Objectives: Early detection of sepsis is critical in clinical practice since each hour of delayed treatment has been associated with an increase in mortality due to irreversible organ damage. This study aimed to develop an explainable artificial intelligence model for early predicting sepsis by analyzing the electronic health record data from ICU provided by the PhysioNet/Computing in Cardiology Challenge 2019. Design: Retrospective observational study. Setting: We developed our model on the shared ICUs publicly data and verified on the full hidden populations for challenge scoring. Patients: Public database included 40,336 patients’ electronic health records sourced from Beth Israel Deaconess Medical Center (hospital system A) and Emory University Hospital (hospital system B). A total of 24,819 patients from hospital systems A, B, and C (an unidentified hospital system) were sequestered as full hidden test sets. Interventions: None. Measurements and Main Results: A total of 168 features were extracted on hourly basis. Explainable artificial intelligence sepsis predictor model was trained to predict sepsis in real time. Impact of each feature on hourly sepsis prediction was explored in-depth to show the interpretability. The algorithm demonstrated the final clinical utility score of 0.364 in this challenge when tested on the full hidden test sets, and the scores on three separate test sets were 0.430, 0.422, and –0.048, respectively. Conclusions: Explainable artificial intelligence sepsis predictor model achieves superior performance for predicting sepsis risk in a real-time way and provides interpretable information for understanding sepsis risk in ICU.

A systematical evaluation work was performed on ten widely used and high-efficient QRS detection algorithms in this study, aiming at verifying their performances and usefulness in different application situations. Four experiments were carried on six internationally recognized databases. Firstly, in the test of high-quality ECG database versus low-quality ECG database, for high signal quality database, all ten QRS detection algorithms had very high detection accuracy (F1 >99%), whereas the F1 results decrease significantly for the poor signal-quality ECG signals (all 95% except RS slope algorithm with 94.24% on normal ECG database and 94.44% on arrhythmia database). Thirdly, for the paced rhythm ECG database, all ten algorithms were immune to the paced beats (>94%) except the RS slope method, which only output a low F1 result of 78.99%. At last, the detection accuracies had obvious decreases when dealing with the dynamic telehealth ECG signals (all <80%) except OKB algorithm with 80.43%. Furthermore, the time costs from analyzing a 10 s ECG segment were given as the quantitative index of the computational complexity. All ten algorithms had high numerical efficiency (all <4 ms) except RS slope (94.07 ms) and sixth power algorithms (8.25 ms). And OKB algorithm had the highest numerical efficiency (1.54 ms).

Heartbeat classification is a crucial step for arrhythmia diagnosis during electrocardiographic (ECG) analysis. The new scenario of wireless body sensor network- (WBSN-) enabled ECG monitoring puts forward a higher-level demand for this traditional ECG analysis task. Previously reported methods mainly addressed this requirement with the applications of a shallow structured classifier and expert-designed features. In this study, modified frequency slice wavelet transform (MFSWT) was firstly employed to produce the time-frequency image for heartbeat signal. Then the deep learning (DL) method was performed for the heartbeat classification. Here, we proposed a novel model incorporating automatic feature abstraction and a deep neural network (DNN) classifier. Features were automatically abstracted by the stacked denoising auto-encoder (SDA) from the transferred time-frequency image. DNN classifier was constructed by an encoder layer of SDA and a softmax layer. In addition, a deterministic patient-specific heartbeat classifier was achieved by fine-tuning on heartbeat samples, which included a small subset of individual samples. The performance of the proposed model was evaluated on the MIT-BIH arrhythmia database. Results showed that an overall accuracy of 97.5% was achieved using the proposed model, confirming that the proposed DNN model is a powerful tool for heartbeat pattern recognition.

Automatic feature extraction and classification are two main tasks in abnormal ECG beat recognition. Feature extraction is an important prerequisite prior to classification since it provides the classifier with input features, and the performance of classifier depends significantly on the quality of these features. This study develops an effective method to extract low-dimensional ECG beat feature vectors. It employs wavelet multi-resolution analysis to extract time-frequency domain features and then applies principle component analysis to reduce the dimension of the feature vector. In classification, 12-element feature vectors characterizing six types of beats are used as inputs for one-versus-one support vector machine, which is conducted in form of 10-fold cross validation with beat-based and record-based training schemes. Tested upon a total of 107049 beats from MIT-BIH arrhythmia database, our method has achieved average sensitivity, specificity and accuracy of 99.09%, 99.82% and 99.70%, respectively, using the beat-based training scheme, and 44.40%, 88.88% and 81.47%, respectively, using the record-based training scheme.

