Predictive Modeling in Biomedical Data Mining and Analysis

By Valentina Emilia Balas and Mamta Mittal

Predictive Modeling in Biomedical Data Mining and Analysis presents major technical advancements and research findings in the field of machine learning in biomedical image and data analysis. The book examines recent technologies and studies in preclinical and clinical practice in computational intelligence. The authors present leading-edge research in the science of processing, analyzing and utilizing all aspects of advanced computational machine learning in biomedical image and data analysis. As the application of machine learning is spreading to a variety of biomedical problems, including automatic image segmentation, image classification, disease classification, fundamental biological processes, and treatments, this is an ideal reference.

Machine Learning techniques are used as predictive models for many types of applications, including biomedical applications. These techniques have shown impressive results across a variety of domains in biomedical engineering research. Biology and medicine are data-rich disciplines, but the data are complex and often ill-understood, hence the need for new resources and information.

• Includes predictive modeling algorithms for both Supervised Learning and Unsupervised Learning for medical diagnosis, data summarization and pattern identification
• Offers complete coverage of predictive modeling in biomedical applications, including data visualization, information retrieval, data mining, image pre-processing and segmentation, mathematical models and deep neural networks
• Provides readers with leading-edge coverage of biomedical data processing, including high dimension data, data reduction, clinical decision-making, deep machine learning in large data sets, multimodal, multi-task, and transfer learning, as well as machine learning with Internet of Biomedical Things applications

• Language: English
• Publisher: Academic Press
• Release date: Aug 28, 2022
• ISBN: 9780323914451

Book preview

Preface

Sudipta Roy, Maharashtra, India

Lalit Mohan Goyal, Faridabad, India

Valentina E. Balas, Romania

Basant Agarwal, Jaipur, India

Mamta Mittal, New Delhi, India

Machine learning (ML) techniques are used as predictive models for many applications including those in the field of biomedicine. These techniques have shown impressive results across a variety of domains in biomedical engineering research. Biology and medicine are data-rich disciplines, but the data are complex and often not properly understood. Most biomedical data are categorized into structured, semi-structured, and unstructured types with very high volume. The volume and complexity of these data present new opportunities, but also pose new challenges. Automated algorithms that extract meaningful patterns could lead to actionable knowledge and change how we develop treatments, categorize patients, or study diseases, all within privacy-critical environments. This book addresses the issues described to predict and model biomedical data mining and analysis. The book has been organized into 15 chapters.

Chapter 1 titled Data Mining with Deep Learning in Biomedical Data presents a time-domain study of EEG signals to detect various neurological disorders with a specific focus on epilepsy. The presented convolutional neural network (CNN), long short-term memory network (LSTM), and CNN-LSTM hybrid models were used to detect seizure activities in precisely filtered EEG segments. The experimental results reveal the suitability of the CNN-LSTM hybrid model for accurate and prompt detection of epileptic seizures with an accuracy of 98%, sensitivity of 98.48%, and specificity of 99.19%, so that patients could be saved from major injuries or sudden expected deaths. These models can be useful in the detection of various diseases or disorders such as schizophrenia, Parkinson's disease, and the identification of breast cancer and bone- or skin-related diseases.

Chapter 2 titled Applications of Supervised Machine Learning Techniques with the Goal of Medical Analysis and Prediction: A Case Study of Breast Cancer analyzes the Wisconsin Breast Cancer Diagnosis dataset for identifying essential features and assessing the performance of some popular machine learning (ML) classifiers in breast cancer prediction. The dataset is first cleaned by eliminating non-numerical values and normalizing the data. The processed data are then visualized to grasp the hidden patterns and non-essential attributes are trimmed. Eight different ML models are trained and tested over the refined data for prediction of the two tumor classes. The presented study identified vital features that were must-haves for the analysis, and the empirical results investigated the superiority of particular ML classifiers over the others.

