Computer Methods and Programs in Biomedicine

Colonic volume content measurements can provide important information about the digestive tract physiology. Development of automated analyses will accelerate the translation of these measurements into clinical practice. In this paper, we test the effect of data dimension on the success of deep learning approaches to segment colons from MRI data. Deep learning network models were developed which used either 2D slices, complete 3D volumes and 2.5D partial volumes. These represent variations in the trade-off between the size and complexity of a network and its training regime, and the limitation of only being able to use a small section of the data at a time: full 3D networks, for example, have more image context available for decision making but require more powerful hardware to implement. For the datasets utilised here, 3D data was found to outperform 2.5D data, which in turn performed better than 2D datasets. The maximum Dice scores achieved by the networks were 0.898, 0.834 and 0.794 respectively. We also considered the effect of ablating varying amounts of data on the ability of the networks to label images correctly. We achieve dice scores of 0.829, 0.827 and 0.389 for 3D single slices ablation, 3D multi-slice ablation and 2.5D middle slice ablation. In addition, we examined another practical consideration of deep learning, that of how well a network performs on data from another acquisition device. Networks trained on images from a Philips Achieva MRI system yielded Dice scores of up to 0.77 in the 3D case when tested on images captured from a GE Medical Systems HDxt (both 1.5 Tesla) without any retraining. We also considered the effect of single versus multimodal MRI data showing that single modality dice scores can be boosted from 0.825 to 0.898 when adding an extra modality.

Central venous pressure (CVP), a key component of hemodynamic monitoring, is widely used to guide fluid resuscitation in critically ill patients. It is typically measured using central venous line catheterization, which is the gold standard, but this method is invasive, time-consuming, and associated with complications. This study aims to investigate whether machine learning (ML)-based analysis of features extracted from a non-invasive, standard-of-care waveform--the photoplethysmography (PPG) signal--can identify patients with elevated CVP. We trained Light Gradient-Boosting Machine (LightGBM) model using a large perioperative dataset (MLORD), containing 17,327 surgical patients from 2019 to 2022 at UCLA. For this study, we selected 1665 patients with both PPG and CVP waveforms available. A total of 843 PPG features per cardiac cycle (CC) were extracted from the PPG waveforms using a signal processing-based feature extraction tool, along with the simultaneous maximum value calculated from the corresponding CCs in the CVP waveform. Additionally, for each patient, the average and standard deviation of each PPG feature, as well as the mean of the maximum CVP values, were calculated across all cardiac cycles, resulting in 843 averaged PPG features, 843 PPG feature standard deviations, and one averaged maximum CVP value per patient. The average maximum CVP value was used as the ground truth to classify patients as either normal (5 [≤] CVP [≤] 15 mmHg) or elevated (CVP > 15 mmHg). Of the 1,665 patients, 1,182 were normal and 483 were elevated. The dataset was split into 90% for training (1,063 normal and 435 elevated) and 10% for testing (119 normal and 48 elevated). From the 1686 PPG features (843 averaged and 843 standard deviation), 246 were selected for model development using the Recursive Feature Elimination with Cross-Validation (RFECV) approach. To further enhance performance, hyperparameters were tuned through 5-fold cross-validation on the training set. Finally, the best-performing configuration was retrained on the full training data, and its performance was evaluated on the held-out test set. To provide a robust estimate and confidence interval, a bootstrapping procedure with 100 iterations was performed on the test set. The LightGBM classifier achieved a mean area under the receiver operating characteristic curve (AUC) of 0.79 (95% CI: 0.71-0.84) and mean accuracy of 0.71 (95% CI: 0.65-0.77), demonstrating good discriminatory power in distinguishing between patients with normal and elevated CVP. This study highlights the ability of PPG-derived features to discriminate between patients with normal and elevated CVP using ML. These early findings lay the groundwork for future research aimed at developing non-invasive approaches to CVP assessment.

