{"title":"Automated strength-interval curve generation using actors","authors":"Raymond J. Spiteri, Joyce Reimer, Kyle Klenk","doi":"10.1016/j.cmpb.2025.108784","DOIUrl":"10.1016/j.cmpb.2025.108784","url":null,"abstract":"<div><h3>Background and Objective:</h3><div>Strength-interval (SI) curves are used by physiologists to quantify the response of excitable tissue as a function of the strength and timing of an electrical stimulus. In the context of cardiac electrophysiology, SI curves characterize the refractoriness of cardiac tissue as a function of inter-stimulus interval length. Although conventionally collected experimentally, this type of information can now more conveniently be obtained through computational simulation. Nevertheless, the computational generation of SI curves can be labor-intensive and time-consuming due to its iterative nature, the number and size of computations required, and the amount of manual researcher intervention involved. The objective of this study is to use the Actor Model of concurrent computation to automate the process of SI curve generation, relieving much of the burden from the researcher while maximizing the use of available computational resources.</div></div><div><h3>Methods:</h3><div>The C++ Actor Framework is used to create an automated tool for controlling the <em>openCARP</em> simulation platform. An SI curve is generated for the bidomain model of electrophysiology through the use of sophisticated parallelization techniques, e.g., dynamic information passing between parallel simulations, facilitated by the use of actors. Computational resource management is optimized by the dynamic monitoring, assessment, and reallocation based on each actor’s current simulation state in relation to all other actors.</div></div><div><h3>Results:</h3><div>A bidomain SI curve with 31 data points that takes 27.5 h to compute conventionally using 80 CPU cores is now generated in 15.4 h. This is over 40% faster than using conventional parallel programming techniques with MPI. Furthermore, it requires no researcher intervention, which can add significantly to the time to solution.</div></div><div><h3>Conclusion:</h3><div>Novel parallelization techniques enabled via the Actor Model significantly improve the efficiency of computational SI curve generation, both from the viewpoints of computation and labor intensiveness. This improvement in efficiency has implications for future studies involving cardiac refractory tissue, along with other types of excitable tissue, including the rapid generation of both general and patient-specific SI curves and the use of these curves for design and <em>in silico</em> testing of new therapeutic tools such as personalized pacemakers.</div></div>","PeriodicalId":10624,"journal":{"name":"Computer methods and programs in biomedicine","volume":"267 ","pages":"Article 108784"},"PeriodicalIF":4.9,"publicationDate":"2025-04-25","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143886061","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}
Marie Seret , Vincent Uyttendaele , J. Geoffrey Chase , Thomas Desaive
{"title":"In-silico assessment of longer measurement intervals in glycaemic control to match clinical practice","authors":"Marie Seret , Vincent Uyttendaele , J. Geoffrey Chase , Thomas Desaive","doi":"10.1016/j.cmpb.2025.108806","DOIUrl":"10.1016/j.cmpb.2025.108806","url":null,"abstract":"<div><h3>Background and Objective</h3><div>STAR is a patient-specific glycaemic control (GC) framework accounting for both inter- and intra- patient variability to modulate insulin and nutrition in ICU patients. While providing safe, effective control to all patient, the workload induced by STAR represents a clinical burden in some ICUs. This study aims at extending the treatment interval of STAR from 1–3 hourly to 1–6 hourly to reduce the workload associated with STAR and assessing the impact on GC outcomes using virtual trials.</div></div><div><h3>Methods</h3><div>Retrospective data form 606 patients are used to create virtual patients. Insulin sensitivity is identified for each patient using a physiological model and used to build and validate the new stochastic models to provide up to 6-hourly predictions using five-fold cross-validation. Virtual trials are performed and safety, performance, nutrition intake and workload are compared and analysed.</div></div><div><h3>Results</h3><div>The extended STAR protocol 1–6 hourly measurement interval still provided high control safety and efficacy. Results showed slightly reduced %BG within the safe target band 4.4–8.0 mmol/L (from 83.8 to 81.4 %) as the measurement interval increased. It also resulted in an increased risk of hyper- (from 14.5 to 16.9 %BG > 8.0 mmol/L) and severe hypo- (from 0.03 to 0.05 %BG < 2.2 mmol/L) glycaemia. Insulin and nutrition rates decreased (from 3.5 [2.0 5.0] to 2.5 [1.7 3.0] U/h and from 100 [85 100] to 89 [71 100] % goal feed (GF) respectively). The workload was significantly reduced from 12 to 8 measurements per day.