BIOSIGNALS 2022 Abstracts


Full Papers
Paper Nr: 9
Title:

Data Compression for Wireless ECG Devices

Authors:

Elena Merdjanovska, Miha Mohorčič, Matjaž Depolli, Aleksandra Rashkovska and Tomaž Javornik

Abstract: Wireless ECG devices are the latest novelty in the field of electrocardiography. ECG is commonly used in healthcare systems to observe cardiac activity, however wireless devices bring new challenges to the field of ECG monitoring. These challenges include limited battery capacity, as well as increased data storage requirements caused by daily uninterrupted ECG measurements. Both of these issues can be mitigated by introducing an efficient compression technique. This paper explores two direct data compression methods for ECG data: delta coding and Huffman coding, as well as their variations. We performed experiments both on measurements from a wireless ECG sensor – the Savvy ECG sensor, as well as on measurements from a standard public ECG database – the MIT-BIH Arrhythmia Database. We were able to select suitable parameters for delta coding for efficient compression of multiple ECG recordings from the Savvy ECG sensor, with a compression ratio of 1.6.
Download

Paper Nr: 13
Title:

AroNap: A Scent-based Nap Promotion System

Authors:

Mayo Iizuka, Anna Yokokubo and Guillaume Lopez

Abstract: Compared to other countries, Japanese sleep time is insufficient, and many people have reduced work efficiency due to daytime sleepiness caused by lack of sleep. The government recommends taking a nap as a countermeasure, but few Japanese do. This study developed a wearable sleep-onset / wake-up promotion system, ”AroNap,” that supported a short, effective nap and verified its effect. AroNap is a system that promotes falling asleep and waking up during a nap at an appropriate timing by attaching a computer-controlled scent diffusion device to a neck pillow. To verify the usefulness of AroNap and scent, we conducted an evaluation experiment to compare the effect on sleep quality of presence or absence of AroNap, the type of scent, and the sleep state according to the timing of use. We also evaluated the system using the system usability scale (SUS). Proper use of AroNap has been shown to improve sleep quality compared to other cases.
Download

Paper Nr: 15
Title:

Classification of Volatile Compounds with Morphological Analysis of e-nose Response

Authors:

Rita Alves, João Rodrigues, Efthymia Ramou, Susana J. Palma, Ana A. Roque and Hugo Gamboa

Abstract: Electronic noses (e-noses) mimic human olfaction, by identifying Volatile Organic Compounds (VOCs). This work presents a novel approach that successfully classifies 11 known VOCs using the signals generated by sensing gels in an in-house developed e-nose. The proposed signals’ analysis methodology is based on the generated signals’ morphology for each VOC since different sensing gels produce signals with different shapes when exposed to the same VOC. For this study, two different gel formulations were considered, and an average f1-score of 84% and 71% was obtained, respectively. Moreover, a standard method in time series classification was used to compare the performances. Even though this comparison reveals that the morphological approach is not as good as the 1-nearest neighbour with euclidean distance, it shows the possibility of using descriptive sentences with text mining techniques to perform VOC classification.
Download

Paper Nr: 17
Title:

Interpretable High-level Features for Human Activity Recognition

Authors:

Yale Hartmann, Hui Liu, Steffen Lahrberg and Tanja Schultz

Abstract: This paper introduces and evaluates a novel way of processing human activities based on unique combinations of interpretable categorical high-level features with applications to classification, few-shot learning, as well as cross-dataset and cross-sensor comparison, combination, and analysis. Feature extraction is considered as a classification problem and solved with Hidden Markov Models making the feature space easily extensible. The feature extraction is person-independently evaluated on the CSL-SHARE and UniMiB SHAR datasets and achieves balanced accuracies up from 96.1% on CSL-SHARE and up to 91.1% on UniMiB SHAR. Furthermore, classification experiments on the separate and combined datasets achieve 85% (CSL-SHARE), 65% (UniMiB SHAR), and 74% (combined) accuracy. The few-shot learning experiments show potential with low errors in feature extraction but require further work for good activity classification. Remarkable is the possibility to attribute errors and indicate optimization areas easily. These experiments demonstrate the potential and possibilities of the proposed method and the high-level, extensible, and interpretable feature space.
Download

Paper Nr: 19
Title:

Evaluation of Fall Detection Approaches based on Virtual Devices: Leveraging on Motion Capture Data in Unity environments

Authors:

Eduarda Vaz, Heitor Cardoso and Plinio Moreno

Abstract: Realistic fall detection datasets are difficult to acquire due to the high risks, awkward situation of pretending to be falling and limited to young healthy individuals. In this work we propose to leverage on motion capture data acquired for games and animations, to simulate the recordings of accelerometers and orientation sensors. The simulated sensor values are obtained in the Unity environment. Our dataset allows to further evaluate the generalization properties of previously presented methods by including new types of both falling and non-falling samples. Our case study is the fall detection based on wristband devices.
Download

Paper Nr: 21
Title:

Hypoxic-Ischaemic Encephalopathy Prognosis using Susceptibility Weighted Image Analysis based on Histogram Orientation Gradient

Authors:

