Exploring Skin Lesion Image Recognition using Deep Learning: A Comprehensive Study with a Focus on Mpox
Students: Md Tazuddin Ahmed, Tasnim Jahan, Joydip Paul
Supervisor: Dr. Taufiq Hasan, Professor
Abstract:
The outbreak of Mpox (formerly called Monkeypox) in the past year has constantly been a leading public health concern, affecting more than 110 member states world- wide. One of the challenges in effectively diagnosing Mpox is its resemblance to other types of rashes, making it challenging to differentiate accurately and promptly. How- ever, in situations where immediate clinical diagnosis is unavailable, computer-aided predictions have demonstrated their value in facilitating rapid detection. Deep learning methods have been developed to learn intricate representations from data, but their performance heavily relies on the availability of sufficient and diverse datasets for analysis. To address this, we first introduced the Mpox Skin Lesion Dataset (MSLD), a publicly accessible dataset intended to be used for binary classification, comprising images of two classes: Mpox and Others (this ’others’ class included lesion images of chickenpox and measles). Later in the year 2022, we came up with the second version of the dataset, namely, Mpox Skin Lesion Dataset v2.0 (MSLD v2.0). Version 2 involved images of patients with Mpox, as well as images from five non-Mpox classes: chickenpox, measles, hand-foot-mouth disease, cowpox, and healthy skin samples. Several extensive experiments are conducted to evaluate the performance of different modified deep learning models in classifying Mpox and other infectious skin diseases. State-of-the-art pre-trained deep learning models, including VGG16, ResNet50, DenseNet121, MobileNetV2, EfficientNetB3, InceptionV3, and Xception, are employed for the classification tasks. The best result was produced by ResNet50 on experiments on MSLD v1.0, achieving 82.96 ± 4.57% accuracy. While performing experiments on MSLD v2.0, transfer learning is implemented with the help of the HAM10000 dataset, which consists of a large collection of pigmented skin lesion im- i ages. Additionally, two separate image augmentation techniques were applied: Color Space Augmentation and Style Transfer. The best overall accuracy obtained in the classification task on MSLD v2.0 is 83.59 ± 2.11%, achieved in the study involving color space augmentation. These findings demonstrate the potential of deep learning models and the useful- ness of the MSLD dataset in accurately classifying Mpox and distinguishing it from other infectious skin diseases. In addition, the study highlights the effectiveness of leveraging advanced machine learning techniques and openly accessible datasets to aid in the early and accurate detection of Mpox, thus contributing to the efficient management of this public health concern.
Development of HPMC and Pectin-Based Oral Disintegrating Thin Film for Folic Acid Delivery
Student: Saika Afrin Sumona
Supervisor: Dr. Muhammad Tarik Arafat, Professor
Abstract:
The aim of this study was to develop an oral disintegrating thin film (ODF) of HPMC and pectin. Folic Acid was incorporated in the ODF for delivery purpose via oral route . Pectin is a plant based biocompatible natural polymer which can be a good film forming agent when incorporated with HPMC. ODFs with different concentration of HPMC and pectin were evaluated via different characterizations .This study showed that increasing the amount of HPMC in the formulation of polymer base , mechanical properties and flexibility got increased . The hydrophilicity of higher HPMC concentration in ODF decreased the disintegration time which also favors 99% release of folic acid within 3 min . Immediate release kinetics was observed for all ODF formulations . ODF with 50:50 ratio of HPMC and pectin respectively showed better mechanical , chemical properties. It was observed that ODF with polymer blend of HPMC and pectin can provide good polymer matrix for folic acid delivery targeting at both patient who have problem in swallowing and vegan people .
