17th International Conference on Biomedical Engineering

In this paper, space and time-fractional Pennes bioheat equation (FPBE) has been obtained by substituting integer-order derivatives with Caputo fractional derivatives. We have introduced a novel meshless approach for the solution of FPBE in which space and time discretizations are independent. Hence, we have employed two different basis functions in time and space. Numerical results have been obtained using a Kronecker product and compared with analytic one to verify the applicability of the technique.

The mitral valve (MV) is one of the four heart valves located between the left atrium and left ventricle, and it regulates the flow between them. The pressures applied from the left atrium and ventricle on the leaflets can lead to a set of complex structural stresses and strains which are not plausible to be determined experimentally. Numerical models, such as finite element method (FEM), is used to study valve function and help the design and development of repair procedures and replacement devices. These models have evolved from simple two-dimensional (2D) approximations to complex three-dimensional (3D) models. Here, we used a simplified 3D patient- specific finite element (FE) model of the mitral heart valve and papillary muscles to calculate the stresses and strains in the leaflets and muscles during the closuring phase. The highest stresses and strains were located in the anterior leaflets with 1.79 MPa and 49.51%, respectively. The stresses and strain in the muscles where attached to the anterior leaflets were the highest with 7.09 MPa and 14.21%, respectively. The results have implications for the medical and biomechanical experts to understand the stresses and deformations in the MV leaflets and papillary muscles during the closuring of the leaflets.

Automated segmentation of the choroid in optical coherence tomography (OCT) images is important to assess diseases which accompany choroidal changes. Existing methods try to segment the boundaries of the choroid by utilizing the edge information. However, in many cases, these boundaries may not have distinct edges thereby rendering edge-based segmentation inaccurate. This paper addresses this issue by performing a region-based segmentation of the choroid instead of an edge-based segmentation. The proposed method uses a deep-learning architecture called U-Net which utilizes the texture of the choroid to segment it. The proposed method was evaluated on a dataset of 1280 OCT images and achieves an intersection over union (IoU) of 0.85. This was better than the IoU of 0.51 achieved by a related work which uses edge-based segmentation. In addition, we have experimentally assessed the effect of removing the retinal layers before choroid segmentation which was performed in the proposed work. Results show that if the retinal layers were not removed, the IoU drops to 0.81. The proposed method can help in automatic analysis of choroidal changes even when the choroidal boundaries are not clear which is often the case in diseased eyes.

Lower extremity dressing aid has been a useful piece of assistive device for the elderly people and people who suffer from monoplegia. However, some of the commercially available devices take too much time and even require the patient to ask help from others to operate the device which defeats the purpose of providing a sense of independence. Other devices may also demand the patient to bend and exert more effort than necessary. To address these problems, we designed an innovative structure for the assistive device to include a moving garment hanger. In this design concept, the hanger will be controlled using electromechanical parts by means of push buttons to trigger the rotation of a single motor. It has been successfully simulated using SolidWorks for the mechanical parts and Proteus 8 Professional for the electrical mechanism. The electrical part includes the motor drives which takes input from the push buttons and gives output to the motor for rotation. The motor is connected to the pulley which moves the belt vertically along with the garment hanger. This automates the whole movement of the lower garment in either upward or downward direction for dressing and undressing. Furthermore, the garment hanger is made to be adjustable to accommodate different waist sizes. It is also designed to be portable for convenience and efficiency. The simulation showed a feasible automated lower extremity dressing aid for monoplegic and elderly people that will address their dressing problems brought by frailty or disability.

