Keywords: Decision support systems for the control of physiological and clinical variables, Quantification of physiological parameters for diagnosis assessment, Healthcare management and delivery, disease control, critical care
Abstract: Determining physiological mechanisms contributing to circulatory failure can be challenging, contributing to the difficulties of delivering effective hemodynamic management in critical care. Measured or estimated Frank-Starling curves could potentially make it much easier to assess patient response to interventions, and thus to manage circulatory failure. This study combines non-additionally invasive model-based methods to estimate left ventricle end-diastolic volume (LEDV) and stroke volume (SV) during hemodynamic interventions in a pig trial. Frank-Starling curves are created using these metrics and Frank-Starling contractility (FSC) is identified as the gradient. Bland-Altman median bias [limits of agreement (2.5th, 97.5th percentile)] are 0.14 [-0.56, 0.57] for model-based FSC agreement with measured reference method FSC using admittance catheter LEDV and aortic flow probe SV. This study provides proof-of-concept Frank-Starling curves could be non-additionally invasively estimated clinically for critically ill patients to provide clearer insight into cardiovascular function than is currently possible.

Keywords: Healthcare management and delivery, disease control, critical care, Artificial organs and biomechanical systems, Biomedical system modelling
Abstract: Receding Horizon Control (RHC), also known as Model Predictive Control (MPC) is one of the most intensively researched areas of control algorithms applied in the artificial pancreas concept. Nevertheless, MPC algorithms have not yet been implemented in commercially available insulin pumps, mainly due to their high computational demand, their less robust nature, and their instability on account of model’s uncertainty. In this paper, we present a robust adjustable RHC. The proposed RHC controller was tested under known food inputs by applying a high degree of parameter uncertainty to the virtual patient implemented in the controller to test the robustness of the architecture. A particle swarm optimization method was applied to tune the controller. The so-called identifiable virtual patient (IVP) model was used in the tests, supplemented with food absorption and continuous glucose monitoring sensor model. The implementation was performed in Julia. The results showed that the proposed RHC is sufficiently robust under high food intake and parameter uncertainty.

Keywords: Artificial intelligence for decision support systems, Artificial intelligence support in diagnosis and decision making systems, Identification and validation
Abstract: Facial emotion recognition (FER) has gained interest and focus over the years. It can be useful in many different applications and could offer significant benefit as part of feedback systems to help train children with Autism Spectrum Disorder (ASD) who struggle to recognize facial expressions and emotions. This paper explores the effectiveness and significance of image pre-processing in Neural Networks on developing suitable models for classification. Transfer Learning using the popular “AlexNet” architecture was used in the development of the model with three different approaches for image inputs. Model performance was compared using accuracy of randomly selected validation set after training on a different random training set from the Oulu-CASIA database and visualizations of predicted areas of importance analyzed. Image classes were distributed evenly, and accuracies of up to 99.90% were observed with small variation between approaches but significant difference in regions of impact. The visualization process highlighted the importance of image pre-processing prior to network training to improve accuracy and eventual efficacy for this application in ASD.

Keywords: Robotics, Artificial intelligence for decision support systems, Biosignal analysis and interpretation,
Abstract: This paper describes the development of PlatypOUs -- an open-source electromyography (EMG)-controlled mobile robot platform that uses the MindRove Brain Computer Interface (BCI) headset as signal acquisition unit, implementing remote control. Simultaneously with the physical mobile robot, simulation environment is also prepared using Gazebo, within the Robot Operating System (ROS) framework, with the same capabilities as the physical device, from the point of view of the ROS. The purpose of the PlatypOUs project is to create a tool for STEM-based education, and it involves two major disciplines: mobile robotics and machine learning, with several sub-areas included in each. The use of the platform and the simulation environment exposes students to hands-on laboratory sessions, which contribute to their progression as engineers. An important feature of our project is that the platform is made up of open-source and easily available commercial hardware and software components. In this paper, an electromyography (EMG) based controller has been developed using support vector machine (SVM) based classification for robot control purposes.

