;Alexandra Stavrakis;Hani Haider;Nelson F. SooHoo;Jonathan B. Hopkins;Tyler R. Clites
Objective: Total knee arthroplasty (TKA) is a common and highly successful treatment for knee osteoarthritis. Despite its success, some TKA implants still do not last the remaining lifetime of the patient, due in large part to aseptic loosening of the bone-implant interface, most commonly involving the tibial component. In this manuscript, we present a compliant tibial stem with the potential to increase the lifespan of TKA by accommodating rotation of the tibial tray about the tibia’s long axis without introducing an additional high-cycle-count wear surface. Our objective was to refine the design of this implant to support the loads and displacements associated with common activities of daily living (ADLs), and to validate performance of a physical prototype on the benchtop. Methods: We used finite element analysis to sweep a representative parameter space of reasonably-sized caged hinges, and then to refine the mechanism geometry in the context of in vivo knee joint loads. We fabricated a prototype of the refined mechanism, and evaluated performance of that physical prototype under ADL loads and displacements. Results: The refined mechanism supports walking loads and displacements with a safety factor of 1.47 on the target fatigue stress limit. The maximum reaction moment in the prototype was 1.22 Nm during emulated walking, which represents a reduction of approximately 80% from the in vivo reaction moment within a conventional TKA implant rotating to the same angle. Discussion/Conclusion: Our results demonstrate feasibility of a compliant tibial stem with the potential to decrease failure rates and increase longevity of TKA implants.
{"title":"Compliant Tibial Stem for Primary Total Knee Arthroplasty","authors":"Armin W. Pomeroy;Alexander Upfill-Brown;Brandon T. Peterson;Dean Chen;Joel Weisenburger;Alexandra Stavrakis;Hani Haider;Nelson F. SooHoo;Jonathan B. Hopkins;Tyler R. Clites","doi":"10.1109/JTEHM.2025.3596561","DOIUrl":"https://doi.org/10.1109/JTEHM.2025.3596561","url":null,"abstract":"Objective: Total knee arthroplasty (TKA) is a common and highly successful treatment for knee osteoarthritis. Despite its success, some TKA implants still do not last the remaining lifetime of the patient, due in large part to aseptic loosening of the bone-implant interface, most commonly involving the tibial component. In this manuscript, we present a compliant tibial stem with the potential to increase the lifespan of TKA by accommodating rotation of the tibial tray about the tibia’s long axis without introducing an additional high-cycle-count wear surface. Our objective was to refine the design of this implant to support the loads and displacements associated with common activities of daily living (ADLs), and to validate performance of a physical prototype on the benchtop. Methods: We used finite element analysis to sweep a representative parameter space of reasonably-sized caged hinges, and then to refine the mechanism geometry in the context of in vivo knee joint loads. We fabricated a prototype of the refined mechanism, and evaluated performance of that physical prototype under ADL loads and displacements. Results: The refined mechanism supports walking loads and displacements with a safety factor of 1.47 on the target fatigue stress limit. The maximum reaction moment in the prototype was 1.22 Nm during emulated walking, which represents a reduction of approximately 80% from the in vivo reaction moment within a conventional TKA implant rotating to the same angle. Discussion/Conclusion: Our results demonstrate feasibility of a compliant tibial stem with the potential to decrease failure rates and increase longevity of TKA implants.","PeriodicalId":54255,"journal":{"name":"IEEE Journal of Translational Engineering in Health and Medicine-Jtehm","volume":"13 ","pages":"376-386"},"PeriodicalIF":4.4,"publicationDate":"2025-08-07","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=11119545","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144831822","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Objectives: Assessing the bidirectional impacts of heart function impairment and sleep-disordered breathing remains underexplored. Thus, this study analyzed respiratory patterns from a wireless radar framework to explore their associations with echocardiographic (2D-echo) measurements. Methods: Background details, 2D-echo parameters, and biochemical data were collected from patients in a cardiology ward in northern Taiwan. Their radar-based respiratory patterns from the night before and the night of the 2D-echo were obtained, averaged, and used to derive indices such as the respiratory disturbance index (RDI) and periodic breathing (PB) cycle length, representing overall respiratory patterns. Next, retrieved data were grouped based on a 50% left ventricular ejection fraction (LVEF) threshold and analyzed using mean comparisons and regression models to explore relationships. Results: Patients with an LVEF of $le 50$ % demonstrated significantly reduced total sleep time, higher RDI, and longer PB cycles compared to those with LVEF >50%. Each 1-event/h increase in the RDI reduced the LVEF by 0.22% (95% confidence interval [CI]: −0.41% to −0.03%, p <0.05),> $le 50$ % from >50%. Subgroup analysis revealed that the PB cycle length was associated with elevated N-terminal-prohormone-brain-natriuretic-peptide (NT-proBNP) levels. Conclusions: This study demonstrates that a wireless radar framework combined with deep learning can effectively monitor respiratory patterns that are associated with cardiac function. Its contactless nature may support continuous cardiac function assessments. Clinical Impact: This study highlights the effectiveness of a wireless radar and deep learning framework for monitoring respiratory patterns that are associated with cardiac function (e.g., LVEF), underscoring its potential for long-term cardiac and sleep-disorder management.
