Prediction of Complications and Prognostication in Perioperative Medicine: A Systematic Review and PROBAST Assessment of Machine Learning Tools.

BACKGROUND: The utilization of artificial intelligence and machine learning as diagnostic and predictive tools in perioperative medicine holds great promise. Indeed, many studies have been performed in recent years to explore the potential. The purpose of this systematic review is to assess the current state of machine learning in perioperative medicine, its utility in prediction of complications and prognostication, and limitations related to bias and validation. METHODS: A multidisciplinary team of clinicians and engineers conducted a systematic review using the Preferred Reporting Items for Systematic Review and Meta-Analysis (PRISMA) protocol. Multiple databases were searched, including Scopus, Cumulative Index to Nursing and Allied Health Literature (CINAHL), the Cochrane Library, PubMed, Medline, Embase, and Web of Science. The systematic review focused on study design, type of machine learning model used, validation techniques applied, and reported model performance on prediction of complications and prognostication. This review further classified outcomes and machine learning applications using an ad hoc classification system. The Prediction model Risk Of Bias Assessment Tool (PROBAST) was used to assess risk of bias and applicability of the studies. RESULTS: A total of 103 studies were identified. The models reported in the literature were primarily based on single-center validations (75%), with only 13% being externally validated across multiple centers. Most of the mortality models demonstrated a limited ability to discriminate and classify effectively. The PROBAST assessment indicated a high risk of systematic errors in predicted outcomes and artificial intelligence or machine learning applications. CONCLUSIONS: The findings indicate that the development of this field is still in its early stages. This systematic review indicates that application of machine learning in perioperative medicine is still at an early stage. While many studies suggest potential utility, several key challenges must be first overcome before their introduction into clinical practice.

A Deep Learning Approach to Classify Surgical Skill in Microsurgery Using Force Data from a Novel Sensorised Surgical Glove.

Microsurgery serves as the foundation for numerous operative procedures. Given its highly technical nature, the assessment of surgical skill becomes an essential component of clinical practice and microsurgery education. The interaction forces between surgical tools and tissues play a pivotal role in surgical success, making them a valuable indicator of surgical skill. In this study, we employ six distinct deep learning architectures (LSTM, GRU, Bi-LSTM, CLDNN, TCN, Transformer) specifically designed for the classification of surgical skill levels. We use force data obtained from a novel sensorized surgical glove utilized during a microsurgical task. To enhance the performance of our models, we propose six data augmentation techniques. The proposed frameworks are accompanied by a comprehensive analysis, both quantitative and qualitative, including experiments conducted with two cross-validation schemes and interpretable visualizations of the network's decision-making process. Our experimental results show that CLDNN and TCN are the top-performing models, achieving impressive accuracy rates of 96.16% and 97.45%, respectively. This not only underscores the effectiveness of our proposed architectures, but also serves as compelling evidence that the force data obtained through the sensorized surgical glove contains valuable information regarding surgical skill.

An investigation into augmentation and preprocessing for optimising X-ray classification in limited datasets: a case study on necrotising enterocolitis.

PURPOSE: Obtaining large volumes of medical images, required for deep learning development, can be challenging in rare pathologies. Image augmentation and preprocessing offer viable solutions. This work explores the case of necrotising enterocolitis (NEC), a rare but life-threatening condition affecting premature neonates, with challenging radiological diagnosis. We investigate data augmentation and preprocessing techniques and propose two optimised pipelines for developing reliable computer-aided diagnosis models on a limited NEC dataset. METHODS: We present a NEC dataset of 1090 Abdominal X-rays (AXRs) from 364 patients and investigate the effect of geometric augmentations, colour scheme augmentations and their combination for NEC classification based on the ResNet-50 backbone. We introduce two pipelines based on colour contrast and edge enhancement, to increase the visibility of subtle, difficult-to-identify, critical NEC findings on AXRs and achieve robust accuracy in a challenging three-class NEC classification task. RESULTS: Our results show that geometric augmentations improve performance, with Translation achieving +6.2%, while Flipping and Occlusion decrease performance. Colour augmentations, like Equalisation, yield modest improvements. The proposed Pr-1 and Pr-2 pipelines enhance model accuracy by +2.4% and +1.7%, respectively. Combining Pr-1/Pr-2 with geometric augmentation, we achieve a maximum performance increase of 7.1%, achieving robust NEC classification. CONCLUSION: Based on an extensive validation of preprocessing and augmentation techniques, our work showcases the previously unreported potential of image preprocessing in AXR classification tasks with limited datasets. Our findings can be extended to other medical tasks for designing reliable classifier models with limited X-ray datasets. Ultimately, we also provide a benchmark for automated NEC detection and classification from AXRs.