</div></div><div><h3>Conclusions</h3><div>The workload was successfully reduced by extending the measurement interval, approaching clinical practice. High performance and safety are achieved. However, the results also highlight a clear risk and reward trade-off in glycaemic control with the increased risk of hyper- and hypo- glycaemia and the reduced nutrition rates. Choosing an intermediate measurement interval could be an interesting solution. Clinical trials should be conducted to further confirm those results and consider the adoption of longer treatment intervals in STAR GC framework.</div></div>","PeriodicalId":10624,"journal":{"name":"Computer methods and programs in biomedicine","volume":"267 ","pages":"Article 108806"},"PeriodicalIF":4.9,"publicationDate":"2025-04-25","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143918387","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}
Peiwen Li , Tianyu Liu , Heyu Ma , Dan Li , Chengcheng Liu , Dean Ta
{"title":"A multi-task neural network for full waveform ultrasonic bone imaging","authors":"Peiwen Li , Tianyu Liu , Heyu Ma , Dan Li , Chengcheng Liu , Dean Ta","doi":"10.1016/j.cmpb.2025.108807","DOIUrl":"10.1016/j.cmpb.2025.108807","url":null,"abstract":"<div><h3>Background and objective</h3><div>It is a challenging task to use ultrasound for bone imaging, as the bone tissue has a complex structure with high acoustic impedance and speed-of-sound (SOS). Recently, full waveform inversion (FWI) has shown promising imaging for musculoskeletal tissues. However, the FWI showed a limited ability and tended to produce artifacts in bone imaging because the inversion process would be more easily trapped in local minimum for bone tissue with a large discrepancy in SOS distribution between bony and soft tissues. In addition, the application of FWI required a high computational burden and relatively long iterations. The objective of this study was to achieve high-resolution ultrasonic imaging of bone using a deep learning-based FWI approach.</div></div><div><h3>Method</h3><div>In this paper, we proposed a novel network named CEDD-Unet. The CEDD-Unet adopts a Dual-Decoder architecture, with the first decoder tasked with reconstructing the SOS model, and the second decoder tasked with finding the main boundaries between bony and soft tissues. To effectively capture multi-scale spatial-temporal features from ultrasound radio frequency (RF) signals, we integrated a Convolutional LSTM (ConvLSTM) module. Additionally, an Efficient Multi-scale Attention (EMA) module was incorporated into the encoder to enhance feature representation and improve reconstruction accuracy.</div></div><div><h3>Results</h3><div>Using the ultrasonic imaging modality with a ring array transducer, the performance of CEDD-Unet was tested on the SOS model datasets from human bones (noted as Dataset1) and mouse bones (noted as Dataset2), and compared with three classic reconstruction architectures (Unet, Unet++, and Att-Unet), four state-of-the-art architecture (InversionNet, DD-Net, UPFWI, and DEFE-Unet). Experiments showed that CEDD-Unet outperforms all competing methods, achieving the lowest MAE of 23.30 on Dataset1 and 25.29 on Dataset2, the highest SSIM of 0.9702 on Dataset1 and 0.9550 on Dataset2, and the highest PSNR of 30.60 dB on Dataset1 and 32.87 dB on Dataset2. Our method demonstrated superior reconstruction quality, with clearer bone boundaries, reduced artifacts, and improved consistency with ground truth. Moreover, CEDD-Unet surpasses traditional FWI by producing sharper skeletal SOS reconstructions, reducing computational cost, and eliminating the reliance for an initial model. Ablation studies further confirm the effectiveness of each network component.</div></div><div><h3>Conclusion</h3><div>The results suggest that CEDD-Unet is a promising deep learning-based FWI method for high-resolution bone imaging, with the potential to reconstruct accurate and sharp-edged skeletal SOS models.</div></div>","PeriodicalId":10624,"journal":{"name":"Computer methods and programs in biomedicine","volume":"267 ","pages":"Article 108807"},"PeriodicalIF":4.9,"publicationDate":"2025-04-25","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143891344","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}
Pu Wang , Teng-Hui Chen , Mei-Yun Chang , Hai-Yen Hsia , Meng Dai , Yifan Liu , Yeong-Long Hsu , Feng Fu , Zhanqi Zhao