Chapter 3 titled Medical Decision Support System Using Data Mining describes how a medical decision support system can support the medical decision-making processes at both clinical and diagnostic levels. To provide an error-free and accurate service, clinicians must apply relevant computer-based information and decision support systems. Decision support systems can be designed as a system based on knowledge or a system based on learning. Human-engineered mappings to suggestions based on best medical treatments and patient data are known as knowledge-based systems. Learning-based systems utilize data mining, statistics, and ML approaches to map the system. Integrated decision support incorporates both the systems of knowledge and learning to solve the problem of presence of partial information in a realistic situation. This effort aims to assist physicians medically and to apply the medicine prescription specifically. The approach may be utilized for query-based applications, online web browser applications, or mobile applications on numerous terminal interfaces.

Chapter 4 titled Role of AI Techniques in Enhancing Multi-Modality Medical Image Fusion Results outlines the benefits of using AI methods for medical image fusion of different modalities. The modality can be computed tomography, magnetic resonance-T1, magnetic resonance-T2, and Positron emission tomography depending on the suspected malignant region. The aim of fusion is to collaborate each modality’s best information into a single image called a fused image. This chapter addresses the multi-modality medical image fusion using AI techniques like Fuzzy Logic and Adaptive Neuro-Fuzzy Inference System (ANFIS). The study reveals that the AI techniques not only give better results but their learning capabilities will likely make the future work self-driven.

Chapter 5 titled A Comparative Performance Analysis of Backpropagation Training Optimizers to Estimate Clinical Gait Mechanics indicates that the clinical gait analysis of healthy people of different age groups plays a significant role in the early estimation of different physiological and neurological disorders. However, due to complicated data acquisition setups and in-person requirements, the estimation of the gait analysis has been quite tough to follow. To avoid such issues, a ML-based approach has been proposed in this work to estimate the biomechanical gait parameters. Three backpropagation neural network models with Levenberg-Marquardt method, resilient backpropagation method, and gradient descent method optimizers have been designed to estimate the joint angles, joint moments, and ground reaction forces in the sagittal plane. The dataset used in the neural network models has been taken from an open-source repository. The anthropometric, biological, and spatiotemporal parameters of 50 different subjects have been exploited as input dataset.

Chapter 6 titled High-Performance Medicine in Cognitive Impairment: Brain–Computer Interfacing for Prodromal Alzheimer’s Disease suggests that Alzheimer’s disease is frequently misdiagnosed as normal aging because it has always been difficult to detect early on. Mild cognitive impairment (MCI) can be identified, but there is little that can be done at that time because no medicine can reverse the effect of MCI; instead, it can only slow down the progression. Alzheimer’s disease is difficult to diagnose medically, especially in its early stages. As a response, a method for early diagnosis of Alzheimer’s disease is urgently needed even now. In this chapter, the authors have proposed a strategy for detecting Alzheimer’s disease in its early stage using noninvasive brain-computer interface technology. Electroencephalography (EEG) brain wave patterns were used for three groups (Alzheimer’s disease—AD, mild cognitive impairment—MCI, and healthy subjects—HS) of test subjects in this research. The proposed framework was evaluated with 46 test subjects, with an accuracy of 86.47% and a precision of 0.801.

Chapter 7 titled Brain Tumor Classifications by Gradient and XG Bosting Machine Learning Models describes the use of the boosting-type ML algorithms to evaluate the model performance parameters. Model performance is validated using K-fold methods and preliminary results indicate that the XG boosting algorithm yields the highest classification accuracy. Evaluations of this type are largely supportive of biomedical imaging studies and there is scope for future studies using other classification models for achieving the highest prediction accuracy.

Chapter 8 titled Biofeedback Method for Human–Computer Interaction to Improve Elder Caring: Eye-Gaze Tracking proposes how physiological methods of eye-gaze tracking could be used to design and develop natural user interaction techniques. A human user’s tacit intention to use physiological signals for the domestic area’s required activities/requirements may be understood by utilizing nonverbal contact to define the user’s intention to use physiological signals for the domestic area’s necessary activities/requirements. To achieve good accuracy and robustness, traditional gaze monitoring systems depend on explicit infrared lights and high-resolution cameras. Recent advancements in mobile devices, as well as an increasing interest in recording normal human behavior, have shown that tracking eye motions in a non-restricted environment could yield promising results.

