Proceedings of Trends in Electronics and Health Informatics

Inhaltsverzeichnis

1. Frontmatter

2. Artificial Intelligence and Soft Computing

   1. Frontmatter

   2. Experimental Study of High-Frequency Drill String Vibrations Under Different Conditions

      Vladimir Bakhtin, Mikhail Deryabin, Dmitry Kasyanov, Sergey Manakov, Denis Shakurov

      Abstract

      In this paper, the experience of measuring high-frequency drilling string vibrations during drilling is presented. Vibration registration is carried out by a borehole noise recorder which has specially been made for this purpose. Three accelerometers are built into its construction design. The measuring axes of those accelerometers form an orthogonal coordinate system. Vibrations are recorded in the frequency band from several hertz to 25 kHz. Measurements are carried out during various drilling operations, i.e., rotary drilling, directional drilling, reaming, making a connection, and circulation. The drill string includes a mud motor. Measurements were performed either during horizontal and vertical drilling. The total record involves several days and its size is approximately 250 GB. The recorded data includes oscillograms separately obtained from each of the three accelerometers and the time dependencies of temperature and internal pressure in the borehole. Since the main goal is investigating high-frequency noise during drilling, frequencies below 1 kHz are partially suppressed by a high-pass filter during registration. Some peculiarities of noise distribution in frequency domain under different drilling conditions are presented in this article. Modified Welch’s method is used for spectrum estimation.

   3. Flexible Systolic Hardware Architecture for Computing a Custom Lightweight CNN in CT Images Processing for Automated COVID-19 Diagnosis

      Paulo Aarón Aguirre-Alvarez, Javier Diaz-Carmona, Moisés Arredondo-Velázquez

      Abstract

      Millions of deaths worldwide have been resulted throughout the COVID-19 pandemic, thus the need of diagnosing techniques for early disease stage has been arisen. Although RT-PCR is the standard test to diagnose SARS-COV-2 infection, factors such as the long waiting time for results and its relatively low accuracy have led to the need of new alternative diagnosis methods. The Convolutional Neural Network (CNN), a powerful and efficient deep learning algorithm, can be applied as an automated diagnosis tool by processing chest Computed Tomography (CT) scanning images of patients with suspected infection. Recent works have shown that low-complexity CNNs accompanied by image preprocessing are sufficient to diagnose COVID-19 with high precision. This fact allows the use of low-end hardware, such as Field Programmable Gate Arrays (FPGAs), to compute these compact models in the microsecond range. In this paper, a flexible hardware architecture to compute a lightweight custom CNN to classify chest (CT) scanning images as COVID and non-COVID is proposed. This system is capable of classifying 23 CT images per second with an accuracy of up to 91% and has remarkable adaptability to different hyperparameters of the convolutional layer, as these are computed by a single systolic array-based convolver.

   4. Dimensionality Reduction in Handwritten Digit Recognition

      Mayesha Bintha Mizan, Muhammad Sayyedul Awwab, Anika Tabassum, Kazi Shahriar, Mufti Mahmud, David J. Brown, Muhammad Arifur Rahman

      Abstract

      For visualization, the concept of dimension is normally enclosed to 2–3 degrees in individuals. A computing node can extend it significantly. However, any increase in the number of dimensions usually introduces an extra computational burden, and it becomes more challenging to extract the exact information. Therefore, dimensionality reduction methods are an increasingly important area of study to help identify methods to mitigate challenges associated with high-dimensional feature sets. Handwritten digit recognition is one of the most relevant fields of study due to the variety of issues faced such as the age of texts, the professional context and norms in which the text is written, and individual differences in writing styles. Research on handwritten digit recognition using various algorithms has been conducted in a variety of languages. In the Bangla character set, there are ten digits. Due to geometry, complicated forms, and similarities between the individual numerals, individual characters are difficult to identify. Also, there are limited open datasets available to researchers to conduct Bangla digit recognition upon. This work discusses dimensionality reduction techniques used in the Bangla Handwritten Digit dataset NumtaDB. Principal Component Analysis (PCA), Neighborhood Component Analysis (NCA), and Linear Discriminant Analysis (LDA) algorithms are examined as feature extraction techniques. CNN is a deep learning technique to classify the input automatically. Over the years, CNN has found a good grip over classifying images for computer visions and now it is being used in other domains too. The numeric digits are then classified using CNN utilizing the lower dimension vectors acquired. These models can recognize most of the digits successfully with a satisfactory level of performance for identifying different digits.