Advanced visual computing solutions and 3D printing are starting to move from the engineering and development stage to being integrated into clinical pipelines for training, planning and guidance of complex interventions. Commonly, clinicians make decisions based on the exploration of patient-specific medical images in 2D flat monitors using specialised software with standard multi-planar reconstruction (MPR) visualisation. The new generation of visual computing technologies such as 3D imaging, 3D printing, 3D advanced rendering, Virtual Reality and in-silico simulations from Virtual Physiological Human models, provide complementary ways to better understand the structure and function of the organs under study and improve and personalise clinical decisions. Cardiology is a medical field where new visual computing solutions are already having an impact in decisions such as the selection of the optimal therapy for a given patient. A good example is the role of emerging visualisation technologies to choose the most appropriate settings of a left atrial appendage occluder (LAAO) device that needs to be implanted in some patients with atrial fibrillation having contraindications to drug therapies. Clinicians need to select the type and size of the LAAO device to implant, as well as the location to be deployed. Usually, interventional cardiologists make these decisions after the analysis of patient-specific medical images in 2D flat monitors with MPR visualisation, before and during the procedure, obtain manual measurements characterising the cardiac anatomy of the patient to avoid adverse events after the implantation. In this paper we evaluate several advanced visual computing solutions such as web-based 3D imaging visualisation (VIDAA platform), Virtual Reality (VRIDAA platform) and computational fluid simulations and 3D printing for the planning of LAAO device implantations. Six physicians including three interventional and three imaging cardiologists, with different level of experience in LAAO, tested the different technologies in preoperative data of 5 patients to identify the usability, friendliness, limitations and requirements for clinical translation of each technology through a qualitative questionnaire. The obtained results demonstrate the potential impact of advanced visual computing solutions to improve the planning of LAAO interventions but also a need of unification of them in order to be able to be uses in a clinical environment.

BACKGROUND & AIMSBarretts epithelium measurement using widely accepted Prague C&M criteria is highly operator dependent. By reconstructing the surface of the Barretts area in 3D from endoscopy video, we propose a novel methodology for measuring the C&M score automatically. This 3D reconstruction provides an extended field of view and also allows to precisely quantify the Barretts area including islands. We aim to assess the accuracy of the extracted measurements from phantom and demonstrate their clinical usability. METHODSAdvanced deep learning techniques are utilised to design estimators for depth and camera pose required to map standard endoscopy video to a 3D surface model. By segmenting the Barretts area and locating the position of the gastro-oesophageal junction (GEJ) we measure C&M scores and the Barretts oesophagus areas (BOA). Experiments using a purpose-built 3D printed oesophagus phantom and high-definition video from 98 patients scored by an expert endoscopist are used for validation. RESULTSEndoscopic phantom video data demonstrated a 95 % accuracy with a marginal {+/-} 1.8 mm average deviation for C&M and island measurements, while for BOA we achieved nearly 93 % accuracy with only {+/-} 1.1 cm2 average deviation compared to the ground-truth measurements. On patient data, the C&M measurements provided by our system concord with the reference provided by expert upper GI endoscopists. CONCLUSIONSThe proposed methodology is suitable for extracting Prague C&M scores automatically with a high degree of accuracy. Providing an accurate measurement of the entire Barretts area provides new opportunities for risk stratification and the assessment of therapy response.

This paper introduces a Lightweight U-Net (LWU-Net) method for efficient gastro-intestinal tract segmentation in resource-constrained environments. The proposed model seeks to strike a balance between computational efficiency, memory efficiency, and segmentation accuracy. The model achieves competitive performance while reducing computational power needed with improvements including depth-wise separable convolutions and optimised network depth. The evaluation is conducted using data from a Kaggle competition-UW Madison gastrointestinal tract image segmentation, demonstrating the models effectiveness and generalizability. The findings demonstrate that the LWU-Net model has encouraging promise for precise medical diagnoses in resource-constrained settings, enabling effective image segmentation with slightly less than a fifth of as many trainable parameters as the U-Net model.

Arterial blood pressure (ABP) and photoplethysmography (PPG) waveforms contain valuable clinical information and play a crucial role in cardiovascular health monitoring, medical research, and managing medical conditions. The features extracted from PPG waveforms have various clinical applications ranging from blood pressure monitoring to nociception monitoring, while features from ABP waveforms can be used to calculate cardiac output and predict hypertension or hypotension. In recent years, many machine learning models have been proposed to utilize both PPG and ABP waveform features for these healthcare applications. However, the lack of standardized tools for extracting features from these waveforms could potentially affect their clinical effectiveness. In this paper, we propose an automatic signal processing tool for extracting features from ABP and PPG waveforms. Additionally, we generated a PPG feature library from a large perioperative dataset comprising 17,327 patients using the proposed tool. This PPG feature library can be used to explore the potential of these extracted features to develop machine learning models for non-invasive blood pressure estimation.