Machine Learning Analysis of Post-Operative Tumour Progression in Non-Functioning Pituitary Neuroendocrine Tumours: A Pilot Study.

Post-operative tumour progression in patients with non-functioning pituitary neuroendocrine tumours is variable. The aim of this study was to use machine learning (ML) models to improve the prediction of post-operative outcomes in patients with NF PitNET. We studied data from 383 patients who underwent surgery with or without radiotherapy, with a follow-up period between 6 months and 15 years. ML models, including k-nearest neighbour (KNN), support vector machine (SVM), and decision tree, showed superior performance in predicting tumour progression when compared with parametric statistical modelling using logistic regression, with SVM achieving the highest performance. The strongest predictor of tumour progression was the extent of surgical resection, with patient age, tumour volume, and the use of radiotherapy also showing influence. No features showed an association with tumour recurrence following a complete resection. In conclusion, this study demonstrates the potential of ML models in predicting post-operative outcomes for patients with NF PitNET. Future work should look to include additional, more granular, multicentre data, including incorporating imaging and operative video data.

Post-Operative Medium- and Long-Term Endocrine Outcomes in Patients with Non-Functioning Pituitary Adenomas-Machine Learning Analysis.

Post-operative endocrine outcomes in patients with non-functioning pituitary adenoma (NFPA) are variable. The aim of this study was to use machine learning (ML) models to better predict medium- and long-term post-operative hypopituitarism in patients with NFPAs. We included data from 383 patients who underwent surgery with or without radiotherapy for NFPAs, with a follow-up period between 6 months and 15 years. ML models, including k-nearest neighbour (KNN), support vector machine (SVM), and decision tree models, showed a superior ability to predict panhypopituitarism compared with non-parametric statistical modelling (mean accuracy: 0.89; mean AUC-ROC: 0.79), with SVM achieving the highest performance (mean accuracy: 0.94; mean AUC-ROC: 0.88). Pre-operative endocrine function was the strongest feature for predicting panhypopituitarism within 1 year post-operatively, while endocrine outcomes at 1 year post-operatively supported strong predictions of panhypopituitarism at 5 and 10 years post-operatively. Other features found to contribute to panhypopituitarism prediction were age, volume of tumour, and the use of radiotherapy. In conclusion, our study demonstrates that ML models show potential in predicting post-operative panhypopituitarism in the medium and long term in patients with NFPM. Future work will include incorporating additional, more granular data, including imaging and operative video data, across multiple centres.

Shifted-windows transformers for the detection of cerebral aneurysms in microsurgery.