{"title":"Evaluating Cardiac Impairment From Abnormal Respiratory Patterns: Insights From a Wireless Radar and Deep Learning Study","authors":"Chun-Chih Chiu;Wen-Te Liu;Jiunn-Horng Kang;Chun-Chao Chen;Yu-Hsuan Ho;Yu-Wen Huang;Zong-Lin Tsai;Rachel Chien;Ying-Ying Chen;Yen-Ling Chen;Nai-Wen Chang;Hung-Wen Lu;Kang-Yun Lee;Arnab Majumdar;Shu-Han Liao;Ju-Chi Liu;Cheng-Yu Tsai","doi":"10.1109/JTEHM.2025.3588523","DOIUrl":"https://doi.org/10.1109/JTEHM.2025.3588523","url":null,"abstract":"Objectives: Assessing the bidirectional impacts of heart function impairment and sleep-disordered breathing remains underexplored. Thus, this study analyzed respiratory patterns from a wireless radar framework to explore their associations with echocardiographic (2D-echo) measurements. Methods: Background details, 2D-echo parameters, and biochemical data were collected from patients in a cardiology ward in northern Taiwan. Their radar-based respiratory patterns from the night before and the night of the 2D-echo were obtained, averaged, and used to derive indices such as the respiratory disturbance index (RDI) and periodic breathing (PB) cycle length, representing overall respiratory patterns. Next, retrieved data were grouped based on a 50% left ventricular ejection fraction (LVEF) threshold and analyzed using mean comparisons and regression models to explore relationships. Results: Patients with an LVEF of <inline-formula> <tex-math>$le 50$ </tex-math></inline-formula>% demonstrated significantly reduced total sleep time, higher RDI, and longer PB cycles compared to those with LVEF >50%. Each 1-event/h increase in the RDI reduced the LVEF by 0.22% (95% confidence interval [CI]: −0.41% to −0.03%, p <0.05),> <tex-math>$le 50$ </tex-math></inline-formula>% from >50%. Subgroup analysis revealed that the PB cycle length was associated with elevated N-terminal-prohormone-brain-natriuretic-peptide (NT-proBNP) levels. Conclusions: This study demonstrates that a wireless radar framework combined with deep learning can effectively monitor respiratory patterns that are associated with cardiac function. Its contactless nature may support continuous cardiac function assessments. Clinical Impact: This study highlights the effectiveness of a wireless radar and deep learning framework for monitoring respiratory patterns that are associated with cardiac function (e.g., LVEF), underscoring its potential for long-term cardiac and sleep-disorder management.","PeriodicalId":54255,"journal":{"name":"IEEE Journal of Translational Engineering in Health and Medicine-Jtehm","volume":"13 ","pages":"323-332"},"PeriodicalIF":3.7,"publicationDate":"2025-07-14","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=11079612","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144695521","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Chronic wounds affect 8.5 million Americans, especially the elderly and patients with diabetes. As regular care is critical for proper healing, many patients receive care in their homes from visiting nurses and caregivers with variable wound expertise. Problematic, non-healing wounds should be referred to experts in wound clinics to avoid adverse outcomes such as limb amputations. Unfortunately, due to the lack of wound expertise, referral decisions made in non-clinical settings can be erroneous, delayed or unnecessary. This paper proposes the Deep Multimodal Wound Assessment Tool (DM-WAT), a novel machine learning framework to support visiting nurses by recommending wound referral decisions from smartphone-captured wound images and associated clinical notes. DM-WAT extracts visual features from wound images using DeiT-Base-Distilled, a Vision Transformer (ViT) architecture. Distillation-based training facilitates representation learning and knowledge transfer from a larger teacher model to DeiT-Base, enabling robust performance on our small wound image dataset of 205 wound images. DM-WAT extracts text features from clinical notes using DeBERTa-base, which comprehends context by disentangling content and position information from clinical notes. Visual and text features are combined using an intermediate fusion approach. To overcome the challenges posed by a small and imbalanced dataset, DM-WAT integrates image and text augmentation along with transfer learning via pre-trained feature extractors to achieve high performance. In rigorous evaluation, DM-WAT achieved an accuracy of 77% $pm ~3$ % and an F1 score of 70% $pm ~2$ %, outperforming the prior state of the art and all baseline single-modality and multimodal approaches. Additionally, to interpret DM-WAT’s recommendations, the Score-CAM and Captum interpretation algorithms provided insights into the specific parts of the image and text inputs that the model focused on during decision-making.