Vision impairment remains a serious global health concern in poor and developing countries. Indonesia has the highest blindness rate in the Southeast Asian region. The majority of the roughly 3.0 million visually impaired individuals have Islamic faith. A Salah means “worship”, is an obligatory religious duty for every Muslim which is observed 5 times a day and performed in a whole series of movements. It requires them to face the direction of the Qibla that is oriented towards the Kaaba in the city of Mecca that connects the believers. In this research, a portable, lightweight, and low-cost assistive device was placed in a traditional Muslim headwear Peci, that will help them not only to avoid head collisions but also to determine the correct Qibla prayer direction. The module integrated obstacle detection, direction finder, and notification system. First, the ultrasonic sensor will be embedded in the front of the Peci for obstacle detection. Second, a digital compass and global positioning system receiver will be utilized to determine the coordinates of the person and to calculate the direction of the Qibla. Third, selection control between the obstacle detector and the Qibla finder functions will be done through the use of switches. Fourth, notification is done via a vibration motor. Fifth, a Blynk mobile application will be developed to provide family members with the real-time location of the visually impaired loved one.

Chronic wounds contribute a substantial social and economic burden on the healthcare system. The global cost of wound treatment was about $19.8 Billion USD in 2019. Healing of chronic wounds takes typically more than 3 months. Current treatments are ineffective and do not always promote wound closure, which requires the activation of multiple cell types. Extracellular vesicles (EVs) contain multiple biomolecules that influence surrounding cells and thus have large capacity to promote tissue repair. To harness the chemoattractant properties of EVs, we developed an extracellular vesicle-laden hydrogel (HydroGEV) with optimized stiffness to promote functional tissue repair, since both mechanical and biological factors influence cell growth and subsequent tissue repair. EVs were isolated and purified from placental stem cells, characterized and incorporated into a gelatin-based hydrogel (GHPA) with different relative stiff-nesses (low, medium and high) determined by crosslinking density. The EVs were found to increase the migration capability of cells in a migration assay, confirming their strong chemoattractant properties and supporting their application for cell recruitment in wound healing. When incorporated into GHPA hydrogels, the EVs effectively improved cell attachment regardless of the stiffness of the hydrogels. Importantly, we demonstrated that by optimizing hydrogel stiffness it was possible to achieve higher cell proliferation and more phenotypic morphology. These promising results support the potential of HydroGEV as a better therapeutic option for patients with acute or chronic wounds.

State-of-the art machine learning (ML) methods show immense promise for medical applications, offering significant improvement in prediction capabilities for electronic health record (EHR) data. However, the models’ black-box nature make it difficult for an end-user to trace the decision-making process, a key challenge for the application of machine learning in healthcare. Explainable machine learning (ex-ML) methods is a complementary tool, identifying key features associated with a model’s predictions. Extending from ex-ML methods, we show how EHR data can be used to make explainable machine learning predictions, and generate insights to minimise an unfavourable healthcare outcome. Firstly, our clinical explainability step outlines how feature importance values change across the range of possible physiological variables, for a single patient and across a population. The output is a target feature value for which the feature importance values are minimised. Secondly, we show that explanations across different machine learning models are not guaranteed, and we provide a method for aggregation. Finally, we develop a set of actionable recommendations on features that can be changed, providing a target treatment order that doctors can follow, to reduce a patient’s likelihood of disease.

Low back pain is a disabling condition that imposes an enormous socioeconomic burden. Intervertebral disc (IVD) regeneration strategies are ideal goals aiming to provide a better solution. While regeneration occurs with tensile loading of degenerated IVDs, no long-term longitudinal data exists to show the sustained impact of tensile loading under physiological conditions. The aim of this study is to develop an MRI-compatible rabbit IVD distraction model to enable longitudinal monitoring of nutrient supply and regeneration process of degenerated discs. Rabbits were divided into two groups. (1) IVD stabbing, no treatment (2) IVD stabbing, distractor implant on 6 weeks post-stabbing, treated by 120 N tensile force. Under C-arm guidance, L4-5 IVD was stabbed by bone marrow aspirate needle to induce disc degeneration. At 6 weeks post stabbing, baseline IVD degeneration was verified by MRI. Under C-arm guidance, path was created by stainless steel k-wire on L4-5 vertebra and replaced by zirconia k-wire. Distractor was connected to k-wire and 120 N force was applied on rabbit spine. Rabbit disc health was evaluated by MRI at 7, 11- and 15-weeks post distractor implant. Rabbit was euthanized after the last MRI scan. From T2 STIR MRI, disc distraction can arrest the progression of disc degeneration in treated group but further deteriorate in control. Post-contrast MRI showed the improvement of nutrient flow in distracted disc. The development of MRI compatible distractor will allow a longitudinal study of the degenerated disc regeneration process under tensile distraction.