PURPOSE: Microsurgical Aneurysm Clipping Surgery (MACS) carries a high risk for intraoperative aneurysm rupture. Automated recognition of instances when the aneurysm is exposed in the surgical video would be a valuable reference point for neuronavigation, indicating phase transitioning and more importantly designating moments of high risk for rupture. This article introduces the MACS dataset containing 16 surgical videos with frame-level expert annotations and proposes a learning methodology for surgical scene understanding identifying video frames with the aneurysm present in the operating microscope's field-of-view. METHODS: Despite the dataset imbalance (80% no presence, 20% presence) and developed without explicit annotations, we demonstrate the applicability of Transformer-based deep learning architectures (MACSSwin-T, vidMACSSwin-T) to detect the aneurysm and classify MACS frames accordingly. We evaluate the proposed models in multiple-fold cross-validation experiments with independent sets and in an unseen set of 15 images against 10 human experts (neurosurgeons). RESULTS: Average (across folds) accuracy of 80.8% (range 78.5-82.4%) and 87.1% (range 85.1-91.3%) is obtained for the image- and video-level approach, respectively, demonstrating that the models effectively learn the classification task. Qualitative evaluation of the models' class activation maps shows these to be localized on the aneurysm's actual location. Depending on the decision threshold, MACSWin-T achieves 66.7-86.7% accuracy in the unseen images, compared to 82% of human raters, with moderate to strong correlation. CONCLUSIONS: Proposed architectures show robust performance and with an adjusted threshold promoting detection of the underrepresented (aneurysm presence) class, comparable to human expert accuracy. Our work represents the first step towards landmark detection in MACS with the aim to inform surgical teams to attend to high-risk moments, taking precautionary measures to avoid rupturing.

Catheter manipulation analysis for objective performance and technical skills assessment in transcatheter aortic valve implantation.

PURPOSE: Transcatheter aortic valve implantation (TAVI) demands precise and efficient handling of surgical instruments within the confines of the aortic anatomy. Operational performance and dexterous skills are critical for patient safety, and objective methods are assessed with a number of manipulation features, derived from the kinematic analysis of the catheter/guidewire in fluoroscopy video sequences. METHODS: A silicon phantom model of a type I aortic arch was used for this study. Twelve endovascular surgeons, divided into two experience groups, experts ([Formula: see text]) and novices ([Formula: see text]), performed cannulation of the aorta, representative of valve placement in TAVI. Each participant completed two TAVI experiments, one with conventional catheters and one with the Magellan robotic platform. Video sequences of the fluoroscopic monitor were recorded for procedural processing. A semi-automated tracking software provided the 2D coordinates of the catheter/guidewire tip. In addition, the aorta phantom was segmented in the videos and the shape of the entire catheter was manually annotated in a subset of the available video frames using crowdsourcing. The TAVI procedure was divided into two stages, and various metrics, representative of the catheter's overall navigation as well as its relative movement to the vessel wall, were developed. RESULTS: Experts consistently exhibited lower values of procedure time and dimensionless jerk, and higher average speed and acceleration than novices. Robotic navigation resulted in increased average distance to the vessel wall in both groups, a surrogate measure of safety and reduced risk of embolisation. Discrimination of experience level and types of equipment was achieved with the generated motion features and established clustering algorithms. CONCLUSIONS: Evaluation of surgical skills is possible through the analysis of the catheter/guidewire motion pattern. The use of robotic endovascular platforms seems to enable more precise and controlled catheter navigation.

Detecting Elementary Arm Movements by Tracking Upper Limb Joint Angles With MARG Sensors.

This paper reports an algorithm for the detection of three elementary upper limb movements, i.e., reach and retrieve, bend the arm at the elbow and rotation of the arm about the long axis. We employ two MARG sensors, attached at the elbow and wrist, from which the kinematic properties (joint angles, position) of the upper arm and forearm are calculated through data fusion using a quaternion-based gradient-descent method and a two-link model of the upper limb. By studying the kinematic patterns of the three movements on a small dataset, we derive discriminative features that are indicative of each movement; these are then used to formulate the proposed detection algorithm. Our novel approach of employing the joint angles and position to discriminate the three fundamental movements was evaluated in a series of experiments with 22 volunteers who participated in the study: 18 healthy subjects and four stroke survivors. In a controlled experiment, each volunteer was instructed to perform each movement a number of times. This was complimented by a seminaturalistic experiment where the volunteers performed the same movements as subtasks of an activity that emulated the preparation of a cup of tea. In the stroke survivors group, the overall detection accuracy for all three movements was 93.75% and 83.00%, for the controlled and seminaturalistic experiment, respectively. The performance was higher in the healthy group where 96.85% of the tasks in the controlled experiment and 89.69% in the seminaturalistic were detected correctly. Finally, the detection ratio remains close ( ±6%) to the average value, for different task durations further attesting to the algorithms robustness.

