Dissertations/Theses - Department of Biomedical Engineering

Browse

Now showing 1 - 16 of 16

Automated denoising and classification of lung sounds using end-to-end deep learning models
(Department of Biomedical Engineering, BUET, 2025-04-19) Samiul Based Shuvo; Hasan Al Banna, Dr. Taufiq
Lung sound auscultation is essential for monitoring respiratory health, especially in regions facing a shortage of skilled healthcare workers. Automated analysis of respi- ratory sounds has the potential to provide significant clinical support in such settings. However, lung sounds (LS) are frequently contaminated by various noise sources such as heart sounds, background conversations, and movement artifacts which makes accu- rate interpretation challenging. Conventional denoising techniques often fail to address these challenges due to the spectral overlap between noise and respiratory signals in real-world clinical recordings. In addition to that, while respiratory sound classification has been widely studied in adults, its application in pediatric populations, particularly in children aged <=6 years, remains a complex and underexplored area. The developmen- tal changes in pediatric lungs considerably alter the acoustic properties of respiratory sounds, necessitating specialized classification approaches tailored to this age group. To address these challenges and advance automated lung sound analysis, this thesis is divided into two major components. In the first part, a specialized deep-denoiser model (Uformer) has been proposed for lung sound denoising. The proposed Uformer model consists of three modules: a Convolutional Neural Network (CNN) encoder module dedicated to extracting latent features, a Transformer encoder module employed to en- hance the encoding of unique LS features further and effectively capture intricate long- range dependencies, and a CNN decoder module employed to generate the denoised signals. The performance of the proposed Uformer model has been evaluated on lung sounds induced with different types of synthetic and real-world noise. The proposed model showed an average output SNR of 16.51 dB when evaluated with -12 dB LS signals. Our end-to-end model, with an average output SNR of 19.31 dB, outperforms the existing model, achieving nearly double the performance, when evaluated with am- bient noise and fewer parameters. Based on the qualitative and quantitative findings in this study, it can be stated that the proposed denoising model is robust and gener- alized to assist with monitoring respiratory conditions. The second part of the thesis focuses on the classification of pediatric respiratory sounds. A multistage hybrid CNN- Transformer architecture has been proposed for detecting respiratory diseases from both entire recordings and individual breath cycles using scalogram images of the signals. To fill the gap in pediatric classification, the SPRSound dataset, comprising recordings from children with an average age of 5.5 years, has been utilized for two-level classifica- tion tasks. The proposed classification model utilizes CNN-extracted respiratory sound features from scalogram images, integrating an attention framework to improve predic- tive performance. The proposed framework achieved a score of 0.9039 in binary event classification and 0.8448 in multiclass event classification. At the record level, ternary classification yielded a score of 0.720, while multiclass record classification attained 0.571. However, our proposed method consistently demonstrates a 3.81% and 5.94% performance gain over the existing best model, respectively, in these record-level clas- sification tasks. These proposed approaches can significantly aid in the diagnosis and prediction of the severity of respiratory diseases in both developing and underdeveloped nations.
Computational hemodynamic analysis of bypass graft for patients with severe femoral periphheral arterial disease
(Biomedical Engineering, 2023-07-16) Nowreen Afsana; Muhammad Tarik Arafat, Dr.
Peripheral arterial disease (PAD) is a condition in which narrowed arteries reduce blood flowtothearmsorlegs.Thismaycauselegpainwhenwalking(claudication)and othersymptoms.PADisusuallyasignofabuildupoffattydepositsinthearteries (atherosclerosis).Atherosclerosis causes narrowing of the arteries that can reduce blood flowinthelegsand,sometimes,thearms.Femoral-bypasssurgeriesareacommon modeoftreatmentforthosewhoaresufferingfromseverePAD.Neointimalhyperpla- sia (NIH) occurs when the innermost layer of arteries is abnormally thickened as a result ofinjuryorcertainmedicalprocedureslikebypasssurgeries.AsaresultofNIH,the blood vessels can re-narrow again, which is known as restenosis.Hence, in numerical studies, theimprovementofthesehemodynamicfactorshasbeenutilizedasanindi- catortoachievesuperiorgeometriesandhigherpatencyratesforbypassgrafts.This studyaimstoconductacomputationalanalysisofdistinctanatomicalandgeometrical characteristics of grafts to determine optimal graft features for femoral bypass surgery,sothatithelpsvascularsurgeonsplantheirsurgerytoachievemaximumfavorable outputfromtheprocedure.Tomimicthereal-lifescenarioofbloodflowingthrough thefemoralbypassgraftsandtoacquirethehemodynamicparametersandresulting mechanicalparameters,acomputationalfluiddynamics(CFD)systemhasbeenused throughoutthisdissertation.Clinicalimagingmodalitieshavebeenusedinthisstudy toattainthepatient-specificmodelsofarteriesforcomputationalanalysis,whichwere used to create virtual femoral bypass surgery models.Parameters like wall shear stress (WSS),timeaveragewallshearstress(TAWSS),oscillatoryshearindex(OSI),high oscillatory low magnitude shear (HOLMES), pressure drop, flow velocity, and average helicity have been measured to analyze the performance of the graft under different ge- ometrical and anatomical variation.Through the results obtained in this research thesis,it shows that computational fluid analysis can be used in the medical field to aid doctors withhelpfulinformationandhelpthemtoservetheirpatients.