Zhen Tang, Sasan Mahmoodi, Angela Darekar and Brigitte Vollmer

Abstract: The aim of this study is to analyse the susceptibility-weighted magnetic resonance images (SWI) by using Histogram of Oriented Gradients (HOG) as a global feature to identify areas of the neonatal brain affected by Hypoxic-ischaemic encephalopathy (HIE). 42 infants with neonatal HIE have undergone under SW imaging in the neonatal period and have been investigated through neurodevelopmental assessment at 24 months of age. HOG features are used to represent the whole brain SW images and the region of interest separated from the brain image registration algorithm. We use k-nearest neighbours (kNN) and random forest to classify the SWI images into normal and abnormal groups, and then we compare our results to our previous work. The result shows an effective classification, which achieved an accuracy of 76.25±10.9. Our research suggests that automated analysis of neonatal SWI images can identify brain regions affected by HIE on SWI images and predict motor and cognitive outcomes.
Download

Paper Nr: 24
Title:

Wavelet based Method of Mapping the Brain Activity Waves Travelling over the Cerebral Cortex

Authors:

Bozhokin Sergey, Suslova Irina and Tarakanov Daniil

Abstract: The brain electroencephalogram is treated as a set of electrical activity bursts in various spectral ranges. Spectral integrals calculated by the wavelet transform method are used to study time-frequency properties of such bursts. The mathematical theory has been developed to describe quantitatively the change in the shape of EEG bursts while their propagating along the cerebral cortex. The proposed model of neural activity uses nonlinear approximation of EEG record as a sum of several Gaussian peaks moving along different trajectories with different speeds. Such model together with the continuous wavelet transform provides an opportunity to receive analytical solutions. The proposed method allows us to draw the maps showing the trajectories of EEG bursts moving along the cerebral cortex. It also becomes possible to study the change in the shape of bursts in the process of their motion. The method was applied to study EEG records of a healthy subject at rest with his eyes closed.
Download

Paper Nr: 30
Title:

Bone Conduction Eating Activity Detection based on YAMNet Transfer Learning and LSTM Networks

Authors:

Wei Chen, Haruka Kamachi, Anna Yokokubo and Guillaume Lopez

Abstract: The trivial eating behaviors affect our health and sometimes lead to obesity and other health problems. We propose an automatic human eating behavior estimation system , which performs real-time inferences using a sound event detection (SED) deep learning model. In addition, We customized YAMNet, a pre-trained deep neural network by 521 audio event classes based on Mobilenet v1 depthwise-separable convolution architecture from Tensorflow. We used transfer learning shaped YAMNet as a feature extractor for acoustic signals and applied an LSTM network as a classification model that can effectively handle time-series environmental acoustic signal. Dietary events including chewing, swallowing, talking, and other (silence and noises), were collected on 14 subjects. The classification results show that our proposed method can validly perform semantic analysis of acoustic signals of eating behavior. The overall accuracy and overall F1 scores were both 93.3% in frame level, respectively. The classifier established in this study provided a foundation for preventing premature eating and a healthier eating behavior monitoring system.
Download

Paper Nr: 40
Title:

Paroxysmal Atrial Fibrillation Detection by Combined Recurrent Neural Network and Feature Extraction on ECG Signals

Authors:

Xinqi Bao, Fenghe Hu, Yujia Xu, Mohamed Trabelsi and Ernest Kamavuako

Abstract: Paroxysmal atrial fibrillation (AFib) or intermittent atrial fibrillation is one type of atrial fibrillation which occurs rapidly and stops spontaneously within days. Its episodes can last several seconds, hours, or even days before returning to normal sinus rhythm. A lack of intervention may lead the paroxysmal into persistent atrial fibrillation, causing severe risk to human health. However, due to its intermittent characteristics, it is generally neglected by patients. Therefore, real-time monitoring and accurate automatic algorithms are highly needed for early screening. This study proposes a two-stage algorithm, including a BiLSTM network to classify healthy and atrial fibrillation, followed by a feature-extraction-based neural network (NN) to identify the persistent, paroxysmal atrial fibrillation onsets. The extracted features include the entropy and standard deviation of the RR intervals. The two steps can achieve 90.14% and 92.56% accuracy in the validation sets on small segments. This overall algorithm also has the advantage of the low computing load, which shows a high potential for a portable embedded device.
Download

Paper Nr: 42
Title:

Adaptive Control of Cardio-respiratory Training in a Virtual Reality Hiking Simulation: A Feasibility Study

Authors:

Rodrigo Lima, Muhammad Asif, Honorato Sousa and Sergi Bermúdez i Badia

Abstract: Adaptive Virtual Reality applications are a novel way to enhance and promote higher levels of physical activity and cardiorespiratory fitness, leading to a healthier lifestyle and avoid cardiovascular diseases. In this study, we developed a system using a virtual hiking simulator, the Levadas from Madeira Island, that aims to increase the compliance of recommendations levels of exertion by implementing a closed-loop adaptation according to the heart rate. The system was tested with a sample of twenty healthy young adults on a repeated measures design, comparing the adaptive VR, a non-adaptive VR version of the software, and a non-VR version. Perceived exertion, presence, usability and intrinsic motivation were assessed. The results from the study reveal that the adaptive control according to the heart rate promoted approximately 20% more time of exertion in the recommended target heart rate zone, while perceiving lower levels of exertion by the participants, compared to the non-adaptive condition.
Download

Short Papers
Paper Nr: 1
Title:

A New Method of Dimensionality Reduction for Large Time Series Applied to Accelerometer Wristbands’ Signals

Authors:

Alihuén García-Pavioni and Beatriz López

Abstract: Feature extraction for high-dimensional time series has become a topic of great importance in recent years. In the medical field, the information needed to predict emotions, stress, epileptic seizures, heart attacks, Parkinson, fall detection in the elderly, and other diseases, can be provided by body sensors in the form of time series signals. The commercial usage of wearable accelerometers has also made the study of time series activity recognition gain much attention. Thus, as the time series provided by the accelerometers could be really long, consuming a lot of storage data and also hamming the machine learning classifier accuracy results, it is important to identify which features are relevant in this particular context, so the data stored can consume the least amount of memory possible in the device, while at the same time the activity classification performance would be satisfactory. This work intends to provide a way for these devices to save the relevant information needed for the machine learning activity classification, by defining a new feature extraction method. The method proposed in this work, called State Changes Representation for Time Series (SCRTS), relies on the relevant data associated with the “state changes” in the time series. These changes are identified according to the conditional probabilities of passing from one state to another during the time, and the “relevance” of each state. We show the results of this method with an experiment based on accelerometers data recorded by the ©ActiGraph wGT3X-BT wristband to recognize sedentary behavior. After applying this method, it was achieved to reduce time series frames of dimension 360, to vectors of dimension 12; while their classification accuracy was 84%.
Download

Paper Nr: 2
Title:

Contactless Heart Rate Measurement using Image Processing

Authors:

Gaganjot Kaur and Jeff Kilby

Abstract: Non-contact methods of determining the human body’s heart rate are of interest for clinical use. This research used a video magnification technique on the individual frames from a 15-second video taken using a digital single-lens reflex (DSLR) camera at 30 frames per second. It was possible to determine the heart rate beats per minute by extracting the green spectrum from a region of interest information from the video frames. In this paper, three methods are presented using this colour change between the frames transform as a signal to find the heart rate. While capturing the video’s using the camera, a commercially available pulse oximeter was used to obtain the pulse rate from the participant’s finger to validate the values calculated from the image processing techniques presented. The results show that it is possible to get a heart rate in terms of pulse rate reading using a camera and the developed MATLAB code.
Download

Paper Nr: 4
Title:

Exploiting EEG-extracted Eye Movements for a Hybrid SSVEP Home Automation System

Authors:

Tracey Camilleri, Jeanluc Mangion and Kenneth Camilleri

Abstract: Detection of eye movements using standard EEG channels can allow for the development of a hybrid BCI (hBCi) system without requiring additional hardware for eye gaze tracking. This work proposes a hierarchical classification structure to classify eye movements into eight different classes, covering both horizontal and vertical eye movements, at two different gaze angles in each of four directions. Results show that the highest eye movement classification was obtained with frontal EEG channels, achieving an accuracy of 98.47% for two directions, 74.38% with four directions and 58.31% with eight directions. Eye movements can also be classified reliably in four directions using occipital electrodes with an accuracy of 47.60% which increases to around 80% if three frontal channels are also included. The latter result was used to develop a hybrid SSVEP home automation system which exploits the EEG-extracted eye movement information. Results show that a sequential hBCI gave an average accuracy of 82.5% when compared to the 69.17% obtained with a standard SSVEP based BCI system.
Download

Paper Nr: 5
Title:

The Usage of Data Augmentation Strategies on the Detection of Murmur Waves in a PCG Signal

Authors:

José Torres, Jorge Oliveira and Elsa F. Gomes

Abstract: Cardiac auscultation is a key screening tool used for cardiovascular evaluation. When used properly, it speeds up treatment and thus improving the patient’s life quality. However, the analysis and interpretation of the heart sound signals is subjective and dependent of the physician’s experience and domain knowledge. A computer assistant decision (CAD) system that automatically analyse heart sound signals, can not only support physicians in their clinical decisions but also release human resources to other tasks. In this paper, and to the best of our knowledge, for the first time a SMOTE strategy is used to boost a Convolutional Neural Network performance on the detection of murmur waves. Using the SMOTE strategy, a CNN achieved an overall of 88.43%.
Download

Paper Nr: 6
Title:

12-Lead ECG Reconstruction via Combinatoric Inclusion of Fewer Standard ECG Leads with Implications for Lead Information and Significance

Authors:

Utkars Jain, Adam A. Butchy, Michael T. Leasure, Veronica A. Covalesky, Daniel Mccormick and Gary S. Mintz

Abstract: The electrocardiogram (ECG) is the most widely used, non-invasive, cardiovascular test. There exist many lead variations including a one, three, six, and 12-lead device. In this work, we use ECGio, a validated deep learning model for the assessment of coronary artery disease, to reconstruct ECG signals with various combinations of leads, ranging from a single lead, to the full 12-leads. We are able to show 0.6536 R2, and 0.0747 mean absolute error (MAE) in the accurate reconstruction of a full 12-lead signal from just lead II. We go one step further and look at which individual leads, and in what combinations, yield the most accurate reconstructions as measured by R2 and MAE. As you would expect, the larger the quantity of leads included, the more accurate the reconstruction. Overall, the mean performance across all possible lead combinations is 0.8335 R2, and 0.0538 MAE. This work opens the door for seeing if ECGio can handle systematic noise injection and missing or misplaced leads.
Download