A statistical index for early diagnosis of ventricular arrhythmia from the trend analysis of ECG phase-portraits.

In this paper, we propose a novel statistical index for the early diagnosis of ventricular arrhythmia (VA) using the time delay phase-space reconstruction (PSR) technique, from the electrocardiogram (ECG) signal. Patients with two classes of fatal VA-with preceding ventricular premature beats (VPBs) and with no VPBs-have been analysed using extensive simulations. Three subclasses of VA with VPBs viz. ventricular tachycardia (VT), ventricular fibrillation (VF) and VT followed by VF are analyzed using the proposed technique. Measures of descriptive statistics like mean (µ), standard deviation (σ), coefficient of variation (CV = σ/µ), skewness (γ) and kurtosis (ß) in phase-space diagrams are studied for a sliding window of 10 beats of the ECG signal using the box-counting technique. Subsequently, a hybrid prediction index which is composed of a weighted sum of CV and kurtosis has been proposed for predicting the impending arrhythmia before its actual occurrence. The early diagnosis involves crossing the upper bound of a hybrid index which is capable of predicting an impending arrhythmia 356 ECG beats, on average (with 192 beats standard deviation) before its onset when tested with 32 VA patients (both with and without VPBs). The early diagnosis result is also verified using a leave one out cross-validation (LOOCV) scheme with 96.88% sensitivity, 100% specificity and 98.44% accuracy.

Effect of Trunk Support on Upper Extremity Function in People With Chronic Stroke and People Who Are Healthy.

BACKGROUND: Trunk control is thought to contribute to upper extremity (UE) function. However, this common assumption in neurorehabilitation has not been validated in clinical trials. OBJECTIVE: The study objectives were to investigate the effect of providing external trunk support on trunk control and UE function and to examine the relationship between trunk control and UE function in people with chronic stroke and people who were healthy. DESIGN: A cross-sectional study was conducted. METHODS: Twenty-five people with chronic stroke and 34 people who were healthy and matched for age and sex were recruited. Trunk control was assessed with the Trunk Impairment Scale (TIS), and UE impairment and UE function were assessed with the UE subsection of the Fugl-Meyer Assessment (FMA-UE) and the Streamlined Wolf Motor Function Test (SWMFT), respectively. The TIS and SWMFT were evaluated, with and without external trunk support; the FMA-UE was evaluated without trunk support. RESULTS: With trunk support, people with stroke showed improvement from 18 to 20 points on the TIS, a reduction in SWMFT performance times from 37.20 seconds to 35.37 seconds for the affected UE, and improvement from 3.3 points to 3.4 points on the SWMFT Functional Ability Scale for the function of the affected UE. With trunk support, the SWMFT performance time for people who were healthy was reduced from 1.61 seconds to 1.48 seconds for the dominant UE and from 1.71 seconds to 1.59 seconds for the nondominant UE. A significant moderate correlation was found between the TIS and the FMA-UE (r=.53) for people with stroke. LIMITATIONS: The limitations included a nonmasked assessor and a standardized height of the external trunk support. CONCLUSIONS: External trunk support improved trunk control in people with chronic stroke and had a statistically significant effect on UE function in both people with chronic stroke and people who were healthy. The findings suggest an association between trunk control and the UE when external trunk support was provided and support the hypothesis that lower trunk and lumbar stabilization provided by external support enables an improvement in the ability to use the UE for functional activities.

Prompt and accurate diagnosis of ventricular arrhythmias with a novel index based on phase space reconstruction of ECG.