Computational Optimization of Patient-Specific Cryoablation Procedure for Breast Cancer Treatment
Students: Tanzila Akter, Md. Rakib Hossen
Abstract:
Cryoablation is a minimally invasive therapy option for early-stage breast cancers and is the most frequent malignancy in women worldwide. On the other hand, the success of cryoablation strongly depends on the positioning of cryoprobes within the tumor and adjacent tissues. The success of this surgery significantly depends on the knowledge and expertise of the doctor, which can lead to variable treatment results. Optimization approaches will probably offer a potential answer to this problem by supplying a cryoablation plan that is patient-specific. These plans are created utilizing numerical simulations, magnetic resonance imaging (MRI), computed tomography (CT), and advanced imaging methods to forecast the results of the cryoablation treatment. An overview of creating patient-specific cryoablation programs for breast cancer treatment is given in this abstract. The optimization process takes into account the thermal characteristics and perfusion rates of the surrounding tissues as well as the size, geometry, and location of the tumor. To ensure a successful and secure process, they also add a variety of limits, such as the size of cryoprobes and their placement inside the tumor. This study’s application to the creation of patient-specific cryoablation strategies will result in promising improvements to treatment efficiency and a reduction in associated dangers. These treatments can help doctors adjust the process to the unique requirements of each patient and offer a more precise forecast of the course of treatment.
Incorporation and Controlled Delivery of Doxorubicin by Synthesized & Necessarily Modified Hydroxyapatite
Student: Nuzhat Arman
Supervisor: Dr. Muhammad Tarik Arafat, Professor
Abstract:
Hydroxyapatite is a ceramic material with Ca/P ratio of 1.67. Usually HA is well known for its bone regeneration capability. Apart from this HA has applications in drug delivery too. For my experimental work hydroxyapatite was synthesized at first by two different methods. They are: Chemical precipitation method & Thermal sintering method. Samples synthesized by these methods had different crystallinity. Then both the samples were loaded with doxorubicin which is an anti-cancer drug. Doxorubicin is said to have side effects if released in uncontrolled way. That’s why incorporation of doxorubicin is necessary. However, the loaded sample were released in three different PH medium. Then there release behaviour was observed. They had different release profiles and the profiles obtained from two samples were different. The samples were then modified by silane which is biocompatible and suitable. After modification the samples were again loaded with drug and their release profiles were compared with the normal ones. Clearly difference was observed. Characterizations like FTIR, TEM, SEM, EDX were performed for the normal and modified samples to obtain the distinctions among them.
Pediatric Pneumonia Detection from Chest X-ray Images Using Deep Learning
Students: Awsaf Rahman, Jannatul Ferdous
Supervisor: Dr. Taufiq Hasan, Professor
Abstract:
Pneumonia is a major cause of illness and mortality in children under the age of five. Although chest X-ray imaging (CXR) is a frequent diagnostic technique for pneumonia, it requires professional interpretation and is not widely available in low- resource settings. An early and correct diagnosis is critical for prompt and efficient treatment of pneumonia. Deep learning algorithms have yielded encouraging results when it comes to automating CXR analysis and detecting pneumonia. In this thesis, we present a deep-learning algorithm for detecting pediatric pneumonia in CXR pictures. To extract information from photos and classify them into normal or pneumonia cases, we employ a convolutional neural network (CNN) with residual connections and attention methods. We evaluate our model on a publicly available dataset of CXR images of children with probable pneumonia, obtained from numerous hospitals in different countries . We compare our model to many state-of-the-art approaches and demonstrate that it yields equivalent or superior accuracy, sensitivity, and specificity. In addition, we undertake ablation tests to assess the impact of each component of our model and give visualizations to explain its predictions. Our model illustrates the promise of deep learning for improving pediatric pneumonia diagnosis using CXR pictures while also minimizing the load on healthcare systems.
Development of Kaolin Incorporated Gelatin/Alginate Based Hemostatic Sponge by Freeze-Gelation Technique.