{"title":"An explainable artificial intelligence framework for weaning outcomes prediction using features from electrical impedance tomography","authors":"Pu Wang , Teng-Hui Chen , Mei-Yun Chang , Hai-Yen Hsia , Meng Dai , Yifan Liu , Yeong-Long Hsu , Feng Fu , Zhanqi Zhao","doi":"10.1016/j.cmpb.2025.108811","DOIUrl":"10.1016/j.cmpb.2025.108811","url":null,"abstract":"<div><h3>Background</h3><div>Prolonged mechanical ventilation (PMV) might cause ventilator-associated pneumonia and diaphragmatic injury, and may lead to worsening clinical weaning outcomes. The present study proposes a comprehensive machine learning (ML) framework for predicting the weaning outcomes of patients with PMV, without relying on ventilator data, by utilizing features from electrical impedance tomography (EIT).</div></div><div><h3>Methods</h3><div>EIT data from 58 patients with PMV were analyzed. Extracted EIT image features were standardized using the min-max method. The Boruta method was employed to select significant features for the ML model. To balance the data, the SMOTE method was utilized. Ten ML algorithms commonly used in clinical prediction were compared. The SHAP and LIME methods were used to explain the ML models. Feature selection, data balancing, and hyperparameter adjustment all adopt the Leave-One-Out cross-validation method to avoid overfitting.</div></div><div><h3>Results</h3><div>The area under the receiver operating characteristic (AUC), specificity, and precision of the ML model with SMOTE balance were significantly improved (<em>p</em> < 0.05) compared to unbalanced data. However, the sensitivity was reduced considerably (<em>p</em> = 0.02). The optimal ML model, extreme gradient boost (XGBoost), demonstrated excellent performance: AUC = 0.862, sensitivity = 0.923, specificity = 0.800, accuracy = 0.889, precision = 0.923, and f-score = 0.923. Decision Curve Analysis and calibration curve evaluation indicated that the model has high clinical generality and reliability. The SHAP and LIME methods enabled model interpretation at both the global and individual sample levels.</div></div><div><h3>Conclusion</h3><div>The weaning outcome prediction model based on EIT data does not rely on ventilator data, which is suitable for a broader range of weaning scenarios. We proposed a comprehensive ML framework for weaning outcome prediction and incorporated the SHAP and LIME methods, which significantly improved the interpretability of the model.</div></div>","PeriodicalId":10624,"journal":{"name":"Computer methods and programs in biomedicine","volume":"267 ","pages":"Article 108811"},"PeriodicalIF":4.9,"publicationDate":"2025-04-25","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143918388","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}
{"title":"Prescription data and demographics: An explainable machine learning exploration of colorectal cancer risk factors based on data from Danish national registries","authors":"Abdolrahman Peimankar , Olav Sivertsen Garvik , Bente Mertz Nørgård , Jens Søndergaard , Dorte Ejg Jarbøl , Sonja Wehberg , Søren Paludan Sheikh , Ali Ebrahimi , Uffe Kock Wiil , Maria Iachina","doi":"10.1016/j.cmpb.2025.108774","DOIUrl":"10.1016/j.cmpb.2025.108774","url":null,"abstract":"<div><h3>Objectives:</h3><div>Despite substantial advancements in both treatment and prevention, colorectal cancer continues to be a leading cause of global morbidity and mortality. This study investigated the potential of using demographics and prescribed drug information to predict risk of colorectal cancer using a machine learning approach.</div></div><div><h3>Methods:</h3><div>Five different machine learning algorithms, including Logistic Regression, XGBoost, Random Forests, kNN, and Voting Classifier, were initially developed and evaluated for their predictive capabilities across various time horizons (3, 6, 12, and 36 months). To enhance transparency and interpretability, explainable techniques were employed to understand the model’s predictions and identify the relative contributions of factors like age, sex, social status, and prescribed medications, promoting trust and clinical insights. While all developed models, including simpler ones such as Logistic Regression, demonstrated comparable performance, the Voting Classifier, as an ensemble model, was selected for further investigation due to its inherent diversity and generalizability. This ensemble model combines predictions from multiple base models, reducing the risk of overfitting and improving the robustness of the final prediction.</div></div><div><h3>Results:</h3><div>The model demonstrated consistent performance across these time horizons, achieving a precision consistently above 0.99, indicating high ability in identifying patients at risk. However, the recall remained relatively low (around 0.6), highlighting the model’s limitations in comprehensively identifying all at risk patients, despite its high precision. This suggests additional investigations in future studies to further enhance the performance of the proposed model.</div></div><div><h3>Conclusion:</h3><div>Machine learning models can identify individuals at higher risk for developing colorectal cancer, enabling earlier interventions and personalized risk management strategies. However, further studies are needed before implementation in clinical practice.</div></div>","PeriodicalId":10624,"journal":{"name":"Computer methods and programs in biomedicine","volume":"267 ","pages":"Article 108774"},"PeriodicalIF":4.9,"publicationDate":"2025-04-24","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143876557","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}