{"title":"Multimodal AI for Home Wound Patient Referral Decisions From Images With Specialist Annotations","authors":"Reza Saadati Fard;Emmanuel Agu;Palawat Busaranuvong;Deepak Kumar;Shefalika Gautam;Bengisu Tulu;Diane Strong","doi":"10.1109/JTEHM.2025.3588427","DOIUrl":"https://doi.org/10.1109/JTEHM.2025.3588427","url":null,"abstract":"Chronic wounds affect 8.5 million Americans, especially the elderly and patients with diabetes. As regular care is critical for proper healing, many patients receive care in their homes from visiting nurses and caregivers with variable wound expertise. Problematic, non-healing wounds should be referred to experts in wound clinics to avoid adverse outcomes such as limb amputations. Unfortunately, due to the lack of wound expertise, referral decisions made in non-clinical settings can be erroneous, delayed or unnecessary. This paper proposes the Deep Multimodal Wound Assessment Tool (DM-WAT), a novel machine learning framework to support visiting nurses by recommending wound referral decisions from smartphone-captured wound images and associated clinical notes. DM-WAT extracts visual features from wound images using DeiT-Base-Distilled, a Vision Transformer (ViT) architecture. Distillation-based training facilitates representation learning and knowledge transfer from a larger teacher model to DeiT-Base, enabling robust performance on our small wound image dataset of 205 wound images. DM-WAT extracts text features from clinical notes using DeBERTa-base, which comprehends context by disentangling content and position information from clinical notes. Visual and text features are combined using an intermediate fusion approach. To overcome the challenges posed by a small and imbalanced dataset, DM-WAT integrates image and text augmentation along with transfer learning via pre-trained feature extractors to achieve high performance. In rigorous evaluation, DM-WAT achieved an accuracy of 77% <inline-formula> <tex-math>$pm ~3$ </tex-math></inline-formula>% and an F1 score of 70% <inline-formula> <tex-math>$pm ~2$ </tex-math></inline-formula>%, outperforming the prior state of the art and all baseline single-modality and multimodal approaches. Additionally, to interpret DM-WAT’s recommendations, the Score-CAM and Captum interpretation algorithms provided insights into the specific parts of the image and text inputs that the model focused on during decision-making.","PeriodicalId":54255,"journal":{"name":"IEEE Journal of Translational Engineering in Health and Medicine-Jtehm","volume":"13 ","pages":"341-353"},"PeriodicalIF":3.7,"publicationDate":"2025-07-11","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=11078373","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144705197","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-07-07DOI: 10.1109/JTEHM.2025.3586470
Chukwuemeka Clinton Atabansi;Hui Li;Sheng Wang;Jing Nie;Haijun Liu;Bo Xu;Xichuan Zhou;Dewei Li
Background: Automatic segmentation of liver regions as well as liver lesions such as hepatocellular carcinoma (HCC) from computed tomography (CT) images is critical for accurate diagnosis and therapy planning. With the advent of deep learning techniques such as transformers, computer-aided diagnostic tools (CADs) have the potential to increase the accuracy of liver tumor diagnosis, progression, and treatment planning. However, two major challenges remain: 1) existing models struggle to extract robust spatial features for accurate liver and liver lesion segmentation, and 2) publicly available liver datasets with HCC annotations are limited. Methods: We first present a new liver dataset acquired from Chongqing University Cancer Hospital (CCH-LHCC-CT) with HCC annotations. Second, we developed a novel deep learning architecture (ICT-Net), which is constructed based on a pretrained transformer encoder in conjunction with an advanced feature upscaling and enhanced convolution-transformer decoder formation. Results: We performed liver and liver tumor segmentation on the CCH-LHCC-CT and three public CT liver datasets. The proposed ICT-Net architecture achieves superior accuracy (higher ACC/DSC/IoU, lower HD95) across all datasets. Conclusions: We construct a novel deep-learning architecture that produces robust information for liver and liver tumor segmentation. The statistical and visual results demonstrate that the proposed ICT-Net outperforms other existing approaches investigated in this study in terms of ACC, DSC, and IoU. Clinical Translation Statement: ICT-Net enhances surgical planning accuracy through precise tumor margin delineation and improves therapy response assessment reliability, which holds meaningful promise to support more precise and effective clinical therapeutic strategies for patients with HCC.