Keywords: Decision support systems and feedback control, Decision support systems for the control of physiological and clinical variables, Intensive and chronic therapy
Abstract: Healthcare and intensive care unit (ICU) medicine in particular, are facing a devastating tsunami of rising demand multiplied by increasing chronic disease and aging demographics, which is unmatched by society’s ability to pay. Digital technologies and automation have brought significant productivity gains to many industries, and manufacturing in particular, but not yet to medicine. In manufacturing, digital twins, model-based optimisation of manufacturing systems and equipment, are a rapidly growing means of further enhancing productivity and quality. This concept intersects well with the model-based decision support and control just beginning to emerge into clinical use, offering the opportunity to personalise care, and improve its quality and productivity. This article presents digital twins in a manufacturing concept and translates it into clinical practice, and then reviews the state of the art in key areas of ICU medicine.

Keywords: Quantification of physiological parameters for diagnosis assessment, Model formulation, Decision support systems for the control of physiological and clinical variables
Abstract: Introduction: Current practice in coaching track cycling sprint athletes is a focus on a very narrow band of power output from 1-4 seconds. However, there is a small oxidative contribution to sprint performance as short as 10-s, and this contribution increases as a rider competes in multiple events. All Olympic track cycling events demand repeated sprint performance! Purpose: This study models sprint-cycling performance to investigate the role of durations requiring a high oxidative contribution to energy supply and their relationship to sprint-cycling power durations. It hypothesizes power at endurance durations are strongly related to power at sprint durations, and further, these relationships may be nonlinear and saturable. Methods: Power meter data was used from 89 participants (192 datasets) to model fit the data using 4 different models (exponential, linear, parabolic, and power) using total least-squares. All data was based on a (0,0) start point acknowledging neither glycolytic or oxidative pathways operate independently. Dependent variables were 15 and 30 second power, and predictor variables 2, 8 and 20 minute power. Results: All four models yielded high r2 values (r2 > 0.81), and the exponential and linear models in particular. Strong correlations for all models demonstrates the role of oxidative power duration on performance over short durations. The linear model was the best model based on consistent, high r2 values and model simplicity, validating the first hypothesis, but nullifying the second. Conclusion: The results show maximal performance in sprint-cycling durations of 15 and 30 seconds are strongly related to maximal performance in 2, 8, and 20 minute power, and training at these durations does not diminish performance, and with a season, training maximally at these durations complements performance. These results match physiological studies showing oxidative pathways play a major role in sprint and repeated sprint efforts.

Keywords: Artificial intelligence support in diagnosis and decision making systems, Biosignal analysis and interpretation,, Artificial intelligence for decision support systems
Abstract: Mechanical ventilated respiratory failure patients may experience asynchronous breathing (AB). Frequent occurrence of AB may impose detrimental effect towards patient’s condition, however, there is lack of autonomous AB detection approach impedes the explication of aetiology of AB causing underestimation of the impact of AB. This research presents a machine learning approach, a dual input convolutional neural network (CNN) to identify 5 types of AB and normal breathing by accepting both airway pressure and flow waveform profiles concurrently. The model was trained with 6,000 breathing cycles and validated with 1,800 isolated data collected from clinical trials. Results show that the trained model achieved a median accuracy of 98.6% in the 5-fold cross-validation scheme. When validated with unseen patient’s data the trained model achieved an accuracy median of 96.2%. However, the model was found to misidentify premature cycling with reverse triggering. The results suggest that it may be difficult to clearly distinguish ABs with similar features and should be trained with more data. Nonetheless, this research demonstrated that a dual input CNN model able to accurately categorise AB which can potentially aid clinicians to better understand a patient’s condition during treatment.

Keywords: Artificial intelligence for decision support systems, Pharmaceutical processes, Healthcare management and delivery, disease control, critical care
Abstract: The present work brings to the reader's attention the benefits and facilities of the PillCrop application. The application enwraps a complete solution for improving the weight accuracy of each dose of medicine given to patients. The application recognizes, using artificial intelligence, the dose recommended by the medical specialist and the standard weight of each pill. If the medicinal drug can be divided it generates two surfaces that represent the necessary division of the pill to fit the weight indicated by the dose. This recommendation uses augmented reality to illustrate the part of the medicine that represents the recommended dose to the patient and where the pill should be divided.