Chua Ming , Geraldine JW Lee , Yao Neng Teo , Yao Hao Teo , Xinyan Zhou , Elizabeth SY Ho , Emma MS Toh , Marcus Eng Hock Ong , Benjamin YQ Tan , Andrew FW Ho
{"title":"Deep learning modelling to forecast emergency department visits using calendar, meteorological, internet search data and stock market price","authors":"Chua Ming , Geraldine JW Lee , Yao Neng Teo , Yao Hao Teo , Xinyan Zhou , Elizabeth SY Ho , Emma MS Toh , Marcus Eng Hock Ong , Benjamin YQ Tan , Andrew FW Ho","doi":"10.1016/j.cmpb.2025.108808","DOIUrl":"10.1016/j.cmpb.2025.108808","url":null,"abstract":"<div><h3>Background</h3><div>Accurate prediction of hospital emergency department (ED) patient visits and acuity levels have potential to improve resource allocation including manpower planning and hospital bed allocation. Internet search data have been used in medical applications like disease pattern prediction and forecasting ED volume. Past studies have also found stock market price positively correlated with ED volume.</div></div><div><h3>Objective</h3><div>To determine whether incorporating Internet search data and stock market price to calendar and meteorological data can improve deep learning prediction of ED patient volumes, and whether hybrid deep learning architectures are better in prediction.</div></div><div><h3>Methods</h3><div>Permutations of various input variables namely calendar, meteorological, Google Trends online search data, Standard and Poor’s (S&P) 500 index, and Straits Times Index (STI) data were incorporated into deep learning models long short-term memory (LSTM), one-dimensional convolutional neural network (1D CNN), stacked 1D CNN-LSTM, and five CNN-LSTM hybrid modules to predict daily Singapore General Hospital ED patient volume from 2010–2012.</div></div><div><h3>Results</h3><div>Incorporating STI to calendar and meteorological data improved performance of CNN-LSTM hybrid models. Addition of queried absolute Google Trends search terms to calendar and meteorological data improved performance of two out of five hybrid models. The best LSTM model across all predictor permutations had mean absolute percentage error of 4.8672 %.</div></div><div><h3>Conclusion</h3><div>LSTM provides strong predictive ability for daily ED patient volume. Local stock market index has potential to predict ED visits. Amongst predictors evaluated, calendar and meteorological data was sufficient for a relatively accurate prediction.</div></div>","PeriodicalId":10624,"journal":{"name":"Computer methods and programs in biomedicine","volume":"267 ","pages":"Article 108808"},"PeriodicalIF":4.9,"publicationDate":"2025-04-24","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143891383","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}
Qing Zhang , Xiaoxiao Wu , Xiang Li , Wei Ma , Tongquan Wu , Liuyue Li , Fan Hu , Yicheng Xie , Xinglong Wu
{"title":"TransAnno-Net: A Deep Learning Framework for Accurate Cell Type Annotation of Mouse Lung Tissue Using Self-supervised Pretraining","authors":"Qing Zhang , Xiaoxiao Wu , Xiang Li , Wei Ma , Tongquan Wu , Liuyue Li , Fan Hu , Yicheng Xie , Xinglong Wu","doi":"10.1016/j.cmpb.2025.108809","DOIUrl":"10.1016/j.cmpb.2025.108809","url":null,"abstract":"<div><h3>Background</h3><div>Single-cell RNA sequencing (scRNA-seq) has become a significant tool for addressing complex issuess in the field of biology. In the context of scRNA-seq analysis, it is imperative to accurately determine the type of each cell. However, conventional supervised or semi-supervised methodologies are contingent on expert labels and incur substantial labeling costs, In contrast self-supervised pre-training strategies leverage unlabeled data during the pre-training phase and utilise a limited amount of labeled data in the fine-tuning phase, thereby greatly reducing labor costs. Furthermore, the fine-tuning does not need to learn the feature representations from scratch, enhancing the efficiency and transferability of the model.