Chapter 9 titled Prediction of Blood Screening Parameters for Preliminary Analysis Using Neural Networks describes various techniques used in the prediction of blood parameters. The prediction of blood screening test features using the backpropagation neural network is presented in detail. The features used in this chapter were fibrinogen and globulin. The normal ranges of fibrinogen and globulin are 2–4 g/L and 20–35 g/L, respectively. Fibrinogen is a glycoprotein that circulates in the blood of all vertebrates. It is observed from the results that the prediction accuracy for fibrinogen is better than that for globulin. To increase the accuracy of the prediction for globulin, the training parameters and activation functions must be modified.

Chapter 10 titled Classification of Hypertension Using an Improved Unsupervised Learning Technique and Image Processing presents an improved nearest neighbor distance clustering algorithm by recognizing the lesions present in the retina. The current approach identifies the symptoms associated with retinopathy for hypertension and classifies the hypertensive retinopathy. This chapter provides an assessment of the hypertensive retinopathy recognition techniques that apply a range of image processing procedures used for feature extraction and classification. The chapter also specifies the existing open databases, containing eye fundus images, which can be used for hypertensive retinopathy research.

Chapter 11 titled Biomedical Data Visualization and Clinical Decision-Making in Rodents Using a Multi-usage Wireless Brain Stimulator With a Novel Embedded Design describes in detail the complete design, biomedical data visualization, and modeling aspects of the stimulator device. The feasibility of this device is successfully tested in in vivo and in vitro stages for a period of more than a month. This embedded design has been developed taking into account cost-effectiveness, user-friendliness, and precision, which are the main focus of this chapter. The brain-computer interface can be useful in taking effective clinical decision-making at an early stage. However, there is limited research in this area so far. Therefore, all the efforts in this direction are extremely important for numerous young flourishing specialists, and aspirations toward the brain-computer interface.

Chapter 12 titled LSTM Neural Network-Based Classification of Sensory Signals for Healthy and Unhealthy Gait Assessment describes the modeling of the long short-term memory (LSTM) deep neural network model and its implementation to classify healthy and unhealthy gait based on a sensory dataset. The reference sensory dataset of 22 subject samples (11 healthy and 11 with knee pathology) is taken from the UCI Irvine Machine Learning Repository. Two different optimizers, namely Stochastic Gradient Descent and Adam, have been exploited in the designed LSTM model with different sets of learning hyperparameters. The classification results of the proposed deep learning model with both optimizers have been compared with each other using several performance measures like precision, recall, and F1 score.

Chapter 13 titled Data-Driven Machine Learning: A New Approach to Process and Utilize Biomedical Data includes a study of precise and accurate diagnostic tools to ease the pressure on medical personnel, simultaneously enhancing efficiency. This chapter explores the development of artificial neural network based diagnostic tools that focus on the challenges described previously. A brief overview of the current scenarios and future prospects of ML in biomedicine is also presented.

Chapter 14 titled Multiobjective Evolutionary Algorithm Based on Decomposition for Feature Selection in Medical Diagnosis presents a mathematical model of a multi-objective evolutionary algorithm based on decomposition (MOEA/D) and its application in feature selection in medical diagnosis. Most of the medical datasets are high dimensional in nature and so there is a need for optimal feature selection, which is a difficult problem. The negative influence may be due to the possibility of irrelevant or many redundant features. Intelligent models including classification, clustering, regression, and boosting techniques are helpful in extracting useful knowledge. The performance of the MOEA/D method is compared with that of state-of-the-art multi-objective optimization methods when applied to most of the datasets.