Keywords: Artificial intelligence for decision support systems, Artificial intelligence support in diagnosis and decision making systems
Abstract: Image-based surgical phase recognition is a fundamental component for developing context-aware systems in future operating rooms (ORs) and thus enhance patient outcomes. To date, phase recognition in laparoscopic videos has been investigated, and spatio-temporal deep learning-based approaches have been introduced. However, phase recognition in laparoscopic videos is still a challenging task and requires ongoing research. In this work, a spatio-temporal deep learning approach for recognising surgical phases is proposed. The proposed framework consists of a convolutional neural network (CNN) and a cascade of three long short-term memory (LSTM) networks. The first and second LSTM networks were trained to learn temporal information from short video clips and the complete video sequence to perform tool detection. The last LSTM was employed to enforce temporal constraints of surgical phases. The proposed approach was thoroughly evaluated on the Cholec80 dataset, and the experimental results demonstrate the high recognition performance of this method.

Keywords: Artificial intelligence support in diagnosis and decision making systems, Biosignal analysis and interpretation,
Abstract: Tissue classification of white or gray matter is a necessary information in the study of brain connectivity. Currently this classification is made by the coregistration of the implanted electrodes in the Magnetic Resonance Imaging (MRI) of the patient. This process is complex and therefore is not always carried out, and is limited by the image resolution and by the accuracy of the coregistration. This paper studies the performance of machine learning (ML) algorithms used with features extracted from Stereo-Electroencephalogram (SEEG) signals recorded from three epileptic patients, for electrode contact classification, to serve as a decision support for specialists and researchers. The features are based on epileptic detection, and are extracted from both time and frequency domain. Accuracy, Area Under Curve and F_1-Score are evaluated for each ML algorithm, and feature importance is assessed by feature permutation. Satisfactory results were achieved, with a maximum of 79% accuracy in group separation for patient specific classification, and 74% in inter-patient classification, indicating high potential in ML techniques for brain tissue classification.

Keywords: Biosignal analysis and interpretation,, Artificial organs and biomechanical systems, Healthcare management and delivery, disease control, critical care
Abstract: Irregularity is the main characteristic of electroencephalographic signals (EEG), which needs a specific analysis method for neurological disease diagnosis. An efficient tool for signal irregularity analysis is Sample Entropy (SampEn). In this context, our paper was elaborated. We used SampEn to design a Machine Learning model for brain state detection based on EEG signals, which allows to differentiate between healthy (H) subjects, epileptic subjects during seizures free intervals (E) and epileptic subjects during seizures (S). Two main novelties are presented in our paper. The first one is related to the outline of the designed machine learning model, signal derivatives are determined as preprocessing step, then extracted features are SampEn and Standard Deviation (STD) from EEG signals and its first and second derivatives. These features are firstly used to train a K-Nearest Neighbor classifier (KNN) and yield high accuracy. After that, we select the most relevant features and we design our proposed classifier that provides better accuracy. The second one is related to the performance of our model to overcome some crucial purposes. In addition to the highest achieved accuracy, 100% for seizure detection, 99.2% for epilepsy detection and 99.86% for three class classification cases, our model used few features and simple classifier which involves fast running time. That is why we can consider our model as a suitable tool for real time applications. Irregularity is the main characteristic of electroencephalographic signals (EEG), which needs a specific analysis method for neurological disease diagnosis. An efficient tool for signal irregularity analysis is Sample Entropy (SampEn). In this context, our paper was elaborated. We used SampEn to design a Machine Learning model for brain state detection based on EEG signals, which allows to differentiate between healthy (H) subjects, epileptic subjects during seizures free intervals (E) and epileptic subjects during seizures (S). Two main novelties are presented in our paper. The first one is related to the outline of the designed machine learning model, signal derivatives are determined as preprocessing step, then extracted features are SampEn and Standard Deviat

Keywords: Artificial intelligence for decision support systems, Intensive and chronic therapy, Biomedical system modelling
Abstract: Tight glycaemic control (TGC) is a treatment in the intensive care in order to avoid stress-induced hyperglycaemia. The insulin sensitivity (SI) prediction is an essential step of the best performing, clinically applied so-called STAR (Stochastic-TARgeted) TGC protocol. Previous results showed performance improvement of the SI prediction using artificial intelligence methods. This study analyses the clinical performance of distinct artificial intelligence based SI prediction methods (2 different neural network based prediction methods: Classification Deep Network and Mixture Density Network with 3 different parametrizations and 2 variants: sex-specific and non sex-specific for each). In-silico validation was used for evaluation simulating the treatment of 171 virtual patients. Based on the results the number of input parameters involved into the prediction can effectively increase the reliability of the SI prediction. Improvements in the performance are also experienced in several cases by using sex-specific models.