</div></div><div><h3>Methods</h3><div>The proposed methodology is outlined below. The deep learning framework, TransAnno-Net, is based on transfer learning and a Transformer architecture. It has been designed for efficient and accurate cell type annotations in large-scale scRNA-seq datasets of mouse lung organs. Specifically, TransAnno-Net is pre-trained on the scRNA-seq lung data of approximately 100,000 cells to acquire gene-gene similarities via self-supervised learning. It is then migrated to a relatively small number of datasets to fine-tune specific cell type annotation tasks. To address the issue of imbalance in cell types commonly observed in scRNA-seq data, we applied a random oversampling technique is applied to the fine-tuned dataset. This is done to mitigate the impact of distributional imbalance on the annotation outcomes.</div></div><div><h3>Results</h3><div>The experimental findings demonstrate that TransAnno-Net exhibits superior performance with an AUC of 0.979, 0.901, and 0.982, respectively, on three mouse lung datasets, outperforming eight state-of-the-art (SOTA) methods. In addition, TransAnno-Net demonstrates robust performance on cross-organ, cross-platform datasets, and is competitive with the fully supervised learning-based method.</div></div><div><h3>Conclusion</h3><div>The TransAnno-Net method is a highly effective cross-platform and cross-data set single-cell type annotation method for mouse lung tissues and supports cross-organ cell type annotation. This approach is expected to enhance the efficiency of research on the biological mechanisms of complex biological systems and diseases.</div></div>","PeriodicalId":10624,"journal":{"name":"Computer methods and programs in biomedicine","volume":"267 ","pages":"Article 108809"},"PeriodicalIF":4.9,"publicationDate":"2025-04-24","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143891382","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}
Sebastian Bachmann , Gianluca Iori , Kay Raum , Dieter H. Pahr , Alexander Synek
{"title":"Predicting physiological hip joint loads with inverse bone remodeling using clinically available QCT images","authors":"Sebastian Bachmann , Gianluca Iori , Kay Raum , Dieter H. Pahr , Alexander Synek","doi":"10.1016/j.cmpb.2025.108805","DOIUrl":"10.1016/j.cmpb.2025.108805","url":null,"abstract":"<div><h3>Background and objective</h3><div>Assessing joint-level loading conditions <em>in vivo</em> is challenging due to invasive measurement or complex computation. Inverse bone remodeling (IBR) offers a different approach by recovering the loading conditions directly from computed tomography (CT) images of the bone microstructure by finding the magnitudes to a set of load cases that load the bone optimally, i.e., maximally homogeneously. An efficient IBR method was recently proposed based on homogenized finite element (hFE) models. This study compared the hip joint load predictions of hFE-based IBR with clinically feasible CT scans to those obtained with the current gold standard, micro-FE-based IBR.</div></div><div><h3>Methods</h3><div>A set of 20 proximal femora was scanned <em>ex vivo</em>, both with a clinical quantitative CT (QCT) scanner (0.3 mm resolution) and an Xtreme CT II (XCT2) scanner (0.03 mm resolution). Finite element (FE) models with decreasing complexity were automatically created from those images. Micro-FE (µFE) models based on XCT2 images served as a baseline. hFE models based on the QCT images were created as clinically feasible models. Further intermediate models were created to trace sources of errors. IBR was applied to predict the optimal scaling factors of twelve unit load cases distributed over the femoral head.</div></div><div><h3>Results</h3><div>The predicted loads of the newly developed workflow for QCT images within IBR followed a trend seen previously with hFE models created from high-resolution images, such as XCT2. The peak load magnitudes of µFE and hFE-based IBR were well correlated (R²=76.8 %), and the overall distribution of the loads was similar. However, an additional peak load calibration was required to obtain quantitative agreement (CCC=82.8 %).</div></div><div><h3>Conclusions</h3><div>A thorough comparison of µFE-based IBR and hFE-based IBR using QCT data was performed for the first time. A clinically feasible workflow, including a peak calibration, is presented, allowing for fast prediction of physiological peak hip joint loads.</div></div>","PeriodicalId":10624,"journal":{"name":"Computer methods and programs in biomedicine","volume":"267 ","pages":"Article 108805"},"PeriodicalIF":4.9,"publicationDate":"2025-04-24","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143886105","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}
Ciro Mennella , Massimo Esposito , Giuseppe De Pietro , Umberto Maniscalco