Student: Md. Mehadi Hassan Sagor
Supervisor: Dr. Muhammad Tarik Arafat, Professor
Abstract:
Porous foam structures based on gelatin and sodium alginate have been fabricated by a novel freeze-gelation technique that avoids the drawbacks of freeze-drying. The effect of kaolin clay and different cross-linkers (glutaraldehyde and tannic acid) on the physical, chemical, and biological properties of the foam structures has been investigated. FESEM, FTIR, swelling and degradation tests, blood clotting index and hemolysis tests, antimicrobial testing, in vivo rat tail amputation and mice liver laceration models, and a prothrombin time-activated partial thromboplastin time (PT-aPTT) test were all used to characterize the foam structures. The results showed that the foam structures had porous architectures with different pore sizes and distributions depending on the kaolin concentration. The swelling and degradation properties of the foam structures were influenced by the cross-linkers and kaolin content. The FTIR analysis confirmed the intermolecular bonding among the polymer chains and the effective cross-linking by glutaraldehyde and tannic acid. The blood clotting index test revealed that kaolin enhanced the hemostatic properties of the foam structures by activating the intrinsic pathway of coagulation and reducing hemolysis. The in vivo tests verified that the foam structures could effectively stop bleeding in emergency medical conditions. The PT-aPTT test further confirmed that kaolin affected the intrinsic pathway of blood clotting and highlighted the potential of wound care use. The antimicrobial test showed that tannic acid crosslinked foam structures had antibacterial resistance against Escherichia coli and Staphylococcus aureus due to the cell lysis capacity of tannins. Therefore, this study demonstrated that kaolin incorporated gelatin/alginate-based foam structures prepared by freeze-gelation technique are promising hemostatic materials with good swelling, stability, antimicrobial, and blood clotting properties.
Development of Chitosan-Based Foam by Freeze Gelation Method to Induce Hemostasis
Student: Ayesha Binth Humayun
Supervisor: Dr. Muhammad Tarik Arafat, Professor
Abstract:
Uncontrolled bleeding is a serious problem in day-to-day accidents, battlefield injuries, surgical settings which increases the likelihood of death. Hence, it is necessary to prevent bleeding with an external agent which incorporates immediate hemostasis. There are various forms of hemostatic agents, among them foam/sponge is suitable. The material of the sponge should be a biopolymer which is biocompatible such as chitosan. However, preparing chitosan sponge by freeze drying method has several drawbacks such as costly, leaving harmful residue to environment, excessive time consumption, surface skin etc. Preparing chitosan foam with proper process variables solves the previous problems and scaling up of samples is possible here. However, chitosan-based foam alone cannot provide desired mechanical properties. So, an additional polymer is added with chitosan to reinforce several characteristics to the sponge. In this study, a chitosan-based foam was developed by freeze gelation method which is cost and time efficient. Then several morphological, physical, biological characteristics were performed. Afterwards a different type of foam was prepared by adding pectin and all the properties were checked again. Finally, In vivo analysis was performed such as mice liver and rat tail. It was found that chitosan/pectin foam performed better in many cases than only chitosan-based foam. The synergistic effect of chitosan and pectin increases the swelling ratio, antibacterial activity and decreases the time for blood clotting.
Geometric Determinants of Local Aerodynamics in Tracheal Stenosis
Students: Md Abdullah Al Mamun, Umma Habiba
Supervisor: Dr. Jahid Ferdous, Associate Professor
Abstract:
There are many pathological conditions that may lead to obstructive airway flow in the lung. Tracheal stenosis, a reduction in airway cross-sectional area, may develop after tracheostomies are usually recognized late. People having tracheal stenosis are mostly unresponsive but when they are affected, the signs and symptoms might be misinterpreted for those with additional respiratory illnesses. The tracheal geometry, as well as the stenosis geometry, are significantly varied among patients which results in the variation of aerodynamic parameters. For analyzing the parameters and effect of stenosis on the trachea wall, simulation is done in two different ways. One simplified geometry model is generated, and stenosis is introduced at different positions, in different lengths, and in different percentages. These models are simulated for resting condition flow rate and for different varying flow rates. Finally, the results are analyzed. A critical percentage of stenosis for different conditions is collected from this result, such as if the stenosis is located 55mm away from the inlet then it would have a significant impact on the bifurcation. For athletes how aerodynamic parameters varies from normal healthy patient is also explored here. For patient-specific models’ simulation, models are extracted from different CT data. With the extracted model similar simulation is done for transient conditions. All aerodynamic parameters are explored, and the time average plot of the parameters is also plotted. For twenty-six patients’ data, a dataset is formed that was processed using Python. Correlation between different parameters is found in the correlation graph. How aerodynamic parameters are associated with each other is appraised from this correlation graph.