Osteoporotic hip fractures (HFs) in the elderly are a pertinent issue in healthcare, particularly in developed countries such as Australia. Estimating prognosis following admission remains a key challenge. Current predictive tools require numerous patient input features including those unavailable early in admission. Moreover, attempts to explain machine learning [ML]-based predictions are lacking. We developed 7 ML prognostication models to predict in-hospital mortality following minimal trauma HF in those aged [≥] 65 years of age, requiring only sociodemographic and comorbidity data as input. Hyperparameter tuning was performed via fractional factorial design of experiments combined with grid search; models were evaluated with 5-fold cross-validation and area under the receiver operating characteristic curve (AUROC). For explainability, ML models were directly interpreted as well as analyzed with SHAP values. Top performing models were random forests, naive Bayes [NB], extreme gradient boosting, and logistic regression (AUROCs ranging 0.682 - 0.696, p>0.05). Interpretation of models found the most important features were chronic kidney disease, cardiovascular comorbidities and markers of bone metabolism; NB also offers direct intuitive interpretation. Overall, we conclude that NB has much potential as an algorithm, due to its simplicity and interpretability whilst maintaining competitive predictive performance. Author SummaryOsteoporotic hip fractures are a critical health issue in developed countries. Preventative measures have ameliorated this issue somewhat, but the problem is expected to remain in main due to the aging population. Moreover, the mortality rate of patients in-hospital remains unacceptably high, with estimates ranging from 5-10%. Thus, a risk stratification tool would play a critical in optimizing care by facilitating the identification of the susceptible elderly in the community for prevention measures and the prioritisation of such patients early during their hospital admission. Unfortunately, such a tool has thus far remained elusive, despite forays into relatively exotic algorithms in machine learning. There are three major drawbacks (1) most tools all rely on information typically unavailable in the community and early during admission (for example, intra-operative data), limiting their potential use in practice, (2) few studies compare their trained models with other potential algorithms and (3) machine learning models are commonly cited as being black boxes and uninterpretable. Here we show that a Naive Bayes model, trained using only sociodemographic and comorbidity data of patients, performs on par with the more popular methods lauded in literature. The model is interpretable through direct analysis; the comorbidities of chronic kidney disease, cardiovascular, and bone metabolism were identified as being important features contributing to the likelihood of deaths. We also showcase an algorithm-agnostic approach to machine learning model interpretation. Our study shows the potential for Naive Bayes in predicting elderly patients at risk of death during an admission for hip fracture.

Osteoporosis is a significant health and economic issue, as it predisposes patients to a higher risk of bone fracture. Measuring bone mineral density has been shown to be an accurate way to assess the risk for osteoporosis. The most common way for bone density testing is a dual-energy X-ray absorptiometry (DEXA) scan, which may be recommended for patients with increased risk of osteoporosis. Radiograph imaging is widely available in clinical settings and acquired for many reasons, such as trauma or pain. The goal of this project is to extract radiomics information from pelvic X-rays (both the hip and femoral neck regions) to assess the risk of osteoporosis (triaging patients into "normal" vs. "at-risk", or "low risk" vs. "high risk" categories). The motivation here is not to replace the DEXA scan but to proactively identify patients at risk for osteoporosis and appropriately refer them to management options. We apply machine learning-based radiomics techniques on a study cohort of 565 patients. Our preliminary results show that a correlation between the radiomics features extracted from pelvic X-rays and the level of osteoporosis risk derived from the DEXA test results.

ObjectiveTo present Motiro, an unified framework for non-supervised statistical analysis endomicroscopy videos of the colorectal mucosa. Materials and MethodsWe wrote an open-source Python wrapper using ImageJ software with OpenCV, Seaborn and NumPy libraries. It generates a mosaic from the video of the mucosa, evaluates morphometric properties of the crypts, their distribution, and return their statistics. Shannon entropy (and Hellinger distance) are used for quantifying variability (and comparing different mucosa). ResultsThe segmentation process applied to normal mucosa of pre(post)- neoadjuvant patient is presented along with the corresponding statistical analysis of morphometric parameters. DiscussionOur analysis provides estimation of morphometric parameters consistent with available methods, is faster, and, additionally, provides statistical characterization of the mucosa morphometry. Motiro enables the analysis of large amounts of endomicroscopy videos for building a normal rectum features dataset to help on: detection of small variability; classification of post-neoadjuvant recovery; decision about surgical intervention necessity.