{"title":"ICT-Net: An Integrated Convolution and Transformer-Based Network for Complex Liver and Liver Tumor Region Segmentation","authors":"Chukwuemeka Clinton Atabansi;Hui Li;Sheng Wang;Jing Nie;Haijun Liu;Bo Xu;Xichuan Zhou;Dewei Li","doi":"10.1109/JTEHM.2025.3586470","DOIUrl":"https://doi.org/10.1109/JTEHM.2025.3586470","url":null,"abstract":"Background: Automatic segmentation of liver regions as well as liver lesions such as hepatocellular carcinoma (HCC) from computed tomography (CT) images is critical for accurate diagnosis and therapy planning. With the advent of deep learning techniques such as transformers, computer-aided diagnostic tools (CADs) have the potential to increase the accuracy of liver tumor diagnosis, progression, and treatment planning. However, two major challenges remain: 1) existing models struggle to extract robust spatial features for accurate liver and liver lesion segmentation, and 2) publicly available liver datasets with HCC annotations are limited. Methods: We first present a new liver dataset acquired from Chongqing University Cancer Hospital (CCH-LHCC-CT) with HCC annotations. Second, we developed a novel deep learning architecture (ICT-Net), which is constructed based on a pretrained transformer encoder in conjunction with an advanced feature upscaling and enhanced convolution-transformer decoder formation. Results: We performed liver and liver tumor segmentation on the CCH-LHCC-CT and three public CT liver datasets. The proposed ICT-Net architecture achieves superior accuracy (higher ACC/DSC/IoU, lower HD95) across all datasets. Conclusions: We construct a novel deep-learning architecture that produces robust information for liver and liver tumor segmentation. The statistical and visual results demonstrate that the proposed ICT-Net outperforms other existing approaches investigated in this study in terms of ACC, DSC, and IoU. Clinical Translation Statement: ICT-Net enhances surgical planning accuracy through precise tumor margin delineation and improves therapy response assessment reliability, which holds meaningful promise to support more precise and effective clinical therapeutic strategies for patients with HCC.","PeriodicalId":54255,"journal":{"name":"IEEE Journal of Translational Engineering in Health and Medicine-Jtehm","volume":"13 ","pages":"310-322"},"PeriodicalIF":3.7,"publicationDate":"2025-07-07","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=11072178","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144641029","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Introduction: Stroke is one of the most common causes of impaired gait. The use of an ankle-foot orthosis (AFO) is one of the recommended methods to improve gait function in stroke patients. Although the stiffness of the AFO is adjusted for each stroke patient, the effect of stiffness adjustment remains unclear due to the difficulty in measuring the gait parameters in a clinical setting. Objective: This study aimed to investigate the effect of adjusting the AFO stiffness based on the gait ability of stroke patients using a markerless gait analysis method. Methods: A total of 32 individuals with stroke were directed to walk under five conditions: no-AFO and AFO with four different levels of spring stiffness. These springs were used to resist the plantarflexion movements. Moreover, the best gait speed improvement condition (best condition) was determined from the five gait conditions for each participant and was compared with the other conditions, assuming a clinical setting. Spatiotemporal gait parameters such as the gait speed, cadence, step length, stance phase, and swing phase were measured from body keypoints in RGB images. Results and Conclusion: The experimental results showed that the gait speed, cadence, step length on both sides, and stance time on both sides were significantly improved in the best condition compared with the other conditions. This study demonstrated the usefulness of the markerless gait analysis method using a single RGB camera and the effectiveness of AFO stiffness adjustment based on the gait ability of the users.