Paper Nr: 8
Title:

Contactless Measurement of Respiratory Volumes: A Calibration Free Method based on Depth Information

Authors:

Felix Wichum, Jacqueline Hassel, Christian Wiede and Karsten Seidl

Abstract: Measurements of respiratory volumes involve a great deal of effort, either by immobile equipment such as bodyplethysmography or by consumables as with spirometers. Contactless measurement methods can remedy this situation. In this paper, a depth camera is used to generate a contactless respiratory signal. A region of interest is placed over the subject’s upper body and the distance-time curve of respiratory motion is recorded. Via selected signal features and the use of an artificial neural network, we can show that this method is equal to the use of conventional volume determination. From a comparison with a spirometer connected in parallel as a reference, a mean error for tidal volume of −0.10 l and vital capacity of 0.09 l is obtained.
Download

Paper Nr: 11
Title:

Optimization of Tracer Dose for Scintigraphic Imagery

Authors:

C. Bousnah, S. Anebajagane, O. Monsarrat, J.-Ph. Conge, H. Maaref and V. Vigneron

Abstract: Myocardial scintigraphy is a non-invasive isotope examination that has played a central role in the management of these coronary heart diseases for decades.it has proven its performance in nuclear cardiology, mainly for the diagnosis of ischemia by making it possible to analyze the myocardial perfusion, and precisely, to evaluate the quality of the irrigation by the arteries and the coronaries, as well as for the diagnosis of coronary heart disease. It is based on the injection of an intravenous radioactive tracer, which, once injected, is absorbed by the heart muscle. The radiation emitted by the radioactive tracer is converted into an image by computer tomography. However, these scintigraphic images suffer from poor spatial resolution in particular, in obese patients, it is difficult to obtain images of sufficient quality using the recommended standard doses due to the attenuation of γ−rays by soft tissues (fat, fibrous tissues, etc.). This phenomenon prompts the nuclear physician to overdose the tracer and the dose of radiation received exceeds the admissible regulatory limits. In this paper we propose a machine learning model that predict the dose of tracer based on patient’s morphological parameters to obtain images of sufficient quality to support the cardiovascular diagnosis while exposing him to the lowest possible doses of radiation. We show the body weight is not the best-predicting parameter for image quality.
Download

Paper Nr: 14
Title:

RoSe: Robot Sentinel as an Alternative for Medicinal or Physical Fixation and for Human Sitting Vigils

Authors:

Robert Erzgräber, Falko Lischke, Frank Bahrmann and Hans-Joachim Böhme

Abstract: An approach for a Robot Sentinel is described as an alternative to medicinal or physical fixation. The robot offers the opportunity to give the patient some privacy while also offering protection from falling out of bed. This approach is solely based on input data given by a Kinect One. A database with IR data with labels according to the sleep stages of the patient was generated. With given database the presented framework is able to detect the movement of the patient in bed from given input data and therefore warn the staff, if a possible harmful situation occurs. In two different experimental phases the approach could be tested and was able to successfully recognize different sleeping phases of the patient (e.g. unsettled sleep, falling asleep and wakeup phase). An unsettling sleep serves as an indication of waking up and therefore the possible desire of standing up. Recognizing those sleeping phases and counteracting this desire, preserves the patient from falling out of bed and potential injury.
Download

Paper Nr: 16
Title:

Denoising of Dynamic Contrast-enhanced Ultrasound Sequences: A Multilinear Approach

Authors:

Metin Calis, Massimo Mischi, Alle-Jan Van Der Veen and Borbála Hunyadi

Abstract: The recent advances in three-dimensional imaging of contrast-enhanced ultrasound acquisitions enable the characterization of the tissue with a single intravenous injection of microbubbles. Many cancer markers have been extended to cover for the three-dimensional contrast ultrasound. However, most of the signal denoising algorithms do not exploit the added dimensionality and vectorize the spatial dimensions, causing a loss of information about the location of the voxels. This paper proposes a denoising algorithm based on the multilinear singular value decomposition and compares it to the singular value decomposition. The ranks are estimated based on information-theoretic criteria, and improved performance has been observed for modeling the time-intensity curves.
Download

Paper Nr: 20
Title:

The Comparison of Various Correlation Network Models in Studying Mobility Data for the Analysis of Depression Episodes

Authors:

Rama K. Thelagathoti and Hesham H. Ali

Abstract: Depression is a serious mental health disorder affecting millions of people around the world. Traditional diagnostic approaches are subjective including self-reporting feedback from patients and observational evaluation by a trained physician. However, altered motor activity is the central feature for depressive disorder. Moreover, recent studies show that the analysis of motor activity is the best predictor in characterizing psychological disorders including depression. With the advent of wearable devices, an individual’s motor activity can be monitored naturally using body worn sensors and feasible to distinguish depressed persons from healthy individuals. In this manuscript, we hypothesis to apply a methodology that takes advantage of motor activity recorded from wearable devices and process mobility patterns for a given group of subjects. Besides, employed a population analysis approach using correlation networks that evaluates mobility parameters of the population and identify subgroups that exhibit similar motor complexity. We have analyzed the mobility data of the given group by extracting three different sets of features using hour-wise, day-wise, and hybrid mobility data. Also, a comparison study of three models is presented by constructing a correlation graph and finding a cluster of individuals exhibiting similar mobility patterns. We found that mobility data using hour-wise features provides the best results compared to the other two models.
Download