Development and evaluation of gelatin starch based folic acid loaded orally disintegrating film
(Department of Biomedical Engineering (BME), BUET, 2024-07-13) Farin, Moontaha; Arafat, Dr. Muhammad Tarik
This study aims to develop a platform of orally disintegrating film (ODF) with gelatin and modified sweet potato starch. The novelty of this work lies on investigating the influence of sweet potato derived starch and its modification on the functional properties of gelatin based orally disintegrating film. Starch was extracted from native sweet potato and two types of physical modification, pregelatinization and alcohol alkaline treatment (AA) were employed on the sweet potato starch. These modification processes changed the surface morphology as well as physiochemical properties of starch such as crystallinity, amylose content, swelling and viscosity which were measured in this study. Incorporating this modified starch, a gelatin-starch based matrix was developed with tunable properties. Structural characterization and possible interactions of gelatin-modified starch matrix were confirmed by field emission scanning electron microscope (FESEM), attenuated total reflection fourier transform infrared spectroscopy (ATR-FTIR) and simultaneous thermal analysis (STA). After the incorporation of pregelatinized sweet potato starch, ODFs showed enhanced tensile strength, on the other hand incorporation of alcohol-alkaline treated starch showed enhanced elongation value of the films. However, both of the starch modification resulted in higher hydrophilicity of films which was represented in the contact angle (57% less) and disintegration time (37% decrease) values compared to gelatin films. In vitro drug dissolution was assessed with folic acid, a B vitamin, which followed immediate release profile. Later hemolytic assay of the ODFs assured that the developed ODFs were biocompatible. Owing to improved hydrophilicity, mechanical properties and biocompability, this gelatin-starch based ODFs present a wide range of possibilities for being used as a rapid release platform for therapeutics.
Development of antimicrobial peptide loaded PVA-based nanofibrous matrix to combat bacterial wound infection
(Department of Biomedical Engineering, BUET, 2024-08-31) Firdous, Syeda Omara; Arafat, Dr. Muhammad Tarik
Wound infections caused by antibiotic resistant bacterial strains, present a significant challenge in healthcare worldwide due to the uncontrolled use of antibiotics. A promising alternative to conventional antibiotics is antimicrobial peptides (AMPs) that provide potent, broad spectrum antibacterial activity with a reduced likelihood of resistance development due to their action mechanism. This research focuses on developing and evaluating AMP loaded PVA nanofibrous matrix as an innovative approach to treating these infections. AMPs extracted from corn seeds, were stabilized within PVA nanofibers through in situ incorporation facilitated by dopamine, resulting in robust and stable nanofibrous matrices. Physical and chemical characterizations, including FTIR, XRD, and tensile test, indicated that the 2:1-PVA/AMP formulation offered superior structural integrity with a tensile strength of 3.2 ± 0.3 MPa, making it the an effective option for further testing. This formulation showed a significant zone of inhibition (ZOI) of 19 ± 0.3 mm against Staphylococcus aureus, 16 ± 0.3 mm against Escherichia coli, 48 ±0.1 mm against Proteus spp., and 16 ± 0.1 mm against Pseudomonas aeruginosa, outperforming the 1:1-PVA/AMP. Comparatively, ceftazidime (CAZ) showed no antibacterial activity across all tested strains. Hemocompatibility assessments revealed that all PVA/AMP samples exhibited a hemolysis ratio below 3%, confirming their safety for blood contact. In vivo experiments using an infected wound model in mice demonstrated that the 2:1-PVA/AMP nanofibers significantly accelerated wound healing, with markedly reduced pus formation and superior tissue regeneration compared to both the control and PVA/CAZ groups. Histological analysis showed a marked reduction in the epithelial gap and enhanced re-epithelialization, with 87 ± 2% wound closure observed by day 14, compared to 44 ± 1% in the PVA/CAZ group. Additionally, a significant reduction in inflammatory cell infiltration was observed in the 2:1-PVA/AMP group which indicates better management of infection and inflammation. This contributed to more effective tissue regeneration compared to both the control and PVA/CAZ groups. These results underscore the potential of AMP loaded PVA nanofibers as a superior alternative to conventional antibiotic treatments in wound care, particularly for managing infections caused by antibiotic resistant bacteria.