Arterial blood pressure (ABP) and photoplethysmography (PPG) waveforms both contain vital physiological information for the prevention and treatment of cardiovascular diseases. Extracted features from these waveforms have diverse clinical applications, including predicting hyper- and hypo-tension, estimating cardiac output from ABP, and monitoring blood pressure and nociception from PPG. However, the lack of standardized tools for feature extraction limits their exploration and clinical utilization. In this study, we propose an automatic feature extraction tool that first detects temporal location of landmarks within each cardiac cycle of ABP and PPG waveforms, including the systolic phase onset, systolic phase peak, dicrotic notch, and diastolic phase peak using the iterative envelope mean method. Then, based on these landmarks, extracts 852 features per cardiac cycle, encompassing time-, statistical-, and frequency-domains. The tools ability to detect landmarks was evaluated using ABP and PPG waveforms from a large perioperative dataset (MLORD dataset) comprising 17,327 patients. We analyzed 34,267 cardiac cycles of ABP waveforms and 33,792 cardiac cycles of PPG waveforms. Additionally, to assess the tools real-time landmark detection capability, we retrospectively analyzed 3,000 cardiac cycles of both ABP and PPG waveforms, collected from a Philips IntelliVue MX800 patient monitor. The tools detection performance was assessed against markings by an experienced researcher, achieving average F1-scores and error rates for ABP and PPG as follows: (1) On MLORD dataset: systolic phase onset (99.77 %, 0.35 % and 99.52 %, 0.75 %), systolic phase peak (99.80 %, 0.30 % and 99.56 %, 0.70 %), dicrotic notch (98.24 %, 2.63 % and 98.72 %, 1.96 %), and diastolic phase peak (98.59 %, 2.11 % and 98.88 %, 1.73 %); (2) On real time data: systolic phase onset (98.18 %, 3.03 % and 97.94 %, 3.43 %), systolic phase peak (98.22 %, 2.97 % and 97.74 %, 3.77 %), dicrotic notch (97.72 %, 3.80 % and 98.16 %, 3.07 %), and diastolic phase peak (98.04 %, 3.27 % and 98.08 %, 3.20 %). This tool has significant potential for supporting clinical utilization of ABP and PPG waveform features and for facilitating feature-based machine learning models for various clinical applications where features derived from these waveforms play a critical role.

IntroductionTo propose an improved U-Net-based segmentation model for colorectal polyp segmentation, aiming to address the challenges of variable lesion morphology, ambiguous boundaries, complex background interference, and insufficient cross-level feature fusion in endoscopic images [5,12]. MethodsAn improved network termed MCA-UNet was developed based on U-Net [5]. The model incorporates a multi-scale context convolution block (MCCB) to enhance multi-scale feature extraction and an attention-guided feature fusion module (AGFF) to optimize skip-feature selection and fusion in the decoder. Experiments were conducted on publicly available colorectal polyp image datasets, including Kvasir-SEG and CVC-ClinicDB [13-15]. Four models, including U-Net, U-Net+MCCB, U-Net+AGFF, and MCA-UNet, were compared, and all models were trained for 100 epochs. Dice, intersection over union (IoU), and mean absolute error (MAE) were used as the main evaluation metrics [20]. ResultsOn the mixed validation set, the Dice scores of U-Net, U-Net+MCCB, U-Net+AGFF, and MCA-UNet were 0.742, 0.771, 0.754, and 0.783, respectively; the corresponding IoU values were 0.603, 0.635, 0.618, and 0.649; and the MAE values were 0.102, 0.090, 0.097, and 0.086. Compared with the baseline U-Net, MCA-UNet improved Dice and IoU by 5.53% and 7.63%, respectively, while reducing MAE by 15.69%. Comparisons on the Kvasir-SEG and CVC-ClinicDB validation subsets further demonstrated the more stable performance of the proposed model. ConclusionBy jointly integrating multi-scale contextual modeling and attention-guided feature fusion, MCA-UNet effectively improves the accuracy and robustness of colorectal polyp segmentation and may provide useful support for intelligent endoscopic image analysis [12,17,18].

