International Journal of Biomedical Engineering and Technology (15 papers in press)

Regular Issues

• S-R2F2U-Net: A single-stage model for teeth segmentation  
  by Mrinal Kanti Dhar, Mou Deb 
  Abstract: Precision tooth segmentation is crucial in the oral sector because it provides location information for orthodontic therapy, clinical diagnosis, and surgical treatments. In this paper, we investigate residual, recurrent, and attention networks to segment teeth from panoramic dental images. Based on our findings, we suggest three models: single recurrent R2U-Net (S-R2U-Net), single recurrent filter double R2U-Net (S-R2F2U-Net), and single recurrent attention enabled filter double (S-R2F2-Attn-U-Net). Particularly, S-R2F2U-Net, as emphasised in the title, outperforms state-of-the-art models in terms of accuracy and dice score. A hybrid loss function combining cross-entropy loss and dice loss is used in training. In addition, it reduces around 45% of model parameters compared to the original R2U-Net. Models are trained and evaluated on the UFBA-UESC dataset that contains 1,500 extra-oral panoramic X-ray images and divided into ten categories based on the structural variations. S-R2F2U-Net achieves 97.31% accuracy and 93.26% dice score. Codes are available on https://github.com/mrinal054/teethSeg_sr2f2u-net.
  Keywords: tooth segmentation; semantic segmentation; deep learning; recurrent module; attention module.
  DOI: 10.1504/IJBET.2024.10063262
   
  
• Alterations of generic musculoskeletal models to incorporate realistic knee joint and muscle geometry for biomechanical analyses during healthy gait: A Narrative Literature Review  
  by Shivangi Giri, Ravi Prakash Tewari 
  Abstract: Knee is an important weight-bearing 6-degree-of-freedom (DOF) joint that is essential for stable locomotion. However, the majority of lower limb musculoskeletal (MSK) models only include one DOF and hence fail to represent the true biomechanics of the knee. For models with multi-DOF knee anatomically realistic modelling of muscle architecture is crucial in representing the mechanical stability of the entire limb. The purpose of this narrative review, therefore, was to report state-of-the-art knowledge on the existing generic rigid-body MSK models that have incorporated: 1) multi-DOF knee; 2) multi-line characterisation of muscle geometry, to analyse normal healthy human gait. 15 studies accommodated multi-DOF knee joint, however majority of them retained the single-line oversimplified muscle geometry. Those that focused on better characterising muscle geometry (n = 8) used only single-DOF knee joints. Most importantly, this review showed that no generic MSK model exists that incorporates realistic representations of both knee and muscle volume.
  Keywords: musculoskeletal modelling; knee; muscle redundancy; review; biomechanics.
  DOI: 10.1504/IJBET.2024.10063535
   
  
• A Deep Learning approach for the augmented diagnosis and prediction of infectious Lung Diseases  
  by Geetha R, Umarani Srikanth, Kamalanaban E 
  Abstract: The pandemic coronavirus is an alarming threat to public health nowadays causing severe acute lung and bronchial infection that incurs a high fatality rate in humans. Researchers vigorously work in this area to find solutions for this critical issue by all means. On the other hand, tests to be carried out to determine the survival time of coronavirus infection across different communities of the population is a long-term need. Herein, this research describes a robust deep neural network to diagnose the suspicious patient’s chest X-ray (CXR) in detecting the presence of infection rapidly. This simple and rapid scalable approach has the capacity of immediate application in coronavirus diagnosis as well as predicting the spread and infection probability for every individual put under-diagnosis depending upon their health and societal parameters. Our robust deep neural network yields the best result of 97.87% accuracy and is user-friendly compared to existing methods.
  Keywords: deep learning; convolutional neural network; CNN; chest X-ray; CXR; rectified linear units; ReLUs; infectious lung disease.
  DOI: 10.1504/IJBET.2024.10063639
   