   5. Obtaining Fractal Dimension for Gene Expression Time Series Using an Artificial Neural Network

      Marco Antonio Esperón Pintos, Jorge Velázquez Castro, Benito de Celis Alonso

      Abstract

      In this work, a stochastic dynamic model of a minimal gene regulatory network is used to simulate the characteristic dynamics of protein concentration. By changing the model’s parameters [11], it will be possible to simulate different cellular conditions that will be used for training some artificial neural networks that recognize the cellular state of disease of health with the information in the protein concentration series [4]. In particular, the hurst exponent and the fractal dimension of the signal will be analyzed. The hurst exponent is relevant in diagnosis as it determines the autocorrelation of the time series and allows different labeling states of cellular health. Some studies have shown that bacteria such as E. Coli and fungi such as S. Cerevisiae show a healthy cellular state when the time series of transcription factors have a hurst exponent greater than 0.5 [3]. This research evaluates the efficiency and feasibility of using an artificial neural network to diagnose cellular states by means of the dynamics of protein concentrations.

   6. Grouping by Mixture of Normals for Breast Cancer in Two Groups, Benign and Malignant

      Gerardo Martínez Guzmán, María Beatriz Bernábe Loranca, Rubén Martínez Mancilla, Carmen Cerón Garnica, Gerardo Villegas Cerón

      Abstract

      The diagnosis of cancer cells from biopsies is mainly based on the analysis of the morphological changes of the nuclear structure as the increase in nuclear size, which probably occurs due to the deregulation of cell cycle, as well as the cell growth. The increase of the nuclear size is observed in biopsies of patients with benign and malignant diagnosis. A radius_mean variable (mean of distances from the center to points on the perimeter), related with the increase of nuclear size in patients with benign and malignant diagnosis, is studied in this work. An analysis of this variable proves, by the algorithm of unsupervised learning, Expectation–maximization (EM). That said variable has a mixture of normals with two components type behavior. Such an algorithm is able to discriminate the data in two groups (malignant and benign), the model shows a 97.8% of coincidence for benign cases and 66.5% for malignant cases.

   7. A Smart Automation System for Controlling Environmental Parameters of Poultry Farms to Increase Poultry Production

      Md. Kaimujjaman, Md. Mahabub Hossain, Mst. Afroza Khatun

      Abstract

      Agriculture and poultry must be addressed as the backbone of the economic growth of any developing country like Bangladesh. Furthermore, agricultural progress and economic prosperity are inextricably linked. Technology advancements and new technical developments have ushered in a new era of real-time animal health monitoring. This study focuses on a sensor-based solution for minimal-cost, capital-saving, value-oriented, and productive chicken farm management in order to increase the value of the broiler farm economy index (BFEI). The goal of this research was to see if an Intelligent System based on an Embedded Framework could be utilized to monitor chicken farms and adjust environmental conditions using smart devices and technology. This study also looked into how different temperatures (ranges from 25 to 33 °C) affected broiler performance efficiency factor (BPEF), livability, and feed efficiency (FCR). It was discovered that the group reared at higher temperatures had greater broiler performance efficiency factor, livability, and lower feed efficiency.

   8. A New Model Evaluation Framework for Tamil Handwritten Character Recognition

      B. R. Kavitha, Noushath Shaffi, Mufti Mahmud, Faizal Hajamohideen, Priyalakshmi Narayanan

      Abstract

      The robustness of any pattern recognition model relies heavily on the availability of comprehensive samples. Until last year, the Tamil Handwritten Character Recognition (HWCR) works relied on the solitary HPL Tamil dataset [1]. Recently, a new benchmarking for Tamil HWCR was published [2] comprising 94000 samples in total. The efficiency of corroboration using multiple standardized databases is crucial in advancing any research area. Towards this aim, in this paper, we showed different ways of experimentation with these datasets. For this purpose, we utilized transfer learning, and a custom deep neural network, a recently published work for Tamil HWCR [3]. Different experimental setups were suggested that involved independent, cross-testing, and mixed modes of model building and evaluation using two standardized datasets. These setups form a rigorous testing framework for analyzing Tamil HWCR tasks. The work presented in this paper is the first to report the results of Tamil HWCR using two standardized datasets and sets a new model evaluation benchmark. For rapid reproducibility and dissemination, the code and materials used in this study are available at https://github.com/Kavitha-BR-VIT/Tamil-HWCR.