Development of gum-based chitosan nanoparticle with improved functional properties for conjunctivitis treatment
(Department of Biomedical Engineering, BUET, 2025-02-05) Afiya Mubasharah; Tarik Arafat, Dr. Muhammad
Conjunctivitis, commonly known as pink eye, is a widespread eye condition characterized by inflammation or infection of the conjunctiva, the thin membrane covering the white part of the eye and the inner eyelids. Treating conjunctivitis effectively can be challenging due to the limitations of traditional drug delivery methods, such as eye drops, which often do not provide sufficient drug retention and can lead to systemic side effects. This study presents an innovative dual-drug delivery approach that involves encapsulating two powerful medications—dexamethasone (DEX), a corticosteroid that reduces inflammation, and moxifloxacin (MOX), an antibiotic effective against bacterial infections—within chitosan nanoparticles (CSNPs). These nanoparticles are further enhanced with various natural gums used as crosslinker, including Arabic gum, tamarind gum, guar gum, gellan gum, and xanthan gum. The inclusion of these gums aims to improve the mucoadhesive properties of the formulations, which helps them adhere better to the ocular surface and increases the availability of the drugs where they are needed most. The study conducted comprehensive evaluations through various tests—in vitro, ex vivo, and in vivo to assess the effectiveness, biocompatibility, and ability of these formulations to retain drugs on the eye's surface. Among the formulations tested, Arabic gum-based chitosan nanoparticles (CS-A-0.03) showed optimal characteristics, including a particle size of ~200 nm and high stability. These nanoparticles demonstrated excellent drug encapsulation efficiency (97 ± 1.96%. for DEX, 58.72 ± 5% for MOX) and controlled release over 5 hours. The formulation exhibited biocompatibility with only 0.28±0.09% hemolysis and no corneal tissue damage. Notably, the CS-A nanoparticles showed 73.91±0.25% mucoadhesion, significantly higher than commercial eye drops (64±3.5%). In vivo tests revealed faster and more effective inflammation reduction compared to conventional treatments. The long-term stability study showed that CS-A based formulations remained stable for 12 months at room temperature. The results demonstrated significant improvements in therapeutic outcomes when compared to conventional eye drop treatments. Specifically, these new formulations showed enhanced drug retention time on the ocular surface, addressing critical issues associated with traditional treatments.
Development of sodium alginate nanocarriers for the treatment of ophthalmic infection and inflammation
(Department of Biomedical Engineering (BME),BUET, 2023-11-01) Datta, Nondita; Tarik Arafat, Dr. Muhammad
Limited bioavailability of the current eye drops caused by insufficient corneal penetration and quick drug washout often leads to the requirement of frequent doses during the treatment of ophthalmic diseases. In this study, sodium alginate polymannuronate (SA) nanocarriers were prepared using ionotropic gelation method to provide better bioavailability through mucoadhesivity by binding to the ocular mucus layer. It has been assumed till this date that only G blocks of SA takes part in ionic crosslinking. This study has revealed that M blocks can also take part in crosslinking during ionic gelation in the absence of G blocks and show better encapsulation efficiency due to self-assembly. This comprehensive analysis of the only M block containing sodium alginate (SA) nanoparticles demonstrated how to modify their properties by meticulously choosing the appropriate crosslinker type and concentration, and also highlighted their novel drug delivery behavior. Four crosslinkers were initially studied at various concentrations and the Taguchi design of experiment demonstrated that the crosslinker type and concentration had a statistically significant effect on particle size and stability. By comparing the zeta potential and particle size values, the optimal combination with the necessary microstructural characteristics for the research purpose was found. Dexamethasone (DEX) and Ciprofloxacin (CIP) were loaded into the nanoparticles. Polymeric micelles formed through self-assembly of the polymer showed very high encapsulation efficiency upon the incorporation of the two hydrophobic drugs. Sustained and simultaneous release of the drugs was obtained. The formulation was proved to be biocompatible and cured the scleral inflammation of the rabbit uveitis models significantly within only 3 days. However, semi-gel formulationsare frequently sticky, opaque, and too viscous to use comfortably. Thus, a nanosuspension based formulation is required to be prepared for overcoming the issues while maintaining the beneficial qualities of the semi-gel.Thus, nanosuspension has been formulated mimicking human tear fluid properties.It is expected that this in-depth study will pave the way for future works using the nanocarriers, in the field of targeted drug and protein delivery.