Keywords: Artificial intelligence support in diagnosis and decision making systems, Artificial intelligence for decision support systems
Abstract: COVID-19 has spread around the world rapidly causing a pandemic. In this research, a set of Deep Learning architectures, for diagnosing the presence or not of the disease have been designed and compared; such as, a CNN with 4 incremental convolutional blocks; a VGG-19 architecture; an Inception network; and, a compact CNN model known as MobileNet. For the analysis and comparison, transfer learning techniques were used in forty-five different experiments. All four models were designed to perform binary classification, reaching an accuracy above 95%. A set of different scores were implemented to compare the performance of all models, showing that VGG-19 and Inception configurations performed the best.

Keywords: Biosignal analysis and interpretation,, Artificial intelligence for decision support systems, Robotics
Abstract: This work proposes an end-to-end model architecture, from feature extraction to classification using an Artificial Neural Network. The feature extraction process starts from an initial set of signals acquired by electrodes of a Brain-Computer Interface (BCI). The proposed architecture includes the design and implementation of a functional six Degree-of-Freedom (DOF) prosthetic hand. A Field Programmable Gate Array (FPGA) translates electroencephalography (EEG) signals into movements in the prosthesis. We also propose a new technique for selecting and grouping electrodes, which is related to the motor intentions of the subject. We analyzed and predicted two imaginary motor-intention tasks: opening and closing both fists and flexing and extending both feet. The model implemented with the proposed architecture showed an accuracy of 93.7% and a classification time of 8.8us for the FPGA. These results present the feasibility to carry out BCI using machine learning techniques implemented in a FPGA card.

Keywords: Artificial intelligence support in diagnosis and decision making systems, Artificial intelligence for decision support systems
Abstract: The White Matter Hyperintensities (WMHs) are usually associated with diabetes which is relevant in medical research to understand the long-term affection of diabetes. However, there is not enough evidence to distinguish whether the WMHs observed in diabetes subjects are structurally different from those observed in healthy subjects. This work aims to recognize the patterns associated with diabetes using the WMHs features of diabetic patients. We used Machine Learning models, such as Logistic Regression (LR), Support Vector Machines (SVM), Random Forest (RF), and a Multilayer perceptron (MLP) Neural Network to classify the features extracted from the WMH segments from T1 and FLAIR sequences of Magnetic Resonance Images (MRI) obtained from diabetic patients. Four classification models were evaluated and compared in their performance and Logistic Regression showed the best results, with an accuracy of 88%, as belonging or not to a diabetic class. Our results showed that diabetic patients have WMH patterns that are structurally different from controls, which may be useful for patients follow up

Keywords: Artificial intelligence support in diagnosis and decision making systems, Healthcare management and delivery, disease control, critical care
Abstract: Parkinson's disease is the second most common neurodegenerative disorder and affects more than 7 million people globally. In this work, we classify subjects with Parkinson's disease using data from finger-tapping on a keyboard. We use a free database by Physionet with more than 9 million records, preprocessed to delete atypical data. In the feature extraction stage, we obtained 48 features. We use Google Colaboratory to train, validate, and test nine supervised learning algorithms that detect the disease. As a result, we achieve a degree of accuracy higher than 98%.