Paper Nr: 23
Title:

A Calibration-free Blood Pressure Measurement on a Scale: Concept and Challenges

Authors:

Christian Wiede, Carolin Wuerich and Anton Grabmaier

Abstract: Two health parameters are most relevant for self-monitoring of hypertension: blood pressure and body weight. Blood pressure is normally measured with a blood pressure cuff, whereas body weight can be measured with a simple body scale. If it is possible to integrate blood pressure measurement into easy-to-use body scales, patients will benefit from simpler use and lower overall price. The aim of this work is to develop a body scale with which blood pressure can be measured without calibration and without the need for additional devices. This can be realised by considering surrogate parameters for blood pressure. Starting from sensors such as electrodes, photo diodes and pressure transducers, various biosignals such as ECG, BCG, PPG or bioimpedance are extracted from the sole of the foot. The signal is reduced to morphological features which serve as input to a neural network for blood pressure determination. The integrated artificial intelligence (AI) is to be implemented in an energy-efficient way on an embedded system. In addition, the energy-efficient implementation ensures battery operation for several months with daily use. Besides the concept, the strengths, weaknesses, threats and opportunities of this concept are examined in detail within the framework of a SWOT analysis. This includes considerations of hardware, software, data and user experience.
Download

Paper Nr: 25
Title:

Cross-lingual Detection of Dysphonic Speech for Dutch and Hungarian Datasets

Authors:

Dávid Sztahó, Miklós G. Tulics, Jinzi Qi, Hugo Van Hamme and Klára Vicsi

Abstract: Dysphonic voices can be detected using features derived from speech samples. Works aiming at this topic usually deal with mono-lingual experiments using a speech dataset in a single language. The present paper targets extension to a cross-lingual scenario. A Hungarian and a Dutch speech dataset are used. Automatic binary separation of normal and dysphonic speech and dysphonia severity level estimation are performed and evaluated by various metrics. Various speech features are calculated specific to an entire speech sample and to a given phoneme. Feature selection and model training is done on Hungarian and evaluated on the Dutch dataset. The results show that cross-lingual detection of dysphonic speech may be possible on the applied corpora. It was found that cross-lingual detection of dysphonic speech is indeed possible with acceptable generalization ability, while features calculated on phoneme-level parts of speech can improve the results. Considering cross-lingual classification test sets, 0.86 and 0.81 highest F1-scores can be achieved for feature sets with the vowel /E/ included and excluded, respectively and 0.72 and 0.65 highest Pearson correlations can be achieved or severity prediction using features sets with the vowel /E/ included and excluded, respectively.
Download

Paper Nr: 26
Title:

A New Method to Determine Systolic Blood Pressure Indirectly Aided by Parallel Recording of ECG and PPG

Authors:

Péter Nagy and Ákos Jobbágy

Abstract: Raised blood pressure severely increases the risk of lethal cardiovascular diseases. Home monitoring of blood pressure is vital in early detection and treatment of hypertonia. Accuracy of indirect blood pressure measurement methods is sensitive to many physiological factors that are difficult to measure or control. The accuracy can be improved by using further sensors. In this paper, we propose a new method for the estimation of systolic blood pressure based on cuff pressure, ECG and photoplethysmographic (PPG) signals. PPG is measured without hardware filtering keeping the DC-component and avoiding the problem of distorting the signal. The proposed method was validated by applying it to healthy senior and healthy young adults at rest and by making a measurement series containing mild physical exercise for healthy young adults. Results of the tests clearly show the supremacy of the new method to conventional oscillometric procedure.
Download

Paper Nr: 28
Title:

Discovery of Effective Spectrum for Classifying iPS Cells Taken with CARS Microscope

Authors:

Ryouichi Furukawa, Yohei Hayashi, Hideaki Kano, Junichi Matsumoto, Shoichi Honda and Kazuhiro Hotta

Abstract: There is a technique using the CARS (Coherent Anti-Stokes Raman Scattering) microscope to identify iPS cells. CARS microscope can visualize the different molecular structures of iPS cells in each spectrum, so it is possible to identify iPS cells without destroying them. However, the information on molecules in the spectrum obtained by the CARS microscope is so diverse that it takes a great deal of time and effort to identify them. We propose a method to automatically identify the spectrum, which is effective for iPS cell identification, thereby reducing the time and effort required for identification using the CARS microscope. In this paper, we propose a network that handles multi-resolution information in parallel to learn both image classification and segmentation simultaneously. Moreover, the effective spectrum for classifying iPS cells are discovered by using the network gradients and the F-measure for cell segmentation. By the experiments on four kinds of iPS cells, we confirmed that the accuracy of the proposed method for classifying iPS cells achieved 99%. Furthermore, the effective spectrum for each iPS cell could be automatically identified.
Download

Paper Nr: 33
Title:

Generalized Poincaré Plots Analysis of Cardiac Interbeat Intervals in Heart Failure

Authors:

Mirjana M. Platiša, Nikola N. Radovanović, Aleksandar Kalauzi and Siniša Pavlović