{"title":"Multiscale activity recognition algorithms to improve cross-subjects performance resilience in rehabilitation monitoring systems","authors":"Ciro Mennella , Massimo Esposito , Giuseppe De Pietro , Umberto Maniscalco","doi":"10.1016/j.cmpb.2025.108792","DOIUrl":"10.1016/j.cmpb.2025.108792","url":null,"abstract":"<div><h3>Background and Objective:</h3><div>This study introduces multiscale feature learning to develop more robust and resilient activity recognition algorithms, aimed at accurately tracking and quantifying rehabilitation exercises while minimizing performance disparities across subjects with varying motion-related characteristics.</div></div><div><h3>Methods:</h3><div>Advanced architectures designed to process multi-channel time series data using two parallel branches that extract features at different scales were developed and tested.</div></div><div><h3>Results:</h3><div>The results indicate that multiscale algorithms consistently outperform traditional approaches, demonstrating enhanced performance, particularly among patient subjects. Specifically, the multiscale tCNN and multiscale CNN-LSTM achieved accuracies of 91% and 90%, respectively, while the multiscale ConvLSTM maintained strong performance at 89%. Notably, the multiscale Transformer emerged as the most effective model, achieving the best average accuracy of 93%.</div></div><div><h3>Conclusions:</h3><div>This research underscores the need to explore advanced methods for enhancing activity recognition systems in healthcare, where accurate exercise monitoring and evaluation are becoming essential for effective and personalized treatment in telemedicine services.</div></div>","PeriodicalId":10624,"journal":{"name":"Computer methods and programs in biomedicine","volume":"267 ","pages":"Article 108792"},"PeriodicalIF":4.9,"publicationDate":"2025-04-23","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143869989","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}
{"title":"Data-driven sleep structure deciphering based on cardiorespiratory signals","authors":"Ming Huang , Osuke Iwata , Kiyoko Yokoyama , Toshiyo Tamura","doi":"10.1016/j.cmpb.2025.108769","DOIUrl":"10.1016/j.cmpb.2025.108769","url":null,"abstract":"<div><h3>Background and Objective</h3><div>:</div><div>Cardiorespiratory signals provide a novel perspective for understanding sleep structure through the physiological mechanism of cardiopulmonary coupling. This mechanism divides the coupling spectrum into high-frequency (HF) and low-frequency (LF) bands, indicating that signal segments of 4–8 min are optimal for analysis. However, the lack of labels tailored to these signals has led to reliance on the American Academy of Sleep Medicine (AASM) definitions, which are primarily designed for electroencephalogram (EEG) and electrooculogram (EOG) data. This study aims to address the challenge of transitioning from AASM-defined labels to cardiorespiratory-oriented ones and to evaluate the feasibility of using these signals for accurate sleep structure recognition.</div></div><div><h3>Methods:</h3><div>To align with the physiological characteristics of cardiorespiratory signals, AASM labels were modified by excluding the N2 stage due to its overlap of stable and unstable non-rapid eye movement (NREM) phases, which introduces ambiguity. The modified dataset focused on the wake, N1, deep sleep (N3), and rapid eye movement (REM) stages. A physiologically-inspired deep-learning model (PIDM) was developed to extract features from cardiorespiratory time series and classify sleep stages. Post-analysis assessed the physiological validity of the model’s N2 predictions by evaluating the HF-to-LF ratio and respiratory variability.</div></div><div><h3>Results:</h3><div>The pipeline, combining the modified labeling scheme with the PIDM model, achieved balanced accuracy scores of 0.83, 0.86, and 0.78 for wake, deep sleep, and REM stages, respectively in the normal group; and 0.92, 0.95, and 0.90 in the mild and moderate sleep apnea groups. Post-analysis revealed that most N2 samples were attributed to stable NREM sleep, characterized by higher HF-to-LF ratios and lower respiratory variability, aligning with physiological understanding.</div></div><div><h3>Conclusions:</h3><div>This study highlights the physiological relevance of cardiorespiratory signals for sleep structure recognition. By addressing the uncertainty in N2 classification through exclusion and redefinition, the proposed pipeline effectively distinguished wake, deep sleep, and REM stages. These findings demonstrate the potential of cardiorespiratory signals as a robust, practical, and EEG-independent tool for sleep analysis, particularly in home healthcare settings.</div></div>","PeriodicalId":10624,"journal":{"name":"Computer methods and programs in biomedicine","volume":"267 ","pages":"Article 108769"},"PeriodicalIF":4.9,"publicationDate":"2025-04-23","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143886104","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}