{"title":"Single Camera-Based Gait Analysis Using Pose Estimation for Ankle-Foot Orthosis Stiffness Adjustment on Individuals With Stroke","authors":"Masataka Yamamoto;Koji Shimatani;Daisuke Matsuura;Yusuke Murakami;Naoya Oeda;Hiroshi Takemura","doi":"10.1109/JTEHM.2025.3585442","DOIUrl":"https://doi.org/10.1109/JTEHM.2025.3585442","url":null,"abstract":"Introduction: Stroke is one of the most common causes of impaired gait. The use of an ankle-foot orthosis (AFO) is one of the recommended methods to improve gait function in stroke patients. Although the stiffness of the AFO is adjusted for each stroke patient, the effect of stiffness adjustment remains unclear due to the difficulty in measuring the gait parameters in a clinical setting. Objective: This study aimed to investigate the effect of adjusting the AFO stiffness based on the gait ability of stroke patients using a markerless gait analysis method. Methods: A total of 32 individuals with stroke were directed to walk under five conditions: no-AFO and AFO with four different levels of spring stiffness. These springs were used to resist the plantarflexion movements. Moreover, the best gait speed improvement condition (best condition) was determined from the five gait conditions for each participant and was compared with the other conditions, assuming a clinical setting. Spatiotemporal gait parameters such as the gait speed, cadence, step length, stance phase, and swing phase were measured from body keypoints in RGB images. Results and Conclusion: The experimental results showed that the gait speed, cadence, step length on both sides, and stance time on both sides were significantly improved in the best condition compared with the other conditions. This study demonstrated the usefulness of the markerless gait analysis method using a single RGB camera and the effectiveness of AFO stiffness adjustment based on the gait ability of the users.","PeriodicalId":54255,"journal":{"name":"IEEE Journal of Translational Engineering in Health and Medicine-Jtehm","volume":"13 ","pages":"333-340"},"PeriodicalIF":3.7,"publicationDate":"2025-07-02","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=11063274","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144704943","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
In recent years, visual cancer information retrieval using Artificial Intelligence has been shown to be effective in diagnosis and treatment. Especially for a modern liver-cancer diagnosis system, the automated tumor annotation plays a crucial role. So-called tumor annotation refers to tagging the tumor in Biomedical images by computer vision technologies such as Deep Learning. After annotation, the tumor information such as tumor location, tumor size and tumor characteristics can be output into a clinical report. To this end, this paper proposes an effective approach that includes tumor segmentation, tumor location, tumor measuring, and tumor recognition to achieve high-quality tumor annotation, thereby assisting radiologists in efficiently making accurate diagnosis reports. For tumor segmentation, a Multi-Residual Attention Unet is proposed to alleviate problems of vanishing gradient and information diversity. For tumor location, an effective Multi-SeResUnet is proposed to partition the liver into 8 couinaud segments. Based on the partitioned segments, the tumor is located accurately. For tumor recognition, an effective multi-labeling classifier is used to recognize the tumor characteristics by the visual tumor features. For tumor measuring, a regression model is proposed to measure the tumor size. To reveal the effectiveness of individual methods, each method was evaluated on real datasets. The experimental results reveal that the proposed methods are more promising than the state-of-the-art methods in tumor segmentation, tumor measuring, tumor localization and tumor recognition. Specifically, the average tumor size error and the annotation accuracy are 0.432 cm and 91.6%, respectively, which suggest potential for reducing radiologists’ workload. In summary, this paper proposes an effective tumor annotation for an automated diagnosis support system. Clinical and Translational Impact Statement—The proposed methods have been evaluated and shown to significantly improve the efficiency and accuracy of liver tumor annotation, reducing the time required for radiologists to complete reports on tumor segmentation, liver partition, tumor measuring and tumor recognition. By integrating into existing clinical decision support systems, it has the potential to reduce diagnostic errors and treatment delays, thereby improving patient outcomes and clinical workflow.