Real-world studies based on electronic health records often require manual chart review to derive patients clinical phenotypes, a labor-intensive task with limited scalability. Here, we developed and compared computable phenotyping based on rules using the spaCy frame-work and a Large Language Model (LLM), GPT-4, for disease behavior and age at diagnosis of Crohns disease patients. We are the first to describe computable phenotyping algorithms using clinical texts for these complex tasks with previously described inter-annotator agreements between 0.54 and 0.98. The data comprised clinical notes and radiology reports from 584 Mount Sinai Health System patients. Overall, we observed similar or better performance using GPT-4 compared to the rules. On a note-level, the F1 score was at least 0.90 for disease behavior and 0.82 for age at diagnosis. We could not find statistical evidence for a difference to the performance of human experts on this task. Our findings underline the potential of LLMs for computable phenotyping. Graphical Abstract O_FIG O_LINKSMALLFIG WIDTH=200 HEIGHT=57 SRC="FIGDIR/small/23297099v2_ufig1.gif" ALT="Figure 1"> View larger version (20K): org.highwire.dtl.DTLVardef@20c846org.highwire.dtl.DTLVardef@3c92b5org.highwire.dtl.DTLVardef@c3e8cborg.highwire.dtl.DTLVardef@1e89f36_HPS_FORMAT_FIGEXP M_FIG C_FIG

3D Slicer, an open-source software platform for the analysis and 3D visualization of medical imaging data, aims to make cutting-edge research tools available to clinical researchers and scientists worldwide. Until recently, the platform was only available in English. This study describes the development of an ad hoc methodology that addresses key linguistic challenges, including domain-specific vocabulary, acronyms, word order, passive voice, syntagms, and adaptation of technical terms. The translation process focuses on ensuring textual uniformity, cohesion, and accuracy while minimizing errors in biomedical computational contexts. The methodology presented may serve as a framework for similar translation efforts in diverse non-English speaking countries. Author summaryPaulo Eduardo de Barros Veiga, born in Ribeirao Preto, Sao Paulo (Brazil), holds degrees in Music, Language, and Literature, focusing on Criticism and Translation. He completed a postdoctoral fellowship (Process No. 2018/01418-2, Sao Paulo Research Foundation - FAPESP). He served as a collaborator and temporary professor in the Department of Music at FFCLRP, University of Sao Paulo, where he worked on the History and Philosophy of Art. Currently, he is involved in a research project within the Department of Computing and Mathematics at the same university, focusing on developing a biomedical imaging data platform (3D Slicer Software). Under the coordination of Prof. Sonia Pujol (Harvard University) and Prof. Luiz Murta (USP), he leads the translation of the 3D Slicer software into Brazilian Portuguese and proposes solutions and methods in Portuguese. He also engages in music, translation, research, and education projects freelance. Additionally, he is a member of the Poesis Critica research group under NAPI-CIPEM. Paulo holds a bachelors, masters, and doctoral degree in Literary Studies from the FCLAr at UNESP (Sao Paulo State University). He received a CAPES scholarship during his Masters and was awarded the Emerging Leaders in the Americas Program by the Canadian government, having studied at the University of Winnipeg. He has also lived in London.