R-peak detection is crucial in electrocardiogram (ECG) signal analysis. This study proposed an adaptive and time-efficient R-peak detection algorithm for ECG processing. First, wavelet multiresolution analysis was applied to enhance the ECG signal representation. Then, ECG was mirrored to convert large negative R-peaks to positive ones. After that, local maximums were calculated by the first-order forward differential approach and were truncated by the amplitude and time interval thresholds to locate the R-peaks. The algorithm performances, including detection accuracy and time consumption, were tested on the MIT-BIH arrhythmia database and the QT database. Experimental results showed that the proposed algorithm achieved mean sensitivity of 99.39%, positive predictivity of 99.49%, and accuracy of 98.89% on the MIT-BIH arrhythmia database and 99.83%, 99.90%, and 99.73%, respectively, on the QT database. By processing one ECG record, the mean time consumptions were 0.872 s and 0.763 s for the MIT-BIH arrhythmia database and QT database, respectively, yielding 30.6% and 32.9% of time reduction compared to the traditional Pan-Tompkins method.

Muscle synergies have been proposed as a strategy employed by the central nervous system to control movements. Muscle synergy analysis is a well-established framework to examine the pathophysiological basis of neurological diseases and has been applied for analysis and assessment in clinical applications in the last decades, even if it has not yet been widely used in clinical diagnosis, rehabilitative treatment and interventions. Even if inconsistencies in the outputs among studies and lack of a normative pipeline including signal processing and synergy analysis limit the progress, common findings and results are identifiable as a basis for future research. Therefore, a literature review that summarizes methods and main findings of previous works on upper limb muscle synergies in clinical environment is needed to i) summarize the main findings so far, ii) highlight the barriers limiting their use in clinical applications, and iii) suggest future research directions needed for facilitating translation of experimental research to clinical scenarios. Articles in which muscle synergies were used to analyze and assess upper limb function in neurological impairments were reviewed. The literature research was conducted in Scopus, PubMed, and Web of Science. Experimental protocols (e.g., the aim of the study, number and type of participants, number and type of muscles, and tasks), methods (e.g., muscle synergy models and synergy extraction methods, signal processing methods), and the main findings of eligible studies were reported and discussed. 383 articles were screened and 51 were selected, which involved a total of 13 diseases and 748 patients and 1155 participants. Each study investigated on average 15 ± 10 patients. Four to forty-one muscles were included in the muscle synergy analysis. Point-to-point reaching was the most used task. The preprocessing of EMG signals and algorithms for synergy extraction varied among studies, and non-negative matrix factorization was the most used method. Five EMG normalization methods and five methods for identifying the optimal number of synergies were used in the selected papers. Most of the studies report that analyses on synergy number, structure, and activations provide novel insights on the physiopathology of motor control that cannot be gained with standard clinical assessments, and suggest that muscle synergies may be useful to personalize therapies and to develop new therapeutic strategies. However, in the selected studies synergies were used only for assessment; different testing procedures were used and, in general, study-specific modifications of muscle synergies were observed; single session or longitudinal studies mainly aimed at assessing stroke (71% of the studies), even though other pathologies were also investigated. Synergy modifications were either study-specific or were not observed, with few analyses available for temporal coefficients. Thus, several barriers prevent wider adoption of muscle synergy analysis including a lack of standardized experimental protocols, signal processing procedures, and synergy extraction methods. A compromise in the design of the studies must be found to combine the systematicity of motor control studies and the feasibility of clinical studies. There are however several potential developments that might promote the use of muscle synergy analysis in clinical practice, including refined assessments based on synergistic approaches not allowed by other methods and the availability of novel models. Finally, neural substrates of muscle synergies are discussed, and possible future research directions are proposed. This review provides new perspectives about the challenges and open issues that need to be addressed in future work to achieve a better understanding of motor impairments and rehabilitative therapy using muscle synergies. These include the application of the methods on wider scales, standardization of procedures, inclusion of synergies in the clinical decisional process, assessment of temporal coefficients and temporal-based models, extensive work on the algorithms and understanding of the physio-pathological mechanisms of pathology, as well as the application and adaptation of synergy-based approaches to various rehabilitative scenarios for increasing the available evidence.