  
• Development and Validation of Virtual Reality Combined with Shoulder Wheel Device for Active Rehabilitation Training  
  by Mohhamad Reza Hosseini, Hamid Khabiri, Hamid Sharini, Vahab Dehlaghi, Ali Safarpoor 
  Abstract: In light of prevalent shoulder impairments among stroke survivors, this study aimed to evaluate the impact of a virtual reality-based rehabilitation system on enhancing active rehabilitation time during shoulder exercises. Thirty stroke patients were divided into two groups: one received conventional shoulder wheel therapy combined with virtual reality, while control group received only shoulder wheel therapy. Clinical assessments, including action research arm test and Fugl-Meyer Assessment, along with functional testing (torque) were conducted three times over a 15-day interval, with a reaction rate test administered at the end of rehabilitation. Results demonstrated that virtual reality-based rehabilitation significantly improved torque, reaction rate, and the Fugl-Meyer Assessment functional test compared to the control group (p<0.05). However, no significant difference was observed in the Action Research Arm Test assessments. This study suggests that virtual reality can play a crucial role in enhancing shoulder functions and increasing active rehabilitation time for stroke patients.
  Keywords: motor impairment; shoulder wheel; elbow and shoulder rehabilitation; virtual reality; motivational environment; active rehabilitation.
  DOI: 10.1504/IJBET.2024.10063958
   
  
• Improved Diagnosis of Lung Cancer Classification based on Deep learning Method  
  by Amel Feroui, Meriem Saim, Mohammed El Amine Lazouni, Sihem Amel Lazzouni, Zineb Aziza Elaouaber, Mahammed Messadi 
  Abstract: Lung cancer, a globally impactful and severe disease, affects millions worldwide. Uncontrolled growth of abnormal lung tissue cells results in severe complications and high mortality. Early detection is crucial for improved prognosis and survival rates. This study presents two methods for lung cancer classification utilising computed tomography (CT) images, which offer detailed scans of the lungs. The first method employs the VGG16 and VGG19 deep learning architectures. The second method utilises pre-trained VGG16 and VGG19 models for feature extraction, followed by training of supervised learning algorithms SVM, k-NN, and decision (DT) for classification. Evaluation of the proposed methods was conducted on two publicly available databases: the LIDC-IDRI database and the IQ-OTH/NCCD database. The results demonstrated that the VGG19 architecture outperforms the VGG16 architecture in terms of accuracy and precision across both databases. However, VGG16 excels on a hybrid database. Additionally, the k-NN classifier outperforms the SVM and decision tree classifiers, indicating the superiority of transfer learning over deep learning for lung cancer image classification. The proposed system has potential implications for improving patient outcomes through early detection and diagnosis.
  Keywords: lung cancer; CT scan images; deep learning; machine learning; biomedical engineering; artificial intelligence; medical imaging; classification; LIDC-IDRI database; IQ-OTH/NCCD database.
  DOI: 10.1504/IJBET.2024.10064213
   
  
• Surface and Hardness Profiles Of Additively and Conventionally Manufactured CoCrMo Alloy in Dental Application: A Preliminary Analysis  
  by Ahmad Syamil Shahruddin, Muhammad Hussain Ismail, Aini Hayati Abdul Rahim 
  Abstract: Denture has long been a viable and versatile option to replace missing teeth. Denture frameworks are made of cobalt-based alloy conventionally fabricated through casting. As technology evolves, additive manufacturing is used to manufacture metal products allowing denture fabrication to be more predictable. Many aspects of additively manufactured denture frameworks require investigation for it to be clinically acceptable. In this study, denture framework samples from SLM and cast manufacturing underwent sample preparation and were analysed for surface roughness using Alicona infinite focus and hardness tests using Vickers hardness tester, and Mitutoyo MVK-H1. The powder and the raw ingot were analysed for better understanding of the material's microstructure. This study aims to compare the surface roughness and hardness profile of cobalt-chromium denture frameworks produced by metal casting and additive manufacturing. A comparison of hardness and surface roughness for various manufacturing techniques is presented.
  Keywords: additive manufacturing; alloy denture framework; CoCrMo; hardness; surface roughness; selective laser melting.
  DOI: 10.1504/IJBET.2024.10064260
   