Abstract: In this work we applied generalized Poincaré plots (gPp) analysis of interbeat intervals in patients with heart failure. More, we compared gPp with its nearest analogy methods based on existing extended Poincaré plots techniques. Obtained results showed advantages of gPp method over usually used distanced (lagged) Poincaré plots analysis. Only gPp has the potential of three-dimensional visualization of results with quantification of new multiscaling parameters. It is comparable with other methods only in two-dimensional planes where all methods showed a strong negative correlation between patterns of Pearson correlation coefficients and patterns of the SD1/SD2 ratio over the whole range of Pp orders (lags). These results could be used as the basis for further research in new standardization of multiscaling methods in heart rhythm analysis where it is important to follow the pattern of regulatory mechanisms dynamics which is related to the duration of RR intervals.
Download

Paper Nr: 34
Title:

A Subset of Acoustic Features for Machine Learning-based and Statistical Approaches in Speech Emotion Recognition

Authors:

Giovanni Costantini, Valerio Cesarini and Daniele Casali

Abstract: In this paper, a selection of acoustic features, derived from literature and experiments, is presented for emotion recognition. Additionally, a new speech dataset is built by recording the free speech of six subjects in a retirement home, as part of a pilot project for the care of the elder called E-Linus. The dataset is employed along with another widely used set (Emovo) for testing the effectiveness of the selected features in automatic emotion recognition. Thus, two different machine learning algorithms, namely a multi-class SVM and Naïve Bayes, are used. Due to the unbalanced and preliminary nature of the retirement home dataset, a statistical method based on logical variables is also employed on it. The 24 features prove their effectiveness by yielding sufficient accuracy results for the machine learning-based approach on the Emovo dataset. On the other hand, the proposed statistical method is the only one yielding sufficient accuracy and no noticeable bias when testing on the more unbalanced retirement home dataset.
Download

Paper Nr: 36
Title:

Machine Learning-based Study of Dysphonic Voices for the Identification and Differentiation of Vocal Cord Paralysis and Vocal Nodules

Authors:

Valerio Cesarini, Carlo Robotti, Ylenia Piromalli, Francesco Mozzanica, Antonio Schindler, Giovanni Saggio and Giovanni Costantini

Abstract: Dysphonia can be caused by multiple different conditions, which are often indistinguishable through perceptual evaluation, even when undertaken by experienced clinicians. Furthermore, definitive diagnoses are often not immediate and performed only in clinical settings through laryngoscopy, which is an invasive procedure. This study took into account Vocal Cord Paralysis (VCP) and Vocal Nodules (VN) given their perceptual similarity and, with the aid of euphonic control subjects, aimed to build a framework for the identification and differentiation of the diseases. A dataset of voice recordings comprised of 87 control subjects, 85 subjects affected by VN, and 120 subjects affected by VCP was carefully built within a controlled clinical setting. A Machine-Learning framework was built, based on a correlation-based feature selection bringing relevant biomarkers, followed by a ranker and a Gaussian Support Vector Machine (SVM) classifier. The results of the classifications were promising, with the comparisons versus healthy subjects bringing accuracies higher than 98%, while 89.21% was achieved for the differentiation. This suggests that it may be possible to automatically identify dysphonic voices, differentiating etiologies of dysphonia. The selected biomarkers further validate the analysis highlighting a trend of poor volume control in dysphonic subjects, while also refining the existing literature.
Download

Paper Nr: 38
Title:

Preliminary Results on the Use of Classification Trees to Predict Non-suicidal Self-injury with Data Collected through a Mobile App

Authors:

Chiara Capra, Pere Marti-Puig, Daniel V. Moreno, Laia Llunas, Stella Nicolaou, Carlos Schmidt and Jordi Solé-Casals

Abstract: Machine learning (ML) integrated with technology has been a breakthrough in mental health, bringing clinical improvements both for the patient and for the clinician. Among these, real-time patient symptoms’ tracking through ecological momentary assessment (EMA) data can be a valuable tool to forecast symptomatology at the individual-patient level for specific disorders, among which non suicidal self-injury. We aimed at applying classification trees to predict non-suicidal self-injury (NSSI) with EMA data collected through a mobile app. A database of 40 patients diagnosed with borderline personality disorder (BPD) with NSSI (N=22), and a subclinical group of students with NSSI (N=19) was analysed. EMA data was collected by the Sinjur app. Two classification trees were used as models. For the first tree, training results reported 69,7% of accuracy, whereas test results reported 59,3% of accuracy, 87,5% of sensitivity and 58,78% of specificity. For the second tree, training results reported 67,9% of accuracy, whereas test results reported 65,2% of accuracy, 85% of sensitivity and 64,8% of specificity. We concluded that real-time patient monitoring via a mobile app can be a valuable tool for making technology-based predictions at the individual patient level. This promising data needs to be built upon in future studies and needs major translation in the everyday clinical practice to demonstrate its real-world efficacy and later, to be translated to the enterprise world.
Download

Paper Nr: 39
Title:

Spectral Classification of Microplastics using Neural Networks: Pilot Feasibility Study

Authors:

Petr Dolezel, Jiri Rolecek, Daniel Honc, Dominik Stursa and Bruno B. Zanon

Abstract: Microplastics, i.e. synthetic polymers that have particle size smaller than 5 mm, are emerging pollutants that are widespread in the environment. In order to monitor environmental pollution by microplastics, it is necessary to have available rapid screening techniques, which provide the accurate information about the quality (type of polymer) and quantity (amount). Spectroscopy is an indispensable method, if precise classification of individual polymers in microplastics is required. In order to contribute to the topic of autonomous spectra matching when using spectroscopy, we decided to demonstrate the quality and efficiency of neural networks. We adopted three neural network architectures, and we tested them for application to spectra matching. In order to keep our study transparent, we use publicly available dataset of FTIR spectra. Furthermore, we performed a deep statistical analysis of all the architectures performance and efficiency to show the suitability of neural networks for spectra matching. The results presented at the end of this article indicated the overall suitability of the selected neural network architectures for spectra matching in microplastics classification.
Download

Paper Nr: 41
Title:

Wavelet based Machine Learning Approaches towards Precision Medicine in Diabetes Mellitus

Authors:

Adeethyia Shankar, Stephanie Chang, Xiaodi Wang and Yongzhong Zhao

Abstract: It is estimated that 422 million people around the world have diabetes mellitus (DM)—a devastating, complex, and highly heterogeneous disease—requesting better interventions based on disease subtyping. In this research, we utilize the discrete wavelet transform (DWT) to decompose and denoise DM data. Using DWT, we enhance heart rate variability (HRV) based DM diagnosis, data visualization of the disparities in Human Microbiome Project (HMP) data (gut bacteria, metabolomics, proteomics, RNA sequencing, targeted proteomics, and transcriptomics data) using demographic features, and insulin resistance prediction. We also attempt to forecast continuous glucose monitoring (CGM) ahead by 90 minutes because CGM is unable to provide real-time blood glucose measurements. We achieve 91.9% diagnosis accuracy for Type 1 DM using Random Forest on data transformed with DWT, holding the potential for usage in clinics. In addition, our DWT-based t-SNE and UMAP explorative analysis of HMP data support subtypes of prediabetic patients stratified by sex, race, and age. Moreover, DWT-based transformations provide multi-view clustering that any other methods would not provide on metabolomics, proteomics, RNA sequencing, targeted proteomics, and transcriptomics data and outperform those without DWT. Taken together, DWT-based machine learning approaches enable a fine resolution of subtyping DM towards precision medicine.
Download

Paper Nr: 7
Title:

Prediction of Personal Characteristics and Emotional State based on Voice Signals using Machine Learning Techniques

Authors:

Marta B. Guerreiro, Catia Cepeda, Joana Sousa, Carolina Maio, João Ferreira and Hugo Gamboa

Abstract: Voice signals are a rich source of personal information, leading to the main objective of the present work: study the possibility of predicting gender, age, and emotional valence through short voice interactions with a mobile device (a smartphone or remote control), using machine learning algorithms. For that, data acquisition was carried out to create a Portuguese dataset (consisting in 156 samples). Testing Support Vector Machine (SVM), K-Nearest Neighbors (KNN), and Random Forest (RF) classifiers and using features extracted from the audio, the gender recognition model achieved an accuracy of 87.8%, the age group recognition model achieved an accuracy of 67.6%, and an accuracy of 94.6% was reached for the emotion model. The SVM algorithm produced the best results for all models. The results show that it is possible to predict not only someone’s specific personal characteristics but also its emotional state from voice signals. Future work should be done in order to improve these models by increasing the dataset.
Download

Paper Nr: 10
Title:

Biomedical Text Mining: Applicability of Machine Learning-based Natural Language Processing in Medical Database

Authors:

Nafiseh Mollaei, Catia Cepeda, Joao Rodrigues and Hugo Gamboa

Abstract: Machine learning has demonstrated superior performance in solving many problems in various fields of medicine compared to non-machine learning approaches. The aim of this review is to understand how Machine Learning-based Natural Language Processing (ML-NLP) has been applied to the clinical notes databases. Optimization algorithms are listed as examples to demonstrate the simplicity and effectiveness of their applications for clinical notes database. We reviewed the literature in clinical applications of ML-NLP, particularly techniques of deep learning such as mainly in pathology reports of diabetes, schizophrenia, cancer and cardiology, where NLP either on a classical algorithm or with deep learning has been actively adopted. We covered 60 different studies in this domain, focusing on a wide range of medical perspective based algorithms. Machine learning-based approaches combine the benefits of health systems with the expertise and experience of human well-being. From this review, it is clear that these techniques can improve the quantification of diagnosis and prognosis of cases and may create tools to assist patients during diagnosis and treatment. We complete this work by providing guidelines on the applicability of ML-NLP by describing the most relevant libraries to extract medical expressions from clinical reports text that can support clinical decision-making.
Download

Paper Nr: 12
Title:

Iris Segmentation based on an Optimized U-Net

Authors:

Sabry A. M., Lubos Omelina, Jan Cornelis and Bart Jansen

Abstract: Segmenting images of the human eye is a critical step in several tasks like iris recognition, eye tracking or pupil tracking. There are a lot of well-established hand-crafted methods that have been used in commercial practice. However, with the advances in deep learning, several deep network approaches outperform the handcrafted methods. Many of the approaches adapt the U-Net architecture for the segmentation task. In this paper we propose some simple and effective new modifications of U-Net, e.g. the increase in size of convolutional kernels, which can improve the segmentation results compared to the original U-Net design. Using these modifications, we show that we can reach state-of-the-art performance using less model parameters. We describe our motivation for the changes in the architecture, inspired mostly by the hand-crafted methods and basic image processing principles and finally we show that our optimized model slightly outperforms the original U-Net and the other state-of-the-art models.
Download