Keywords: Artificial intelligence support in diagnosis and decision making systems, Artificial intelligence for decision support systems
Abstract: Many of the patients diagnosed with Parkinson’s disease (PD) do not know they have it until the most severe symptoms appear, sometimes they must wait months or even years to get the correct diagnosis, so detection in its early stage is important to improve the quality of life of patients and families. We propose the creation of a model based on supervised learning, to learn the patterns associated with the voice of PD patients. We used 1400 voice recordings of PD patients and controls which were preprocessed, further were obtained 70 features for each recording, and then we used a supervised learning algorithms such as a Multilayer Perceptron (MLP), Random Forest (RF), Logistic Regression (LR), and Support Vector Machines (SVM) to classify the data between patients and controls. From all machine learning models evaluated the SVM model showed the best performance, with an accuracy of 88%. This work presents the possibility to incorporate the voice analysis as digital biomarker to facilitate diagnosis in PD.

Keywords: Biosignal analysis and interpretation,, Artificial intelligence for decision support systems, Control of voluntary movements, respiration, locomotion
Abstract: This work presents the experimental design for recording Electroencephalography (EEG) signals in 20 test subjects submitted to Steady-state visually evoked potential (SSVEP). The stimuli were performed with frequencies of 7, 9, 11 and 13 Hz. Furthermore, the implementation of a classification system based on SSVEP-EEG signals from the occipital region of the brain obtained with the Emotiv EPOC device is presented. These data were used to train algorithms based on artificial intelligence in a Raspberry Pi 4 Model B. Finally, this work demonstrates the possibility of classifying with times of up to 6 ms in embedded systems with low computational capacity.

Keywords: Biomedical imaging systems, Artificial intelligence support in diagnosis and decision making systems
Abstract: Convolutional Neural Networks is one of the most commonly used methods for automatic prostate segmentation. However, few studies focus on the segmentation of the two main zones of the prostate: the central gland and the peripheral zone. This work proposes and evaluates two models for 2D semantic segmentation of these two zones of the prostate. The first model (Model-A) uses an encoder-decoder architecture based on the global U-net and the local U-net architectures. The global U-net segments the whole prostate, whereas the local U-net segments the central gland. The peripheral zone is obtained by subtracting the central gland from the whole prostate. On the other hand, the second model (Model-B) uses an encoder-classifier architecture based on the VGG16 network. Model-B performs segmentation by classifying each pixel of an MRI into three categories: background, central gland, and peripheral zone. Both models are tested using MRIs from the dataset NCI-ISBI 2013 Challenge. The experimental results show a superior segmentation performance for Model-A, encoder-decoder architecture, (DSC = 96.79% ± 0.15% and IoU = 93.79% ± 0.29%) compared to Model-B, encoder-classifier architecture, (DSC = 92.50% ± 1.19% and IoU = 86.13% ± 2.02%).

Keywords: Decision support systems and feedback control, Intensive and chronic therapy, Biomedical system modelling
Abstract: Photothermal therapy has emerged as a potential modality to generate hyperthermia as cancer treatment due to its low invasiveness and its capacity to complement other cancer therapies. Undesired side effects can occur such as damage to surrounding healthy tissue and temperature increase to values that cause tissue carbonization and evaporation. For this reason, this research aims to develop a multivariable fuzzy logic controller that maximizes tumor thermal damage while keeping temperature within the recommended ranges and minimizing neighboring healthy tissue damage. Three inputs were contemplated for the control system: tumor thermal damage, future maximum temperature error and future healthy tissue temperature error; thirteen logic rules were used to determine the controller output, which was established to be the change in laser power. Results showed that the control algorithm successfully accomplished the proposed goals.

Keywords: Artificial intelligence for decision support systems, Quantification of physiological parameters for diagnosis assessment, Artificial intelligence support in diagnosis and decision making systems
Abstract: In mechanically ventilated patients, some lung injuries can be reduced or avoided with therapy individualization, while the lung function is evaluated continuously, breath by breath. However, obtaining information on respiratory mechanics (respiratory system resistance and compliance) in the presence of respiratory effort is challenging, even if using invasive and complex procedures. The contribution of this work is to predict both respiratory system resistance and compliance over time using a convolutional neural network (CNN) and estimate the respiratory effort profile using the respiratory dynamics. Therefore, the approach used in this work was to generate a large amount of simulated data to feed a CNN so it could learn how to predict the correct values of the respiratory system resistance and compliance. Then, the respiratory effort was estimated by solving a first-order linear model. The main results showed a normalized mean squared error of 5.7% for the respiratory system resistance and 11.56% for compliance from Bland-Altman plots derived from the computational simulator. Finally, the method was validated using real data from an active lung simulator within which respiratory mechanics varied, and some ventilator settings were adjusted to mimic actual patient situations. The active lung simulator effort profile was obtained with a normalized mean squared error of 8.31% considering the use of an active lung simulator. The results have shown that the simulated data were valuable for the CNN training, while the performance over the real data suggested that the network was generalized accordingly for estimating respiratory parameters and effort profile.