   9. Integrated Linear Regression and Random Forest Framework for E-Commerce Price Prediction of Pre-owned Vehicle

      Amit Kumar Mishra, Saurav Mallik, Viney Sharma, Shweta Paliwal, Kanad Ray

      Abstract

      The E-Commerce industry has taken the world in its stride. It has been growing at an exponential rate. This paper presents a machine learning-based model to predict the price of a used vehicle for selling and buying purposes. The model proposes two algorithms: Linear Regression and Random Forest Regression and draws a comparison between the two on the basis of standard performance measures. The uniqueness of this work is that it is not only capable to predict the price of a used vehicle but the model can also be extended with minimal effort for any kind of product across various spheres of the E-commerce industry.

   10. Personalized Recommender System for House Selection

       Suneeta Mohanty, Shweta Singh, Prasant Kumar Pattnaik

       Abstract

       Housing is the key to improved health and welfare. Hence it is required to select the house wisely. In this paper, TOPSIS method is used for the ranking of various alternatives of houses as per the user’s personalized requirements and demand.

3. Healthcare Informatics

   1. Frontmatter

   2. Epileptic Seizure Detection from EEG Signal Using ANN-LSTM Model

      Redwanul Islam, Sourav Debnath, Reana Raen, Nayeemul Islam, Torikul Islam Palash, Rahat Ali

      Abstract

      Epilepsy is a well-known neurological disease caused by malfunctioning nerve activity in the brain. These malfunctioning causes episodes called seizures. Seizures in epileptic patients involve uncontrollable movements, loss of sensation, convulsions, and loss of consciousness, which can result in catastrophic injury and even death. Therefore, a computerized seizure recognition system is important to protect epilepsy patients from the risk of seizures. The main reason for this disorder is still unknown. Though the symptoms associated with seizures can be treated manually and the accuracy of the diagnosis depends on the experience of the technician. In this paper, we presented an artificial intelligence-based approach where time–frequency characteristics of EEG signals are used to detect an epileptic seizure. Electroencephalography (EEG) is widely recognized for the diagnosis and evaluation of brain activity and disorders. We preprocessed the EEG signal and converted it into time–frequency information, then passed it through ANN-LSTM architecture for training. The proposed model achieves 99.46% classification accuracy with higher overall sensitivity (99.78%) and specificity (99.13%) and outperformed other similar methods. The findings of the study indicate that our technique has the potential for clinical application. The effectiveness of this strategy may be further assessed by combining it with other epileptic datasets.

   3. Cognitive Assessment and Trading Performance Correlations

      J. Eduardo Lugo, Jocelyn Faubert

      Abstract

      Neuroeconomics and behavioral finance have provided insight on how cognitive processes and emotions combine to influence financial decisions. In trading decision-making, cognitive assessment and its possible increase through training should be better understood. In this preliminary validation investigation, we employed NeuroTracker (3D multiple object tracking or 3D-MOT), a technique extensively used to test and train cognitive processes in performance populations, to investigate whether the metrics on this task relate to trading performance. The findings demonstrate that there are strong relationships between trading metrics and NeuroTracker scores.

   4. Molecular Docking Study of Oxido-Vanadium Complexes with Proteins Involved in Breast Cancer

      Lisset Noriega, María Eugenia Castro, Norma A. Caballero, Gabriel Merino, Francisco J. Melendez

      Abstract

      Breast cancer is the most common type of cancer detected in women worldwide; therefore, identifying compounds with anticancer activity is a research field that can contribute to dealing with this disease. Vanadium complexes are highlighted due to their versatility in medical issues. In this study, three oxido-vanadium (V) complexes were studied with the aim of elucidating their binding to therapeutic protein targets Cdc25b, AKT2, and SHP2. The interaction energies between the proteins and the three complexes were between −6 and −9 kcal/mol, and the interactions observed were mainly hydrogen bond and hydrophobic interactions.