Chapter 15 titled Machine Learning Techniques in Healthcare Informatics: Showcasing Prediction of Type 2 Diabetes Mellitus Disease using Lifestyle Data focuses on the role of the ML paradigms in healthcare analytics and presents the implementation of the framework for developing ML models for type 2 diabetes mellitus (T2DM) disease. In this chapter, lifestyle indicators rather than clinical/pathological parameters have been used for the prediction of T2DM. The study involves different experts like diabetologists, endocrinologists, dieticians, and nutritionists for selecting the contributing lifestyle parameters to promote health and manage diabetes. The study aims to develop an intelligent knowledge-based system for the prediction of T2DM without conducting clinical tests. It can save the patient undue delays caused by unnecessary readmissions and pathological tests in hospitals. The proposed work emphasizes the use of ML techniques, namely K-nearest neighbor (KNN), logistic regression (LR), naïve Bayes (NB), support vector machine (SVM), decision tree (DT), random forest (RF), and artificial neural network (ANN), for the prediction of T2DM disease. The RF technique attained the highest accuracy of 93.56% followed by DT, LR, SVM, NB, ANN, and KNN with accuracies of 92.70%, 91.41%, 90.98%, 89.27%, 87.98%, and 84.54%, respectively.

We are grateful to Elsevier, especially Chris Katsaropoulos, Senior Acquisitions Editor, for providing us the opportunity to edit this book.

1: Data mining with deep learning in biomedical data

Kuldeep Singha; Jyoteesh Malhotrab    a Department of Electronics Technology, Guru Nanak Dev University Amritsar, Punjab, India

b Department of Engineering and Technology, Guru Nanak Dev University Regional Campus Jalandhar, Punjab, India

Abstract

In the era of technological advances, the health-care sector is going through a ground-breaking transition by shifting the traditional approach of physical examination of patients to remote patient monitoring. For this purpose, various machine learning techniques are employed to analyze the biomedical signals acquired from a patient's body for the detection or prediction of health-related disorders. Nowadays, deep learning techniques are widely used in this sector due to their feature extraction independence, ability to manage massive volumes of biomedical signals, including electroencephalogram (EEG) or electrocardiograms, and to deal with their nonstationary character. This chapter presents the time-domain study of EEG signals to detect various neurological disorders with a keen focus on epilepsy. The present work employs the proposed architectures of convolutional neural network (CNN), long short-term memory network (LSTM), and CNN-LSTM hybrid models to detect seizure activities in precisely filtered EEG segments. The experimental results reveal the suitability of CNN-LSTM hybrid model for accurate and prompt detection of epileptic seizures with an accuracy of 98%, sensitivity of 98.48%, and specificity of 99.19%, so that patients could be saved from major injuries or sudden expected deaths.

Keywords

Data mining; Deep learning; Epilepsy; EEG; Health care; Signal analysis

1: Introduction

In the era of Internet of things (IoT) technologies, smart health care is an emerging sector that is attracting the attention of medical personnel, the research community, and patients [1, 2]. These technological advances in association with machine learning and cloud-fog computing capabilities have started revolutionizing the health-care sector by shifting the traditional patient monitoring approach to remote patient monitoring [3]. In this sector, biomedical data analysis is crucial in the detection and diagnosis of a variety of health-related issues such as bacterial and viral infectious diseases; neurological and mental disorders, particularly, epilepsy, schizophrenia, Alzheimer's disease, etc.; cardiovascular diseases; autoimmune diseases; cancer; and skin- or bone-related diseases [2, 4–10]. The biomedical data may include electroencephalogram (EEG) or electrocardiogram (ECG) signals, X-ray, CT scan, MR-based images or microscopic images, etc., which could be analyzed using machine learning or deep learning-based signal analysis techniques [10–13].

Among the aforementioned diseases, neurological and mental disorders are one of the most serious hazards to public health [4]. These disorders have become one of the main causes of disabilities and deaths globally. The social and economic burden of these disorders is more severe in underdeveloped or impoverished countries due to a scarcity of health-care infrastructure. This burden is likely to grow rapidly in forthcoming years as a result of an intensive increase in population and aging [14]. Modern IoT-enabled health-care technologies may be useful in detecting and predicting these neurological and mental problems to save patients’