Keywords: Decision support systems for the control of physiological and clinical variables, Artificial intelligence support in diagnosis and decision making systems, Medical information systems
Abstract: Optimizing the positive end-expiratory pressure remains challenging for any clinician treating a patient with acute respiratory distress syndrome. This paper presents an approach to automate a PEEP titration maneuver and identify the best PEEP according to maximal compliance. The respiratory system was modeled by a single-compartment model, and parameters were estimated using multiple linear regression. A classifier identified the best PEEP using the scaled relative change in compliance between PEEP levels based on empirical data from previous manual PEEP titrations. An experimental system allows the in vivo testing of the automated PEEP titration, including additional safety measures. The complete system was tested in a single animal experiment and correctly identified the best PEEP. The introduced system is a step closer towards an automated, standardized PEEP optimization and closed-loop control of mechanical ventilation.

Keywords: Biomedical imaging systems, Artificial intelligence for decision support systems, Simulation and visualization,
Abstract: Electrical Impedance Tomography (EIT) is a noninvasive, indirect image reconstruction technique which consists in the inference of the distribution of electrical conductivity inside a body or object from the set of electrical potentials measured on its boundary. Several methods have been used for the reconstruction of EIT images, such as Simulated Annealing, Kalman Filter, D-bar, and, more recently, Convolutional Neural Networks (CNN). An issue when using CNN is that the resulting image of the convolution process is smaller than the original input image. Besides that, the values lying on the borders of the input image are used less, hence the importance is overlooked. This problem is usually addressed by the introduction of padding, which is the addition of layers in the borders of the original input image. This work proposes the use of a doubly periodic padding, which is relevant for toroidal image problems such as electric potential distribution measured using EIT. The CNN is trained using a database generated by numerical simulations. The resulting image reconstructions are presented for different noisy potential inputs.

Keywords: Biomedical system modelling, Artificial intelligence support in diagnosis and decision making systems, Medical information systems
Abstract: Governments must detect and treat people with tuberculosis, also prevent the uninfected community. In this sense, must promote the study of algorithms for the prediction of the epidemic trend. This paper addresses the forecasting of tuberculosis cases in Bogota, considering health surveillance system data from 2007-2020. Forecasts are obtained using the Kalman Filter and the Robust Kalman Filter. Results show better performance using the robust filter for six-week tuberculosis cases prediction.

Keywords: Kinetic modelling and control of biological systems, Biomedical imaging systems, Biomedical system modelling
Abstract: The objective of this work consists of studying the feasibility to impose descending stair negotiation in the UGent Knee Rig (UGKR). The force and position reference signals for the control strategy are derived from literature. A mathematical coordinate transformation is developed to make the stair descent possible within the limitations of the UGKR. Kinematic measurements of the six degrees of freedom in the knee joint allow to assess knee instability. Therefore, a kinematic measurement method based on CT-images is used to measure the relative position of the bones during the stair descent. First, the study is performed on a mechanical knee hinge to evaluate the performance of the position and force control. Second, a saw-bone study is performed to assess the kinematic measurement. The results show the feasibility of the developed method to impose stair descent motions in future cadaver studies.