ProKnow is an archive and restore tool for radiation oncology and imaging data, peer review, distributed contouring, study of metrics and discovery of trends; it represents a common ground for plan analysis and comparison due to an integrated industry standard Dose-Volume Histogram (DVH) engine, which can be used for all patient datasets. The aim of this work is to present a deep and easy-to-use implementation of the Elekta ProKnow DS cloud-based Picture Archiving and Communications in Radiotherapy (RT-PACS) system within our department. Two de-identification workflows of the DICOM data are presented, the first one is accomplished via the ProKnow Dicom Agent (PDA) and the second one involves a trusted third-party service. We can access ProKnow not only through the user interface, but also through the Application Programming Interface (API) with scripts written in the Python language to extract information from the uploaded data, calculate and store metrics as well as upload clinical data. We used ProKnow for a retrospective feasibility study of an isotoxic dose-escalated radiotherapy concept for glioblastoma. Furthermore, to ensure protocol-compliant irradiation planning for the preparation of a prospective dose-escalation trial, we conducted a dummy run with 10 collaborating institutes in Germany. RT-structures were automatically downloaded (via the API) and the Dice Score and Hausdorff Distance were calculated and set as metric in ProKnow. A drawback of the currently implemented de-identification process is that in a subsequent clinical data upload, matching the original and de-identified IDs is not possible. We therefore collaborate with the MeDICLMU(Data Integration Center [DIC]) for development and implementation of an automated de-identification process via a trusted third party service. With this architecture, it will be possible to merge clinical data in local DIC databases with de-identified data in ProKnow at any point in time. Author SummaryUntil now, it has been difficult to make scientific comparisons of treatments in radiation oncology because complex dose-volume distributions have to be compared rather than point doses at an isocentre, as is often thought. For the first time, ProKnow software enables the manufacturer-independent comparison of dose-volume distributions in tumour and normal tissue for large cohorts. With the above-described integration, multicentre research with state-of-the-art, secure, cloud-based data flows becomes easy to use.

Artificial Intelligence (AI) is a rapidly progressing technology with its applications expanding exponentially over the past decade. While initial breakthroughs predominantly focused on deep learning and computer vision, recent advancements have facilitated a shift towards natural language processing and beyond. This includes generative models, like ChatGPT, capable of understanding the grammar of software code, analog signals, and molecular structures. This research undertakes a comprehensive examination of AI trends within the biomedical domain, including the impact of ChatGPT. We explore scientific literature, clinical trials, and FDA-approval data, utilizing a thematic synthesis approach and bibliometric mapping of keywords to examine numerous subsets from over a hundred thousand unique records found in prominent public repositories up to mid-July 2023. Our analysis reveals a higher prevalence of general health-related publications compared to more specialized papers using or evaluating ChatGPT. However, the growth in specialized papers suggests a convergence with the trend observed for other AI tools. Our findings also imply a greater prevalence of publications using ChatGPT across multiple medical specialties compared to other AI tools, indicating its rising influence in complex fields requiring interdisciplinary collaboration. Leading topics in AI literature include radiology, ethics, drug discovery, COVID-19, robotics, brain research, stroke, and laparoscopy, indicating a shift from laboratory to emergency medicine and deep-learning-based image processing. Publications involving ChatGPT predominantly address current themes such as COVID-19, practical applications, interdisciplinary collaboration, and risk mitigation. Radiology retains dominance across all stages of biomedical R&D, spanning preprints, peer-reviewed papers, clinical trials, patents, and FDA approvals. Meanwhile, surgery-focused papers appear more frequently within ChatGPT preprints and case reports. Traditionally less represented areas, such as Pediatrics, Otolaryngology, and Internal Medicine, are starting to realize the benefits of ChatGPT, hinting at its potential to spark innovation within new medical sectors. AI application in geriatrics is notably underrepresented in publications. However, ongoing clinical trials are already exploring the use of ChatGPT for managing age-related conditions. The higher frequency of general health-related publications compared to specialized papers employing or evaluating ChatGPT showcases its broad applicability across multiple fields. AI, particularly ChatGPT, possesses significant potential to reshape the future of medicine. With millions of papers published annually across various disciplines, efficiently navigating the information deluge to pinpoint valuable studies has become increasingly challenging. Consequently, AI methods, gaining in popularity, are poised to redefine the future of scientific publishing and its educational reach. Despite challenges like quality of training data and ethical concerns, prevalent in preceding AI tools, the wider applicability of ChatGPT across diverse fields is manifest. This review employed the PRISMA tool and numerous overlapping data sources to minimize bias risks.