AIM: To develop a statistical index based on the phase space reconstruction (PSR) of the electrocardiogram (ECG) for the accurate and timely diagnosis of ventricular tachycardia (VT) and ventricular fibrillation (VF). METHODS: Thirty-two ECGs with sinus rhythm (SR) and 32 ECGs with VT/VF were analyzed using the PSR technique. Firstly, the method of time delay embedding were employed with the insertion of delay "τ" in the original time-series X(t), which produces the Y(t)=X(t-τ). Afterwards, a PSR diagram was reconstructed by plotting Y(t) against X(t). The method of box counting was applied to analyze the behavior of the PSR trajectories. Measures as mean (µ), standard deviation (σ) and coefficient of variation (CV=σ/µ), kurtosis (ß) for the box counting of PSR diagrams were reported. RESULTS: During SR, CV was always <0.05, while with the onset of arrhythmia CV increased >0.05. A similar pattern was observed with ß, where <6 was considered as the cut-off point between SR and VT/VF. Therefore, the upper threshold for SR was considered CVth=0.05 and ßth<6. For optimisation of the accuracy, a new index (J) was proposed: J=wCVCVth+1-wßßth. During SR the upper limit of J was the value of 1. Furthermore CV, ß and J crossed the cut-off point timely before the onset of arrhythmia (average time: 4min 31s; SD: 2min 30s); allowing sufficient time for preventive therapy. CONCLUSION: The J index improved ECG utility for arrhythmia monitoring and detection utility, allowing the prompt and accurate diagnosis of ventricular arrhythmias.

Development of an automated updated Selvester QRS scoring system using SWT-based QRS fractionation detection and classification.

The Selvester score is an effective means for estimating the extent of myocardial scar in a patient from low-cost ECG recordings. Automation of such a system is deemed to help implementing low-cost high-volume screening mechanisms of scar in the primary care. This paper describes, for the first time to the best of our knowledge, an automated implementation of the updated Selvester scoring system for that purpose, where fractionated QRS morphologies and patterns are identified and classified using a novel stationary wavelet transform (SWT)-based fractionation detection algorithm. This stage informs the two principal steps of the updated Selvester scoring scheme--the confounder classification and the point awarding rules. The complete system is validated on 51 ECG records of patients detected with ischemic heart disease. Validation has been carried out using manually detected confounder classes and computation of the actual score by expert cardiologists as the ground truth. Our results show that as a stand-alone system it is able to classify different confounders with 94.1% accuracy whereas it exhibits 94% accuracy in computing the actual score. When coupled with our previously proposed automated ECG delineation algorithm, that provides the input ECG parameters, the overall system shows 90% accuracy in confounder classification and 92% accuracy in computing the actual score and thereby showing comparable performance to the stand-alone system proposed here, with the added advantage of complete automated analysis without any human intervention.

A low-complexity ECG feature extraction algorithm for mobile healthcare applications.

This paper introduces a low-complexity algorithm for the extraction of the fiducial points from the Electrocardiogram (ECG). The application area we consider is that of remote cardiovascular monitoring, where continuous sensing and processing takes place in low-power, computationally constrained devices, thus the power consumption and complexity of the processing algorithms should remain at a minimum level. Under this context, we choose to employ the Discrete Wavelet Transform (DWT) with the Haar function being the mother wavelet, as our principal analysis method. From the modulus-maxima analysis on the DWT coefficients, an approximation of the ECG fiducial points is extracted. These initial findings are complimented with a refinement stage, based on the time-domain morphological properties of the ECG, which alleviates the decreased temporal resolution of the DWT. The resulting algorithm is a hybrid scheme of time and frequency domain signal processing. Feature extraction results from 27 ECG signals from QTDB, were tested against manual annotations and used to compare our approach against the state-of-the art ECG delineators. In addition, 450 signals from the 15-lead PTBDB are used to evaluate the obtained performance against the CSE tolerance limits. Our findings indicate that all but one CSE limits are satisfied. This level of performance combined with a complexity analysis, where the upper bound of the proposed algorithm, in terms of arithmetic operations, is calculated as 2:423N + 214 additions and 1:093N + 12 multiplications for N 861 or 2:553N + 102 additions and 1:093N +10 multiplications for N > 861 (N being the number of input samples), reveals that the proposed method achieves an ideal trade-off between computational complexity and performance, a key requirement in remote CVD monitoring systems.