A Systematic Study of the Effect of Plaque and Vessel Morphological Characteristics on Plaque Vulnerability
Student: Murar E Rabbe Kabbo
Supervisor: Dr. Muhammad Tarik Arafat, Professor
Abstract:
Plaque rupture in coronary artery is the one of the major causes of death in Bangladesh. The mortality rate would be very low if plaque that is prone to rupture is identified before adverse coronary events. Recently identifying vulnerable plaque has been an interesting research topic among the researchers. However, the effect of plaque morphological characteristics and their interrelation on plaque rupture has not been clear. As plaque vulnerability is associated with the hemodynamic parameters (e.g., wall shear stress), the effects of plaque morphological characteristics on these hemodynamic parameters should give a clear idea about their effects on plaque rupture. The interaction between each characteristic will also be considered to visualize their combined effect on plaque rupture. Computational Fluid Dynamics (CFD) simulations have been performed on coronary arteries for many years. However, there is a lack of systematic study in this field. Design of experiments (DOE) has been a useful tool for optimization of parameters. As there are many morphological characteristics in a coronary artery the realistic approach should be to incorporate the DOE in this field. Our novelty is that I implemented DOE to find out the most dominating morphological feature which may be responsible plaque rupture.
Automated Lung Nodule Detection from CT Scan Images
Students: Mehedi Hassan, Md. Ahnaf Tanvir
Supervisor: Dr. Taufiq Hasan, Professor
Abstract:
Lung cancer is the most lethal cancer and the leading cause of death worldwide. In 2020, it caused 2.2 million new cases and 1.8 million deaths. It accounts for 11 percent of cancer-related fatalities in Bangladesh. Late cancer detection is the primary reason for this. Especially in a country of the third world like Bangladesh, the situation is made more complicated by the scarcity of specialist radiologists and the increased workload of those who already work in the field. Deep learning-based applications play a significant role in this endeavor. Early detection can substantially reduce mortality rates. Numerous publicly accessible datasets have inspired researchers to develop deep-learning algorithms for early lung cancer diagnosis. By detecting lung nodules, an early sign of lung cancer, CT scans can identify the morphological changes that occur during lung cancer. To aid in this process, we developed a lung nodule detection system based on the publicly available LUNA16 dataset. This system con- sists of three sequential systems: lung and nodule segmentation, followed by a nodule classification model. Our lung segmentation model obtained a dice score of 98.41 with results comparable to those of cutting-edge models. The sensitivity of the nodule seg- mentation model was 85.57, while the F1 score of the classification model was 90.19. This demonstrates the capability of our solution to effectively segment lung, detect nodule location through segmentation, and then reduce false positives through the classification model. Thus, our system can aid in the early detection of cancer and reduce its mortality rate. This cascaded system can be extremely useful in limited medical resources like our country. Additional work on this system will allow our model to reach its utmost potential and assist physicians.
Development of Mucin Nanoparticle for the Treatment of Eye Inflammation
Student: Wahida Binte Naz Aurthy
Supervisor: Dr. Muhammad Tarik Arafat, Professor
Abstract:
Mucins, which are multifunctional glycosylated proteins, have recently been used as a novel biomaterial. Nanoparticles (NPs) have distinct advantages in drug delivery. There has never been a research of the synthesis mechanisms of mucin nanoparticles using ionic gelation procedures and just using mucin NP for drug administration. Following that, a novel method for forming mucin nanoparticles is provided in this thesis study. Because mucin is an anionic polymer, the cationic crosslinker Ca2+ is used for crosslinking and nanoparticle production. The creation Of nanoparticles is confirmed by morphological characterizations. Furthermore, glycerol has been added to improve the NP’s antibacterial and anti-inflammatory properties. The performed characterizations demonstrate a sustain drug release profile with excellent drug encapsulation efficiency. The anti-inflammatory property of the NP sample is verified effectively by applying it to an ocular inflammation-induced rabbit eye.