This study aims to predict the range of motion (ROM) with secondary parameters such as the intra-disc pressure (IDP) and facet force under complex physiological loading for Anterior Cervical Discectomy and Fusion (ACDF) and degeneration occurring at various functional spine units (FSU) in a human cervical spine, using machine learning models. Multi-target regression is a machine learning (ML) algorithm that outputs an array of values for a given set of input parameters. An anatomically accurate and validated finite element model (FEM) of a human sub-axial spinal column (C2-T1) was used in this study. Material properties for all spine components were taken from literature. An algorithm programmed using Python was interfaced with ABAQUS to automate the calculation of ROM, IDP and facet force generated from the nodal and elemental data of an intact model, a model with ACDF at the C5-C6 level and a model with mild degeneration at C5-C6 separately. The data generated from the FEA results were trained with random forest regression, support vector machines and multiple linear regression algorithms. The results indicated that the R² value was significantly high for the random forest regression model, accounting for a very less Root Mean Square Error (RMSE) and was able to predict more than one target variable unlike the rest. Conclusively, the target variables were predicted under complex loading conditions for clinical conditions of fusion and degeneration with high accuracy and less computational cost.

The natural weight of the head, any head supported mass and various muscle forces result in eccentric compressive loads that should be effectively supported by the human cervical spine to protect the spinal cord and maintain interrelationships among vertebrae. Different experimental studies have used various testing protocols to study the effect of a compressive preload on the biomechanics of the cervical spine. The objective of this study is to investigate the effect of such an in-vivo compressive load on the range of motion (ROM) of the sub-axial column using a novel computational method. An anatomically accurate and validated finite element model (FEM) of a human sub-axial spinal column (C2-T1) was used in this study. Material properties for all spine components were taken from literature. Cortical bone, cancellous core and intervertebral disks were modelled using linear isotropic elements. Ligaments were modelled using shell elements with non-linear material properties. An algorithm programmed using Python was interfaced with ABAQUS to calculate ROMs with nodal data extracted from its workspace. The code used nodal co-ordinates at the endplates to define physiological planes, and this process helped to calculate the angle between the deformed and undeformed model. The ROMs was successfully computed using these definitions, and EPS was measured across all the segmental units.

Magnetic Resonance Imaging is a commonly used modality to diagnose dementia and there is a massive requisite for an automated MRI image classification algorithm to assist the clinician during diagnosis. The main objective of this research work is to categorize the brain MRI images of patients as demented and non-demented using harmony search based hybrid meta-heuristic optimization algorithms. For this analysis, 65 non-demented and 52 demented subjects collected from Open Access Series of Imaging Studies are used. With appropriate modifications on original algorithms, the classification performance of four meta-heuristic techniques namely Particle Swarm Optimization, Artificial Bee Colony, Ant Colony Optimization, Harmony Search are tested individually as transformation technique based classifier. Then Harmony Search will be hybridized with above mentioned other three meta-heuristic techniques and the classification performance improvement is analyzed. Harmony search based hybrid optimization techniques are widely reported for solving numerical optimization problems, feature extraction, clustering and training neural networks. To the best of our knowledge, there are no reports in the literature regarding the usage of harmony search based hybrid optimization techniques as transformation technique to classify medical images. Particle Swarm Optimization hybridized with Harmony Search algorithm provides the highest accuracy of 84% in dementia MRI image classification.