Accurate assessment of consciousness during general anesthesia is crucial for optimizing anesthetic dosage and patient safety. Current electroencephalogram-based monitoring devices can be inaccurate or unreliable in specific surgical contexts (e.g. cephalic procedures). This study investigated the feasibility of using electrocardiogram (ECG) features and machine learning to differentiate between awake and anesthetized states. A cohort of 48 patients undergoing surgery under general anesthesia at the Tours hospital was recruited. ECG-derived features were extracted, including spectral power, heart rate variability and complexity metrics, as well as heart rate fragmentation indices (HRF). These features were augmented by a range of physiological variables. The aim was to evaluate a number of machine learning algorithms in order to identify the most appropriate method for classifying the awake and anesthetized states. The gradient boosting algorithm achieved the highest accuracy (0.84). Notably, HRF metrics exhibited the strongest predictive power across all models tested. The generalizability of this ECG-based approach was further assessed using public datasets (VitalDB, Fantasia, and MIT-BIH Polysomnographic), achieving accuracies above 0.80. This study provides evidence that ECG-based classification methods can effectively distinguish awake from anesthetized states, with HRF indices playing a pivotal role in this classification. Author summaryGeneral anesthesia monitoring is critical for optimizing patient safety and outcomes. While electroencephalogram (EEG)-based systems are commonly used, they have limitations in accuracy and applicability, particularly in cases where EEG electrodes placement is challenging or impossible, such as during cephalic surgeries or when patients have forehead skin lesions. Here, a novel approach using electrocardiogram (ECG) signals and machine learning techniques was used to differentiate between awake and anesthetized states during surgery. A total of 48 patients undergoing surgical procedures under general anaesthesia at the Tours hospital were selected for inclusion in the study. This investigation focused on heart rate fragmentation indices, metrics designed for assessing biological versus chronological age, derived from ECG recordings. The gradient boosting algorithm demonstrates performance comparable to leading methods reported in the literature for this classification task. Importantly, model generalizability was confirm through successful application to publicly available datasets. This article highlights the potential of ECG signals as an alternative source for deriving depth of anesthesia indices, offering increased versatility in clinical settings where EEG monitoring is challenging or contraindicated.

AO_SCPLOWBSTRACTC_SCPLOWThe early phase of acute myocardial ischemia is associated with an elevated risk of ventricular reentrant arrhythmias. Indeed, after partial or total occlusion of a coronary artery, some regions of the heart experience a reduction in myocardial blood flow. This causes metabolic and cellular processes, such as hypoxia, hyperkalemia and acidosis, which lead to changes in the transmembrane ionic dynamics. The effect of such alterations may result in the formation of electrical loops and reentries. Computational models could simulate the generation of arrhythmias, possibly persistent, in condition of ectopic beats and in presence of acute myocardial regions. Since quantitative information (extent, localization, ...) about acute ischemic regions are hardly available from clinics, to date, computational models only integrate imaging data from chronic infarcted ventricles. This may not accurately reflect the acute condition. This work presents a novel patient-specific electrophysiological model, based on images of myocardial blood flow maps acquired during a pharmacologically induced acute ischemic event. The model personalization is obtained with the partitioning of the left ventricle geometries on the basis of the myocardial blood flow maps. First, we aim to numerically investigate the induction and sustainment of reentrant drivers in patient-specific scenarios, in order to assess their arrhythmic propensity. Secondly, we perform an intra-patient sensitivity analysis, where different levels of acute ischemia are virtually depicted for the most arrhythmogenic patient. Our results suggest that the amount of ischemic regions seems to have less influence on arrhythmogenesis than their pattern.

The COVID-19 (SARS-CoV-2) pandemic is overwhelming global healthcare delivery systems due to the exponential spike in cases requiring specialty tests, facilities and equipment, including complex, precision devices like ventilators. In particular, the surge in critically ill patients has revealed a significant deficiency in regional availability of respiratory care ventilators. The authors offer a mathematical framework for ventilator distribution under scarcity conditions using an optimized network model and solver. The framework is interoperable with existing COVID-19 healthcare demand models and scales for different user-defined system sizes, including hospital networks, city, state, regional and national-scale prioritization. The authors approach improves current capabilities for medical device resource management within the existing incident command system while accounting for availability of devices, ventilation treatment time periods, disinfection and cleaning between patients, as well as shipping logistics time. The authors present a proof of concept using a high fidelity COVID-19 data set from Colorado, discusses how to scale nationally, and emphasizes the importance of applying ethical human-in-the-loop decision making when using this or similar approaches to managing medical device resources during epidemic emergencies.