Dialysis Fluid and Dialysate Quality Assessment of Bangladesh
Students: Md. Jahidul Islam, Md. Mahadi Hossain
Supervisor: Dr. Jahid Ferdous, Associate Professor
Abstract:
Hemodialysis is a life-saving treatment for individuals with end-stage renal disease, but ensuring the quality of water used in hemodialysis process is crucial to prevent potential health risks. The thesis overviews water quality analysis in hemodialysis settings, emphasizing the importance of monitoring and maintaining appropriate water standards to safeguard patient well-being. Water quality in hemodialysis facilities is subject to strict regulations and guidelines established by organizations such as the association for the advancement of medical instrumentation (AAMI) and International Organization for Standardization (ISO). These guidelines specify maximum allowable limits for various contaminants such as bacteria, endotoxin and chemical impurities. Regular monitoring of water quality is essential to identify any potential issues or deviations from acceptable standards. Water treatment systems, including reverse osmosis (RO), ultraviolet (UV) disinfection, and carbon filtration, are employed to remove impurities and maintain appropriate water quality levels. Routine maintenance and periodic validation of these systems are essential to ensure their effectiveness and reliable performance. The consequences of inadequate water quality in hemodialysis can be severe, leading to adverse patient outcomes. Contaminated water can introduce bacteria and endotoxins, causing infections and inflammation in patients. Chemical impurities may contribute to systemic toxicity or impair dialysis membrane function. Therefore, strict adherence to water quality guidelines is paramount to minimize the risk of these complications. In our thesis we presented different samples showing various values of microbial, endotoxin and chemical tests which is result due to the hospitals not following standard procedures. In conclusion, water quality analysis is a critical aspect of hemodialysis care. Compliance with established standards, regular monitoring, and maintenance of water treatment systems are fundamental to ensure the safety and efficacy of hemodialysis therapy. By prioritizing water quality, healthcare providers can minimize the potential for adverse events and optimize patient outcomes in the hemodialysis setting.
Design of Tunable Tannic Acid-Gelatin-Dopamine Based Antibacterial Tough Adhesive
Student: Md. Tashdid Hossain Shoudho
Supervisor: Dr. Muhammad Tarik Arafat, Professor
Abstract:
In this study, I have synthesized a new underwater, tough and wet adhesive in acidic medium which contains gelatin, dopamine and tannic acid. Tannic acid is a perfect antibacterial cross linker. The crosslinking is done by centrifugation at room temperature. Cross linking as well as mechanical properties like toughness, strength are tuned by pH of the medium. Two different acids, hydrochloric acid and acetic acid are used for tuning pH of the medium which led to two different samples. The novelty lies in the easy and facile steps in the fabrication process of the adhesive in acidic medium. In order to characterize the adhesive, I have done FT-IR, lap shear stress, cyclic adhesion, underwater adhesiveness testing, antibacterial testing for both S.aureus and E.coli bacteria, hemolysis and self-healing. This research leads to the formation of a tunable, antibacterial and tough adhesive which is a prerequisite for rapid wound healing, specifically, for body parts like elbow, ankle or neck.
Multivalvular Heart Sound Dataset Preparation and Disease Detection
Students: Sumaiya Ohab, Afia Zahin, Rakib Hossen
Supervisor: Dr. Taufiq Hasan, Professor
Abstract:
Phonocardiogram signal can offer crucial prognostic data on the condition of the heart. Thus, automatic heart sound analysis for disease prediction has great po- tential, especially in underdeveloped areas of the world. Heart is not able to pump blood efficiently throughout the body when heart valves are diseased or not in normal condition. The heart has to work more to pump enough amount of blood. Sudden cardiac arrest, failure of the heart and death can be caused by this. These compli- cations can be prevented by early detection of heart valve diseases. In this study, we proposed a method which is based on deep learning and can classify the diseased phonocardiogram data. In this study, we present a classification of the heart sound recordings collected in real life using three classifiers. Heart sound classification means the process of cate- gorizing and analyzing the various sounds produced by the heart during its normal functioning or in the presence of certain abnormalities. Classification of heart sounds involves the application of signal processing techniques and different machine learning algorithms. Electronic stethoscopes or phonocardiogram equipment is used to record the heart sounds at first, then the signals are preprocessed to remove noise and arti- facts. The necessary data is then extracted from the signals using feature extraction techniques. Based on the information retrieved, a variety of machine learning tech- niques like Random Forest (RF), Convolutional Neural Network (CNN) and Deep Neural Network (DNN) are used to categorize heart sounds. Additionally, to increase the precision and robustness of the classification process, ensemble approaches and feature selection techniques may be used.