  
• A Systematic Review of Deep Learning Algorithms Utilising Electroencephalography Signals for Epileptic Seizure Detection  
  by Sunil Choudhary, Tushar Kanti Bera 
  Abstract: Researchers are diligently endeavouring to integrate artificial intelligence (AI) into medical practice to harness the most recent breakthroughs in these fields. Early identification and accurate disease prediction are the primary goals of healthcare to administer the efficient preventative care at any disease or critical illness like epilepsy. The epilepsy is found as a condition that is marked by repeated and unpredictable seizure-activity. The difficulty of accurately predicting epileptic seizures has not been fully solved yet. Recently, the AI tools, have been utilised to help the doctors by providing the disease information extracted from the patient';s datasets. This paper discusses the applications of deep learning algorithms (DLA) for epileptic seizure detection utilising electroencephalography (EEG) signals. The significant obstacles associated with accurately detecting automated epileptic seizures have also been studied using DLA in conjunction with EEG-data. It also analyses the advantages, challenges and limitations of the DLA applied for epileptic-seizure detection (ESD).
  Keywords: EEG signals; classification; deep learning; DL; machine learning; ML; epileptic seizures; detection and diagnosis; deep learning algorithms; DLA.
  DOI: 10.1504/IJBET.2024.10064735
   
  
• Skull Part Relationships and Shape Prediction Toward the Missing Part Completion  
  by Tan-Nhu Nguyen, Ngoc-Bich Le, Xuan-Hien Quach-Nguyen, Thi-Hiep Nguyen, Van-Toi Vo, Tien-Tuan Dao 
  Abstract: Accurate cranial reconstruction needs need clear relation among skull parts due to the asymmetry of the skull structures. Consequently, this study investigated the relation among skull parts for enhancing the skull missing part prediction. The relationship was trained from three-dimensional skull shapes reconstructed from 329 head-and-neck computed tomography images. We automatically defined the skull parts throughout all skull shapes. The skull parts were parameterised using the principal component analysis (PCA). Skull part relations were trained through their PCA-based shape parameters. The output skull parts could be predicted from the input skull parts with the trained shape relation with good and acceptable accuracy in cranial reconstruction. The best and worst mean errors were 1.32 mm and 2.54 mm when the number of missing skull parts was one and ten, respectively. The investigated procedure was employed in a computer-aided system for automatically predicting and printing skull missing parts directly in 3D spaces.
  Keywords: skull part relationship; skull shape prediction; statistical shape modelling; skull part fixing; cranial reconstruction.
  DOI: 10.1504/IJBET.2024.10065355
   
  
• Detection of Acute Lymphoblastic Leukemia Using Extreme Learning Machine based on Deep Features from Microscopic Blood Cell Images  
  by Sunita Chand 
  Abstract: Leukaemia is the medical term for blood cancer. This paper proposes an automatic disease diagnosis model to detect leukaemia from microscopic blood cell images by classifying these images into malignant and benign cells. It uses extreme learning machine (ELM) as the classifier and uses the transfer learning on AlexNet to obtain the 4,096 features required to train the classifier. The training of AlexNet is performed on 864 and 2,080 images, obtained after augmentation. The experiments are repeated five times each for nine different values of number of hidden neurons in the hidden layer of the classifier, to obtain nine average accuracies. The best average accuracy obtained for IDB1 is 99.4% at 3,000 and 4,500 hidden neurons, while for IDB2, it is 99.8% at 3,500 hidden neurons. The grand average is calculated over these nine averages and is found to be 98.6% and 99.2% for IDB1 and IDB2 respectively, while obtaining best accuracy as 100% for both the datasets.
  Keywords: Extreme Learning Machine; Deep Neural Network; Feature Extraction; AlexNet; Transfer Learning; Image Augmentation.
  DOI: 10.1504/IJBET.2024.10065360
   