{"title":"Effective Tumor Annotation for Automated Diagnosis of Liver Cancer","authors":"Yi-Hsuan Chuang;Ja-Hwung Su;Tzu-Chieh Lin;Hue-Xin Cheng;Pin-Hao Shen;Jin-Ping Ou;Ding-Hong Han;Yi-Wen Liao;Yeong-Chyi Lee;Yu-Fan Cheng;Tzung-Pei Hong;Katherine Shu-Min Li;Yi Lu;Chih-Chi Wang","doi":"10.1109/JTEHM.2025.3576827","DOIUrl":"https://doi.org/10.1109/JTEHM.2025.3576827","url":null,"abstract":"In recent years, visual cancer information retrieval using Artificial Intelligence has been shown to be effective in diagnosis and treatment. Especially for a modern liver-cancer diagnosis system, the automated tumor annotation plays a crucial role. So-called tumor annotation refers to tagging the tumor in Biomedical images by computer vision technologies such as Deep Learning. After annotation, the tumor information such as tumor location, tumor size and tumor characteristics can be output into a clinical report. To this end, this paper proposes an effective approach that includes tumor segmentation, tumor location, tumor measuring, and tumor recognition to achieve high-quality tumor annotation, thereby assisting radiologists in efficiently making accurate diagnosis reports. For tumor segmentation, a Multi-Residual Attention Unet is proposed to alleviate problems of vanishing gradient and information diversity. For tumor location, an effective Multi-SeResUnet is proposed to partition the liver into 8 couinaud segments. Based on the partitioned segments, the tumor is located accurately. For tumor recognition, an effective multi-labeling classifier is used to recognize the tumor characteristics by the visual tumor features. For tumor measuring, a regression model is proposed to measure the tumor size. To reveal the effectiveness of individual methods, each method was evaluated on real datasets. The experimental results reveal that the proposed methods are more promising than the state-of-the-art methods in tumor segmentation, tumor measuring, tumor localization and tumor recognition. Specifically, the average tumor size error and the annotation accuracy are 0.432 cm and 91.6%, respectively, which suggest potential for reducing radiologists’ workload. In summary, this paper proposes an effective tumor annotation for an automated diagnosis support system. Clinical and Translational Impact Statement—The proposed methods have been evaluated and shown to significantly improve the efficiency and accuracy of liver tumor annotation, reducing the time required for radiologists to complete reports on tumor segmentation, liver partition, tumor measuring and tumor recognition. By integrating into existing clinical decision support systems, it has the potential to reduce diagnostic errors and treatment delays, thereby improving patient outcomes and clinical workflow.","PeriodicalId":54255,"journal":{"name":"IEEE Journal of Translational Engineering in Health and Medicine-Jtehm","volume":"13 ","pages":"251-260"},"PeriodicalIF":3.7,"publicationDate":"2025-06-05","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=11025471","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144281292","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
京公网安备 11010802042870号
京ICP备2023020795号-1
扫码关注我们
求助内容：
应助结果提醒方式：