Development of Antimicrobial Peptide (AMP) Loaded Gelatin-Dopamine Based Hydrogel for Wound Dressing
Student: Saima Islam
Supervisor: Dr. Muhammad Tarik Arafat, Professor
Abstract:
This research aims to formulate a Gelatin-Dopamine hydrogel network that can transport antimicrobial peptides. The swelling profile and the estimated time of degradation onset were used to optimize the samples. Both the base sample and the sample that received AMP had their effect on red blood cell (RBC) count and blood clotting measured. The effectiveness of AMP was studied using an in vitro antibacterial test and an in vivo mice wound model. Against S. aureus and E. coli, AMP demonstrated effective antibacterial properties. By SEM analysis porous morphology was confirmed. A high rate of solvent retention and water vapor transfer guarantees an improved moist, hydrated, and O2 permeable medium.
Patient-Specific Computational Fluid Dynamics Analysis of Urine Flow Through Ureter
Students: Basudeb Biplob Das, Shakib Mahmud Ayon, Humyra Hossain
Supervisor: Dr. Jahid Ferdous, Associate Professor
Abstract:
Through peristaltic action, the ureter is essential in moving urine from the kidney to the bladder. However, there are a number of pathogenic factors that might impede or obstruct this movement. The goal of this work was to create ureter models with anatomical accuracy using CT scan information from six patients with normal ureters. Investigation and analysis of significant flow characteristics, such as flow velocity, pressure differential, average and maximum wall shear stresses, were the goal. We laid the groundwork for future studies into unhealthily ureteral flow by building these realistic ureter models. The study’s findings serve as a standard against which to compare and assess the flow parameters of sick ureters in comparison to those of a healthy reference. This comparative analysis will be extremely helpful in determining the efficacy of therapeutic therapies and tracking the development of post-therapeutic ureteral flow. Understanding the flow parameters under study, such as flow velocity, can help one understand how quickly and unevenly urine flows down the ureter. We can better understand the forces guiding urine flow and the resistance encountered throughout the ureteral pathway by understanding the pressure difference between the inlet and output. Additionally, by analyzing the average and maximum wall shear stress, it is possible to identify potential weak points or abnormalities in the ureteral wall and learn more about the shear forces that are applied to it. These discoveries have important clinical ramifications. Healthcare providers can more accurately determine the effects of different ureteral diseases on urine flow dynamics by comparing the flow parameters of diseased ureters to those of a healthy ureter. This comparative analysis will help in the formulation of treatment plans, the making of judgments, and the tracking of the development of post-treatment ureteral function. It might also aid in the creation of individualized treatment plans that are adapted to each patient’s unique needs and ureteral abnormalities. This study effectively created anatomically correct ureter models using CT scan data from six patients with healthy ureters. Investigation of the flow parameters yields vital information about the characteristics of the typical ureteral flow. Future studies analyzing abnormal ureteral flow and post-treatment assessments will build on these findings. In the end, this research advances our knowledge of ureteral diseases and moves urology closer to more efficient and individualized treatment options.