  
• Design of a Bionic Arm using EMG Signal Processing and Artificial Intelligence  
  by Olusola Kunle Akinde, Oreoluwa V. Akanbi, Oluseyi A. Adeyemi 
  Abstract: Bionic arm, its development for those who are disadvantaged is the focus of this work. Electrocardiogram (ECG) was utilised in the work for the reception of brain electromyography (EMG) signal functions to control the arm which is later deciphered using digital signal processing (DSP) through empirical mathematical equations to distinguish the EMG signals from different finger movements giving more room for recognising more complex finger movements rather than the commonly used method for transcription of EMG signals. Furthermore, utilising computer vision with python, another functional mode of operation was implemented for technical testing of the arm, controlled through the receptive movements of human fingers read from a personal computer (PC) camera. The architecture employed in this work generally improves the response time of the bionic arm through reduction in time needed to read and translate EMG signals and error calculations before primary usage by the amputees to prevent issues during usage.
  Keywords: Arduino; electromyogram; EMG; digital signal processing; DSP; bionic arm; OpenCV; electrocardiogram; ECG.
  DOI: 10.1504/IJBET.2024.10065936
   
  
• Comparative Biomechanical Analysis of Lumbar spine in Trainees with Varied Barbell Positions: A Finite Element Study  
  by Diwei Chen, Datao Xu, Huiyu Zhou, Xuanzhen Cen, Yang Song, Dong Sun 
  Abstract: Considering various training goals, barbell squat variations like high bar back squat (HBBS) and front barbell squat (FBS) are popular. However, it's crucial to understand their differing effects on lumbar spine biomechanics. The objective is to comprehensively investigate the biomechanical mechanisms of lumbar during squatting movements. Data from 16 participants underwent analysis using musculoskeletal and finite element models. In the FBS, the lumbar vertebral joints exhibit a significantly greater extension angle compared to HBBS. Additionally, significant disparities in extension moment of the lumbar vertebral joints are observed between HBBS and FBS during the 0-26% and 34-77% phases of the movement. Moreover, Von Mises stress values on both the vertebrae and intervertebral discs are lower than those experienced in HBBS. The research results indicate that, with a focus on lumbar spine protection, FBS can effectively reduce the load on the lumbar spine under comparable conditions.
  Keywords: finite element model; lumber spine; squat exercise; lower back pain; kinetics; training injuries.
  DOI: 10.1504/IJBET.2024.10065991
   
  
• Wavelet-Based Methodology for Non-Invasive Detection and Multiclass Classification of Voice Disorders: A Comprehensive Study across Multilingual Datasets  
  by Avinash Shrivas, Shrinivas Deshpande, Girish Gidaye 
  Abstract: Impaired voice function affects 1.2% of the global population and is often diagnosed through invasive procedures. Past efforts in automated voice disorder detection mainly tackled the binary "healthy vs. unhealthy" classification. In this study, we suggest a non-invasive alternative based on speech analysis, diverging from the conventional invasive surgical methods. Both binary and multiclass classification is carried out in the present work by decomposing the speech signal extracted from German, Spanish, English, and Arabic datasets using discrete wavelet transform (DWT). The impact of varying decomposition levels on detection and classification accuracy is evident, with the fifth level of decomposition demonstrating the highest recognition rate of 90% to 99% for tasks involving voice disorder identification and multiclass classification. Results indicate that energy and statistical features derived from DWT offer richer information on pathological voices. Consequently, the proposed system could serve as a valuable adjunct for clinical diagnosis of laryngeal pathologies.
  Keywords: Voice disorder; wavelet transform; statistical features; multiclass classification.
  DOI: 10.1504/IJBET.2024.10066255
   