Paper Nr: 29
Title:

Heart Rate Estimation based on Optical Flow: Enabling Smooth Angle Changes in Ultrasound Simulation

Authors:

Henning Schäfer, Hendrik Damm and Christoph M. Friedrich

Abstract: Ultrasound simulators show previously recorded ultrasound videos from different angles to the trainee. During acquisition, breathing, pulse, and other motion artifacts are involved, which often prevent a smooth image transition between different angles during simulation. In this work, a global motion vector is derived using the Lucas–Kanade method for calculating the optical flow in order to create a motion profile in addition to the recording. This profile allows transition synchronization in ultrasound simulators. For the transition in kidney recordings, the Pearson’s r correlation could be increased from 0.252 to 0.495 by autocorrelating motion profiles and synchronizing them based on calculated delays. Approaches based on tracking and structural similarity were also evaluated, yet these have shown inferior qualitative transition results. In ultrasound videos with visibility of vessels, e.g., thyroid gland with carotid artery or echocardiogram, the heart rate can also be estimated via the optical flow. In the abdominal region, the signal contains respiratory information. Since the motion profile can be generated in real time directly at the transducer position, it could be useful for diagnostic purposes.
Download

Paper Nr: 31
Title:

Personalized Evaluation of Life-threatening Conditions in Chronic Kidney Disease Patients: The Concept of Wearable Technology and Case Analysis

Authors:

Ana Santos Rodrigues, Birutė Paliakaitė, Saulius Daukantas, Andrius Sološenko, Andrius Petrėnas and Vaidotas Marozas

Abstract: The progressive aging of society results in a one-third increase in mortality rates of chronic kidney disease (CKD) patients over the past decade. In the end stage of CKD, 40% of deaths are sudden deaths due to cardiac arrhythmias precipitated by electrolyte imbalance. Unfortunately, there is a lack of technology for unobtrusive long-term monitoring of life-threatening conditions, leading to limited knowledge on arrhythmia characteristics and their relationship with complications. This paper presents a wearable technology prototype to monitor CKD patients between subsequent dialysis procedures. The proposed technology enables at-home monitoring of electrolyte fluctuations and detection of cardiac arrhythmias, such as ventricular tachycardia and extreme bradycardia. A patient uses a wearable wrist-worn device to record continuous photoplethysmogram and intermittent electrocardiogram signals together with a smart device, such as a tablet or a smartphone, to enter meals and medications that may alter electrolyte levels. The application of the proposed wearable technology is demonstrated in a case analysis. The developed wearable technology for monitoring CKD patients in a home environment can be valuable for identifying patients susceptible to dangerous arrhythmias due to electrolyte imbalance.
Download

Paper Nr: 37
Title:

Survival Status Prediction for Non-small Cell Lung Cancer Patients using Machine Learning

Authors:

Aishwarya Mohan and Aleksandar Jeremic

Abstract: Lung cancer is the leading cause among cancer-related deaths worldwide. Clinically, it could be divided into several groups: 1) the non-small cell lung cancer (NSCLC, 83.4%), 2) the small cell lung cancer (SCLC,13.3%), 3) not otherwise specified lung cancer (NOS,3.1%), 4) aarcoma lung carcinoma (0.2%), and 5) other specified carcinoma (0.1%). According to SEER Cancer Statistics Review, 5-year survival rate of patients with advanced non-small cell lung cancer (NSCLC) who received chemotherapy was less than 5%. Our ability to provide survival status at any time in future is important from at least two standpoints: a) from the clinical standpoint it enables clinicians to provide optimal delivery of healthcare and b) from a personal standpoint, by providing patient’s family with opportunities to plan their life ahead and potentially cope with emotional aspect of loss of life. In this paper we propose to utilize machine learning techniques to achieve this goal and evaluate several techniques in order to determine their prediction performance using publicly available dataset.
Download

Paper Nr: 43
Title:

Exploring Classification in Open and Closed Eyes EEG Data for People with Cognitive Disorders

Authors:

Ioanna Chouvarda, Lampros Mpaltadoros, Ioanna Boutziona, George N. Tsakonas, Magda Tsolaki and Konstantinos Diamantaras

Abstract: Cognitive disorders, including Alzheimer’s Disease (AD), are health issues concerning all society. The evolution of technology and Artificial Intelligence (AI)/ Machine Learning (ML) in the health domain promises an earlier and more accurate diagnosis for Alzheimer’s disease and Dementia. In this study, we examine Healthy patients and patients with AD and Mild Cognitive Impairment (MCI), often a prior step of AD. With the use of EEG, we collect data from their brain activity. After a basic processing step, kernel PCA is applied as a dimensionality reduction method using segments of the multichannel signal, and the transformation output is employed as input for the predictive model. Machine learning functions are used to classify data correctly into Healthy, AD, MCI classes, and a postprocessing step allows for classification at the patient level. The results show that the algorithm can predict with an accuracy of 90 percent and more in total, AD or MCI patients vs. Healthy patients.
Download