   5. Multi-Level Stress Detection using Ensemble Filter-based Feature Selection Method

      Arham Reza, Pawan Kumar Singh, Mufti Mahmud, David J Brown, Ram Sarkar

      Abstract

      Stress has become one of the major concerns in modern human life, especially after the outbreak of the COVID-19 pandemic, and it has had a great impact on human daily life activities. Detecting stress from physiological signals at an early stage is crucial as it prevents it from outgrowing severe health issues. Most researchers interested in stress detection have focused on developing new feature extraction methods. In this paper, at first, we have extracted some common statistical features from raw data. Then to remove redundant features, we have proposed an ensemble of filter-based feature selection methods for stress detection. Two filter methods, namely, Mutual Information and Pearson Correlation Coefficient are used to obtain the rank of the features. Based on the selected features, three popular classification models, namely, Decision Tree, Random Forest, and K-nearest neighbors are used for the detection of four stress classes—baseline, stress, amusement, and meditation). The proposed method has been applied to the publicly available standard WESAD dataset which consists of various physiological signals taken from both chest and wrist. We have achieved classification accuracies of 99.9% and 96.8% for subject-dependent and subject-independent cases, respectively.

   6. A Hybrid Transfer Learning and Segmentation Approach for the Detection of Acute Lymphoblastic Leukemia

      Ang Jia Hau, Nazia Hameed, Adam Walker, Md. Mahmudul Hasan

      Abstract

      Acute Lymphoblastic Leukemia (ALL) is a subcategory of leukemia which is a type of blood cancer. ALL is also the most commonly diagnosed form of pediatric cancer, with early detection of the cancer playing a crucial role in the survivability of the diagnosed patient. This paper therefore proposes a Hybrid Transfer Learning eXtreme Gradient Boosting (HTL-XGB) algorithm which exploits Transfer Learning and through utilizing modern developments of state-of-the-art CNNs (Convolutional Neural Networks) effective ability for feature extraction, combined with the robust performance of eXtreme Gradient Boosting for the classification and detection of ALL. An object detection methodology is proposed for the individual detection of activated and non-activated lymphocytes within full blood smear images, using image processing techniques with the proposed HTL-XGB architecture. Several state-of-the-art CNNs were experimented through fair trials, with Xception achieving the highest accuracy of 89.7%.

   7. Logistic Regression Approach to a Joint Classification and Feature Selection in Lung Cancer Screening Using CPRD Data

      Yuan Shen, Jaspreet Kaur, Mufti Mahmud, David J. Brown, Jun He, Muhammad Arifur Rahman, David R. Baldwin, Emma O’Dowd, Richard B. Hubbard

      Abstract

      Lung cancer is one of the most deadly cancers in the world. Its mortality rate is high when the cancer is diagnosed late. Therefore, early detection is a crucial factor for an increase in survival rate, and lung cancer screening is one of the most important intervention tools. However, the screening would be cost-effective only when we can accurately select a sub-population which is at the most risk of lung cancer. It is hypothesised that this selection task can be done cost-effectively when we use clinical data (e.g. demographic, lifestyle and comorbidity variables) rather than low-dose CT. This work used the clinical data extracted from Clinical Practice Research Datalink (CPRD). The goal is to test whether this approach can achieve comparable or even better selection performance when compared to an alternative approach using clinical data from lung cancer screening trials. The latter approach is adopted in [54]. In this paper, we further adapt the logistic regression model for a joint classification and feature selection analysis. The model is implemented in an ‘ensemble learning’ manner to deal with severe ‘class imbalance’ problems. It is observed that the sensitivity and specificity results are slightly better than those reported in [54]. Also, we identified a comorbidity factor COPD and a smoking-related factor smk-status as the two most discriminative features.

   8. HI Applications for ADHD Children: A Case for Enhanced Visual Representations Using Novel and Adapted Guidelines

      Sandesh Sanjeev Phalke, Abhishek Shrivastava

      Abstract

      An effective representational style in a health informatics (HI) application enhances the attention span of children living with ADHD (ChADHD) in a typical learning environment. Designers of HI applications for ChADHD require relevant visual design guidelines to create effective representations. However, observations, such as visual design guidelines, are often unavailable, or are present in a format unsuitable for designers. Subsequently, representations suffer from imperfections born out of either the absence of information or the misinterpretation of existing knowledge. This study sheds more light on this scenario through working with two specific groups: designers and remedial experts. We establish gaps in knowledge through a set of interviews and focus group sessions. We bring in details of concerns raised by designers, and find relevant visual guidelines and/or contradictions in the existing ones, if any. Finally, we conclude by summarizing a mix of existing (but adapted) and novel guidelines to help design appropriate representations in ChADHD applications.