Keywords: Biosignal analysis and interpretation,, Biomedical imaging systems, Tracer kinetic modelling from imaging systems
Abstract: Motion capturing systems produce a large amount of information on the motion of individuals. A growing number of data reduction techniques have been developed to reduce the amount of data while keeping relevant information. An overview that compares and identifies the advantages and disadvantages of these methods on cyclic motion data is, however, lacking. Therefore, this study aims to assess the features of different data reduction techniques by applying them to a large public gait data set. Due to the periodicity of cyclic data, an individual cycle can be isolated and analyzed. The analysis of single cycles requires pre-processing steps to segment and align the individual cycles. The latter is needed to isolate the amplitude variability. Three alignment procedures with different complexity, namely Linear Length Normalization (LLN), Piecewise LNN (PLLN) and Continuous Registration (CR), are assessed based on the amount of resulting variation. Subsequently three data reduction techniques (i.e. Principal Component Analysis (PCA), Principal Polynomial Analysis (PPA) and Multivariate Functional PCA (MFPCA)) are applied to the aligned single gait cycles. The data reduction techniques are evaluated based on the in-sample error, the out-of-sample error, the compactness and the computation time to produce a model. The curves aligned with CR have the lowest remaining variation and thus the lowest amount of remaining phase variation. The differences between the different data reduction techniques appear to be minimal. PPA shows to be the most compact and is therefore recommended when compactness is crucial and out-of-sample performance is less essential. The use of MFPCA is advised when one wants to include data from different sources. PCA is suggested when computation time is key.

Keywords: Model formulation
Abstract: Correct torquing of bone screws is critical for positive patient outcomes in orthopaedic surgery. Under- or over-tightening screws can lead to thread stripping or screw loosening, leading to implant/fixation failure and potential tissue damage or disability. It has been proposed that an automated torque-limiting smart-screwdriver may be able to use model-based methods to determine the properties of bone as screws are inserted, and then use these to determine the optimal tightening torque and provide a torque-indication or -limitation to enforce this limit. Previous work focused on identifying the material properties from sensor data, but this paper will address the unanswered question of torque-limit prediction.

Here we have developed a simple model of screw thread stripping. This model is based on the assumption that overtightening the screw will shear a cylindrical section of the underlying material. This simple assumption is augmented with a stress concentration factor dependant on the screw geometry. This model was tested against experimental stripping-torque data.

We found that without the stress-concentration factor the model produced predicted torques with a strong linear relationship to the experimental values (R² = 0.98), however the magnitude of the predictions was 2-3 times too high. Including the stress concentration factor brought these predictions into the range of the experimental values, but the strong linear relationship from before was disrupted (R² = 0.80).

Overall, this approach is promising for optimal torque prediction, but needs more thorough testing with a range of materials and screws, and has room for improvement with the stress-concentration factor.

Keywords: Surgical robotics and medical nano-robotics, Biological systems and controls, Biomedical system modelling
Abstract: The first closed-loop control of electrosurgical power satisfying a specified tissue damage bound along the desired tissue dissection path is presented. The damage is represented by the 82◦C isotherm corresponding to the admissible tissue denaturation front position in relation to that of the cutting probe tip. The front location is assessed in real time through the infrared temperature readings of the 40◦C isotherm tightly related to the emerging denaturing patch size around the moving probe tip. A control-oriented denaturing hypermodel and its recasting into a form amenable for use in a moving- horizon locally linear model predictive control law are presented. The optimal control action is determined by solving a compound model predictive control problem that targets a number of active one-dimensional domains. This model is obtained from an offline trained nonlinear autoregressive model with exogenous input. To enforce the safety constraints, a supervisor system precedes the path planning control law. This system prevents excessive denaturation by excluding certain system moves, and determines system termination conditions. We experimentally demonstrate the system’s performance in two different line-cutting tasks on ex vivo porcine tissue with a desired denaturation front.

Keywords: Biomedical system modelling
Abstract: A geometric model and related methods to easily define patient specific vertebral body models have been introduced in our previous studies. This paper proposes an angle measurement method that can be fully automated after the definition of the patient specific vertebral body model. A Principal Component Analysis based algorithm allowing the quick identification of the symmetry plane of the human spline is also developed and described. The clinical dataset used to analyse and validate the models and methods introduced consists of 39 patients' lumbar section of the spinal column with 195 vertebrae.

In terms of angle measurement the proposed geometric model and the measurement method is proven to be accurate enough for clinical diagnostics, the average mean value of the measurement error 0.15° and 0.75° comparing the measurements to the two reference datasets. The average standard deviation of the error was around 2.50° that is almost the same as the average standard deviation of the two reference datasets (2.34°).