Optic Nerve Head Shear Stress in Relation to Eye’s Physiological Parameters and Material Properties: A Comprehensive Study
Student: Pritom Saha
Supervisor: Dr. Muhammad Tarik Arafat, Professor
Abstract:
This study presents the findings of numerical simulations investigating the biomechani- cal behavior of the optic nerve head (ONH) under various conditions of intraocular pressure (IOP), intracranial pressure (ICP), and blood pressure (BP), The research also explores the relationship between different geometric and material properties Of the eye structure and their impact on normal IOP and ICP values. The simulations utilize a generic eye model that can be modified to examine the effects of dilTerent geometric and material properties. The results of the simulations indicate that as IOP increases, the maximum shear stress experienced by the lamina cribrosa and post-laminar neural tissue increases. On the other hand, when ICP increases, the maximum shear stress on the post-laminar neural tissue increases. while it decreases for the lamina cribrosa. When blood pressure rises. the maxi- mum shear stress on the lamina cribrosa and post-laminar neural tissue decreases at a slow rate. Regarding the significance of material properties in causing optic nerve damage, the sequence is as follows: the Young’s modulus of the lamina cribrosa has the greatest impact, followed by the Young’s modulus of the pia mater, the Young’s modulus of the post-laminar neural tissue, the Young’s modulus of the vitreous body, the Young’s modulus of the dura mater, and finally. the Young’s modulus of the sclera.
Study of Physics-Informed Neural Networks to Predict Wall Shear Stress in Left Coronary Artery
Student: Angkon Biswas
Supervisor: Dr. Muhammad Tarik Arafat, Professor
Abstract:
Computational fluid dynamics (CFD) based on finite element method (FEM) modeling of blood flow though coronary artery serves as a powerful tool in understanding various medical conditions such as aneurysm, degree of plaque formation and defining the possibilities of plaque formation based on flow parameters, etc.; and developing effective diagnostic methods and treatments. However due to the complexity and high computational cost of the simulations CFD suffers from a relatively long computational time which hinders their transformation from a research tool to a clinical tool. With realistic boundary conditions and complex computational domains, this problem becomes even more severe for image-based, patient-specific CFD simulations. IN This study I proposed a Deep Neural Network Framework as an alternative to CFD for generating hemodynamic parameters from coronary arterial blood flow in real-time. Here, I employed Physics Informed Neural Network (PINN), which provides a flexible approach to integrate mathematical equations governing blood flow which is Navier-Stokes equation, allowed us to overcome the large data requirement constraint in deep learning approach. I used patient specific CTA data of left coronary artery (LCA) and obtained our training data from CFD analysis using a passive scaler. This study shows that PINN can be used to improve WSS quantification in problems where the boundary conditions are unknown by using very few measurement points. Here I used the underlying conservation principles (i.e., for mass, momentum, and energy) to deduce hidden variables of interest like velocity and pressure fields just from spatio-temporal visualizations of a passive scaler. I designed a data assimilation algorithm that does not depend on geometry or initial and boundary conditions. Due to this, it is highly flexible in terms of choosing the specific arterial segment and time domain for data acquisition, training, and prediction. The proposed algorithm achieves accurate predictions of both the pressure and velocity fields. The average accuracy of the wall shear stress prediction v is 91.3%. The proposed framework makes a significant contribution by allowing the prediction of blood flow hemodynamics in any section of the arterial network based solely on geometrical features and velocity measurement data, which can be obtained through non-invasive methods. The findings from evaluating the model on six individualized geometries suggest that the model can serve as a viable alternative to finite element methods and other numerical simulations that may be challenging and time-consuming to implement. In addition, even though the model was initially trained for coronary arteries in a cost-effective and time-efficient manner, it can be readily adapted for other artery types by using a new dataset.
A Computational Study of the Lumbar Spine Model for Various
Students: Manzar Monir, Md. Sohel Rana
Supervisor: Dr. Jahid Ferdous, Associate Professor
Abstract:
The lumbar spine, comprising the lower back region, plays a crucial role in supporting the body’s weight, maintaining posture, and facilitating movement. First, a detailed lumbar spine model is studied, incorporating intricate anatomy and structural components. This model is validated against experimental data and serves as the foundation for subsequent analyses. Next, a range of physiological scenarios like different values of elastic modulus, Poisson ratio, permeability, etc. is evaluated with the computational model of the lumbar spine. Through this comprehensive computational analysis, we aim to enhance our understanding of lumbar spine biomechanics, injury mechanisms, and their clinical implications. The pathological scenarios are also presented with different values.