The objective of this research work is to use Crow Search Algorithm for classification of blood smear microscopic images into two classes: Leukemic B-lymphoblast cells (cancer cells) and normal B-lymphoid precursors (normal cells). Crow Search Algorithm is generally used to crack numerical optimization, training neural networks and feature selection problems. This research work uses Crow Search algorithm as a transformation technique to convert non-linearly separable data points as linearly separable data points. Microscopic image dataset named C-NMC is collected from cancer imaging archives website and the microscopic images of 30 cancer subjects and 30 normal subjects are considered in this analysis. To prove the significant performance of crow search algorithm as transformation technique based classifier, other popular unsupervised classification techniques like K-Means and Fuzzy C Means are used. Remarkably 87% of accuracy is achieved when crow search algorithm is used as classifier.

PDT (photodynamic therapy) is one of the treatments of superficial early cancer. But PDT is not adapted to advanced cancers. The peak absorption wavelength of Talaporfin sodium (Laserphyrin^®) is 405 nm. 405 nm light is not used for Laserphyrin^® PDT because it does not penetrate into tissues deeply interfered with hemoglobin absorption and scattering attenuation in tissues. The next peak absorption wavelength 664 nm is used for Laserphyrin^® PDT now. The penetration depth of 664 nm light in tissue is not enough for advanced cancers.

Photoexcitation is usually performed by single-photon excitation at absorption wavelength. However, high peak power light can occur two-photon excitation. Since the absorption rate at 405 nm of Laserphyrin^® is very high, the effect of two-photon excitation at 810 nm can be expected. Through the absorption rate at 664 nm of Laserphyrin^® is smaller than at 405 nm, light scattering attenuation at 1328 nm in tissues is smaller than at 810 nm. Two-photon excitation PDT at 1328 nm might be expected.

In this study, we investigate a feasibility of two-photon excitation PDT using near infrared light, and compared two-photon excitation PDT at 810 nm with at 1328 nm. We have developed Nd:YAG SHG pumped KTP-OPO that oscillate at both wavelength by replacing the output coupler and adjusting phase matching angle with high peak power. In this paper, we report KTP-OPO configuration and optical properties at 810 nm and 1328 nm. We have accomplished 200 kW peak power 20 ns pulse width 50 Hz repetition rate at both wavelengths.

Skeletal bone age assessment is a common clinical practice to investigate endocrinology, genetic and growth disorders in children. Traditional clinical testing methods are time-consuming and labor-intensive, and there will be operator errors because of the subjective factors of the physicians. The existing automatic bone age detection methods based on automatic extraction of clinical features also has the problems of low accuracy and difficult generalization due to inaccurate feature extraction. In this paper, we propose an end-to-end automatic bone age detection method based on deep learning to process hand bone X-ray images. A Convolutional Block Attention Module (CBAM) is added to the basic model of Inception Resnet v2, the Softmax network layer is changed to the Mean Absolute Error (MAE) index output, and the mean square error loss function is used to evaluate the performance of the bone age detection regression problem. A new stratified k-fold cross validation method is proposed to cover training models on the public dataset of bone age for all races, genders and age ranges. It shows that the MAE between the detected bone age and the labeled bone age is 0.34 years in the results, which is better than the current bone age evaluation method.

Multiple Deep Learning (DL) algorithms have been developed recently and are shown to be achieving very high accuracy in object detection. However, challenges have been reported in detecting small objects within a large image (e.g. > 2000 by 2000 in resolution). Various methods have been proposed using different detection algorithms in order to detect small objects. However, these approaches require high computational resources and are not suitable for edge computing devices that are used for practical applications such as pedestrian traffic light detection. We explored two different methods of detection to evaluate which method is best at detecting small objects. The first method is a two—part procedure with the first step being image processing and the second step, a R-CNN based detection using Edge Boxes algorithm for the extraction of region proposals. The second method is solely Faster R-CNN Object Detection with Instance Segmentation, termed as Mask R-CNN. A total of 4000 streets images of Singapore with pedestrian traffic lights were used as training data. The dimensions of the images range from 1200 by 900 to 4000 by 3000. The small object to be detected is the green or red man within pedestrian traffic lights. We evaluated these methods based on training time required, detection time, accuracy as well as suitability for deployment in edge computing devices. From the results, it is shown that the HSV + R-CNN approach is preferred as it achieves an accuracy of 95.5% and can be deployed in edge devices.