An automated algorithm for online detection of fragmented QRS and identification of its various morphologies.

Fragmented QRS (f-QRS) has been proven to be an efficient biomarker for several diseases, including remote and acute myocardial infarction, cardiac sarcoidosis, non-ischaemic cardiomyopathy, etc. It has also been shown to have higher sensitivity and/or specificity values than the conventional markers (e.g. Q-wave, ST-elevation, etc.) which may even regress or disappear with time. Patients with such diseases have to undergo expensive and sometimes invasive tests for diagnosis. Automated detection of f-QRS followed by identification of its various morphologies in addition to the conventional ECG feature (e.g. P, QRS, T amplitude and duration, etc.) extraction will lead to a more reliable diagnosis, therapy and disease prognosis than the state-of-the-art approaches and thereby will be of significant clinical importance for both hospital-based and emerging remote health monitoring environments as well as for implanted ICD devices. An automated algorithm for detection of f-QRS from the ECG and identification of its various morphologies is proposed in this work which, to the best of our knowledge, is the first work of its kind. Using our recently proposed time-domain morphology and gradient-based ECG feature extraction algorithm, the QRS complex is extracted and discrete wavelet transform (DWT) with one level of decomposition, using the 'Haar' wavelet, is applied on it to detect the presence of fragmentation. Detailed DWT coefficients were observed to hypothesize the postulates of detection of all types of morphologies as reported in the literature. To model and verify the algorithm, PhysioNet's PTB database was used. Forty patients were randomly selected from the database and their ECG were examined by two experienced cardiologists and the results were compared with those obtained from the algorithm. Out of 40 patients, 31 were considered appropriate for comparison by two cardiologists, and it is shown that 334 out of 372 (89.8%) leads from the chosen 31 patients complied favourably with our proposed algorithm. The sensitivity and specificity values obtained for the detection of f-QRS were 0.897 and 0.899, respectively. Automation will speed up the detection of fragmentation, reducing the human error involved and will allow it to be implemented for hospital-based remote monitoring and ICD devices.

Detection of myocardial scar from the VCG using a supervised learning approach.

This paper addresses the possibility of detecting presence of scar tissue in the myocardium through the investigation of vectorcardiogram (VCG) characteristics. Scarred myocardium is the result of myocardial infarction (MI) due to ischemia and creates a substrate for the manifestation of fatal arrhythmias. Our efforts are focused on the development of a classification scheme for the early screening of patients for the presence of scar. More specifically, a supervised learning model based on the extracted VCG features is proposed and validated through comprehensive testing analysis. The achieved accuracy of 82.36% (sensitivity 84.31%, specificity 77.36%) indicates the potential of the proposed screening mechanism for detecting the presence/absence of scar tissue.

On the Detection of Myocadial Scar Based on ECG/VCG Analysis.

In this paper, we address the problem of detecting the presence of a myocardial scar from the standard electrocardiogram (ECG)/vectorcardiogram (VCG) recordings, giving effort to develop a screening system for the early detection of the scar in the point-of-care. Based on the pathophysiological implications of scarred myocardium, which results in disordered electrical conduction, we have implemented four distinct ECG signal processing methodologies in order to obtain a set of features that can capture the presence of the myocardial scar. Two of these methodologies are: 1) the use of a template ECG heartbeat, from records with scar absence coupled with wavelet coherence analysis and 2) the utilization of the VCG are novel approaches for detecting scar presence. Following, the pool of extracted features is utilized to formulate a support vector machine classification model through supervised learning. Feature selection is also employed to remove redundant features and maximize the classifier's performance. The classification experiments using 260 records from three different databases reveal that the proposed system achieves 89.22% accuracy when applying tenfold cross validation, and 82.07% success rate when testing it on databases with different inherent characteristics with similar levels of sensitivity (76%) and specificity (87.5%).