  
• Automatic Part Segmentation for Full Newborn Skeleton-Articulated Geometries using Geometric Deep Learning and 3D Point Cloud  
  by Morgane Ferrandini, Duc-Phong Nguyen, Hien-Duyen Le-Nguyen, Vi-Do Tran, Hoai-Danh Vo, Tan-Nhu Nguyen, Tien-Tuan Dao 
  Abstract: The development of the maternal pelvis model including a detailed foetal model with articulated joints is of great clinical relevance. The objective of the present study is to propose an automatic and fast segmentation workflow of the full newborn skeleton using geometric deep learning. Computed tomography scans of 124 newborn were retrieved and manually segmented. PointNet++, a geometric deep learning algorithm, was trained to perform automatic segmentation on the 3D point clouds of the reconstructed skeletons. This method was compared with the k-means clustering approach. The PointNet++ model provided highly accurate results, with an accuracy of 95.7
  Keywords: skeleton part segmentation; newborn bones; geometric deep learning; CT scan; 3D point cloud.
  DOI: 10.1504/IJBET.2024.10066351
   
  
• Synergistic Augmentation of EtOH and 4-Watt Ultraviolet-C for Rapid Surface Decontamination  
  by Jahanzeb Sheikh, Tan Tian Swee, Syafiqah Saidin, Chua Lee Suan, Sameen Malik 
  Abstract: Bacterial contamination poses significant health risks, especially in densely populated settings like educational institutions. This study in a Malaysian educational institute examined bacterial deposition on frequently touched surfaces and evaluated the efficacy of 70% ethanol (EtOH) with 10-s of contact time, combined with ultraviolet (UV) light irradiation under varied time exposures durations. Results showed that EtOH-only treatment was least effective on lift-2, with a 20% inactivation rate, while other surfaces revealed efficiencies between 69.19% and 84.4%. However, employment of EtOH-UV treatment achieved highest inactivation across all the samples treated within 60-s requiring 0.15 mJ/cm2 of dose. However, swab obtained from lift-1 could sustain 1.41-log10 inactivation under maximum exposure settings. Scanning electron microscopy (SEM) further validated the persistence of Bacillus spp, Staphylococcus spp, and E. coli colonies. This study underscores the need for comprehensive disinfection strategies in educational facilities to reduce bacterial contamination, highlighting the enhanced efficacy of the combined EtOH-UV treatment.
  Keywords: bacteria; decontamination; disinfectants; environment; high-touch surface; low-touch surface; pathogens; Ultraviolet-C.
  DOI: 10.1504/IJBET.2024.10066427
   
  
• Cardiac Disorder Classification: An Efficient Novel Deep Kronecker Neural Network with Sand Cat Swarm Optimisation Algorithm for Feature Selection  
  by Meghavathu S.S. Nayak, Hussain Syed 
  Abstract: Diagnosing a disease takes time and requires highly technical methods. These days, predicting and diagnosing cardiovascular disease (CVD) is crucial to lowering the death rate and catching them in early stages. Prior research employed machine learning (ML) techniques for disease prediction; however, adequate attention should have been paid to feature identification through appropriate methods for selecting features. This research introduced a novel deep learning (DL)-based deep Kronecker neural network (DKNN) for CVD classification. Essential features are extracted using the DenseNet-201 approach, and feature selection techniques help highlight the most important traits while reducing diagnosis costs. Therefore, the Sand Cat Swarm Optimization (SCSO) method is used to identify the most relevant features for diagnosing heart disease. Furthermore, the imbalanced data problem is resolved, and overfitting is decreased through cycle generative adversarial network (CGAN) based data augmentation. Differentiating from other approaches, the proposed approach obtains above 99% accuracy, precision, recall, and F1-score.
  Keywords: cardiovascular disease; CVD; deep Kronecker neural network; DKNN; DenseNet-201; sand cat swarm optimisation; SCSO; cycle generative adversarial network; CGAN.