AO_SCPLOWBSTRACTC_SCPLOWThe electrocardiogram (ECG) is an almost universally accessible diagnostic tool for heart disease. An ECG is measured by using an electrocardiograph, and todays electrocardiographs use built-in software to interpret the ECGs automatically after they are recorded. However, these algorithms show limited performance, and therefore clinicians usually have to manually interpret the ECG, regardless of whether an algorithm has interpreted the ECG or not. Manual interpretation of the ECG can be time-consuming and require specific skills. Therefore, a better algorithm is clearly needed to make correct ECG interpretations more accessible and time efficient. Algorithms based on artificial intelligence have shown promising performance in many fields, including ECG interpretation, over the last few years and might represent an alternative to manual ECG interpretation. In this study, we used a dataset with 88253 12-lead ECGs from multiple databases, annotated with SNOMED-CT codes by medical experts. We employed a supervised convolutional neural network with an Inception architecture to classify 30 of the most frequent annotated diagnoses in the dataset. Each patient could have more than one diagnosis, which makes this a multi-label classification. We compared the Inception models performance while applying different preprocessing methods on the ECGs and different model settings during 10-folded cross-validation. We compared the models classification performance using binary cross-entropy (BCE) loss and double soft F1 loss. Furthermore, we compared the classification performance when downsampling the original sampling rate of the input ECG. Finally, we trained 30 interpretable linear models to provide class activation maps to explain the relative importance of each sample in the ECG with respect to the 30 diagnoses considered in this study. Due to the heavily imbalanced class distribution in our dataset, we placed the most emphasis on the F1 score when evaluating the performance of the models. Our results show that the best performance in terms of F1-score was seen when the Inception model used double soft F1 as the loss function and ECGs downsampled to 75Hz. This model achieved an F1 score of 0.420 {+/-} 0.017, accuracy = 0.954 {+/-} 0.002, and an AUROC score of 0.832 {+/-} 0.019. An aggregation of the generated saliency maps, achieved using Local Interpretable Model-Agnostic Explanations (LIME), showed that the Inception model paid the most attention to the limb leads and the augmented leads and less importance to the precordial leads. One of the more significant contributions that emerge from this study is the use of aggregated saliency maps to obtain ECG lead importance for different diagnoses. In addition, we emphasized the relevance of evaluating different loss functions, and in this specific case, we found double soft F1 loss to be slightly better than BCE. Finally, we found it somewhat surprising that downsampling the ECG led to higher performance compared to the original 500Hz sampling rate. These findings contribute in several ways to our understanding of the artificial intelligence-based interpretation of ECGs, but further studies should be carried out to validate these findings in other datasets from other patient cohorts.

Anterior quadratus lumborum (QL) block is a regional anesthesia technique shown to provide both somatic and visceral pain relief by targeting lower thoracic nerves and the thoracic sympathetic trunk. Despite its clinical benefits, success depends on accurate sonoanatomic identification, which can be challenging due to individual anatomical variations. In this study, we developed an artificial intelligence (AI) model to automatically segment key sonoanatomic landmarks for the anterior QL block. A total of 82 ultrasound videos from 42 healthy volunteers yielded 460 labeled images capturing the vertebral body (L3/L4), posterior renal fascia, transverse abdominal muscle, quadratus lumborum muscle, psoas major muscle, and the injection point. We trained a 2D U-Net-based model (nnU-Net) with five-fold cross-validation. Training data was split into an 80% training set (368 images) and 20% validation set (92 images). The performance of the AI model was tested on images obtained from 20 patients receiving the anterior QL block as a part of standard treatment. The model achieved a moderate-high Dice score of 0.62 across six classes, with especially high segmentation performance for vertebral bodies (Dice 0.90) and the psoas major muscle (Dice 0.85). Low-moderate performance was observed for the posterior renal fascia (Dice 0.35) and the injection point (Dice 0.38), likely reflecting their subtle sonographic appearance. In conclusion, this is the first AI model that can delineate the sonoanatomy of the anterior QL block region. Our findings underscore the potential of AI to improve the precision and consistency of ultrasound-guided anterior QL blocks. KEY MESSAGESO_ST_ABSWhat is already known on this topicC_ST_ABSThe anterior quadratus lumborum (QL) block effectively reduces postoperative opioid consumption and pain but relies on precise sonoanatomic identification. The efficacy of the block has been concluded in a recent systematic review with meta-analyses. What this study addsIt presents the first AI model to segment key landmarks for the anterior QL block in a clinical setting. How this study might affect research, practice or policyAI-based segmentation may improve consistency, reduce operator dependence, and enhance beneficial patient outcomes.