Diagnosis-oriented learned image compression system for chest radiographs with Regions-of-interest control
(Department of Biomedical Engineering, BUET, 2025-05-28) Ibn Sabur Khan Nuhash, Shoyad.; Hasan Al Banna, Dr. Taufiq
This thesis presents DiagLIC, a diagnosis-oriented learned image compression (LIC) framework specifically designed for chest radiographs. Unlike conventional codecs and general-purpose LIC models that prioritize visual fidelity or bitrate efficiency alone, the proposed method in- corporates both diagnostic awareness and region-of-interest (ROI) preservation directly into the compression objective. Leveraging a transformer-based architecture with prompt-conditioned Swin Transformer blocks, the system supports variable-rate compression through a single model and allocates higher representational capacity to clinically significant regions such as lesions, opacities, or cardiomegaly zones. The framework is jointly optimized using a composite loss function that integrates rate-distortion and classification performance, ensuring that compressed images retain features essential for downstream diagnosis. Empirical evaluation on NIH and VinDr-CXR subsets demonstrates that DiagLIC significantly outperforms classical codecs (e.g., JPEG, WebP, JPEG2000) and state-of-the-art learned compression baselines in both PSNR and diagnostic accuracy. Across six quality levels, DiagLIC consistently achieves higher PSNR (e.g., 40.42 dB at 0.0951 bpp) and yields an average weighted PSNR of up to 52.49 dB, with ROI-specific PSNRs exceeding NROI PSNRs by over 1 dB at all levels. For diagnostic classifi- cation, DiagLIC maintains high fidelity, with only a minimal drop in average AUC (0.014) com- pared to uncompressed images, achieving AUC scores above 0.70 across all 8 tested patholo- gies and no degradation in certain categories such as Nodule/Mass. Statistical analysis confirms that the improvements in rate-distortion performance over baselines are significant, while the reduction in diagnostic performance remains clinically negligible. Qualitative comparisons fur- ther reveal that DiagLIC effectively suppresses artifacts and preserves lesion-level details, even at aggressive compression. This work demonstrates the feasibility and clinical value of inte- grating semantic objectives into image compression models, laying the groundwork for future diagnosis-preserving compression techniques in medical imaging.
End-to-end deep learning architecture for multi-class brain tumor segmentation and classification from MRI images
(Biomedical Engineering (BME) BUET, 2024-02-12) Salman Fazle Rabby; Taufiq Hasan Al Banna, Dr.
Brain tumors are severe medical conditions that can prove fatal if not detected andtreated early. Radiologists often use MRI and CT scan imaging to diagnose brain tumorsearly.However, a shortage of skilled radiologists to analyze medical images can beproblematic in low-resource healthcare settings. To overcome this issue, deep learning-based automatic analysis of medical images can be an effective tool for assistive di-agnosis. Conventional methods generally focus on developing specialized algorithmsto address a single aspect, such as segmentation, classification, or localization of braintumors.Inthiswork,anovelmulti-tasknetworkwasproposed,modifiedfromthecon-ventional VGG16, along with a U-Net variant concatenation, that can simultaneouslyachieve segmentation, classification, and localization using the same architecture. Wetrain the classification branch using the Brain Tumor MRI Dataset, and the segmenta-tionbranchusingaBrainTumorSegmentationdataset.Theintegrationofourmethod’soutput can aid in simultaneous classification, segmentation, and localization of fourtypesofbraintumorsinMRIscans.Theproposedmulti-taskframeworkachieved97%accuracy in classification and a dice similarity score of 0.86 for segmentation. In addi-tion,themethodshowshighercomputationalefficiencycomparedtoexistingmethods.Our method can be a promising tool for assistive diagnosis in low-resource healthcaresettingswhereskilledradiologistsarescarce.
Fabrication of Tannic Acid Crosslinked Gelatin-CMC Based Electrospun Matrix for Combatting Antibacterial Resistance in Infected Wounds
(Department of Biomedical Engineering, BUET, 2024-09-14) Jinia, Rafia Hasnat; Arafat, Dr. Muhammad Tarik
The global rise in antibiotic-resistant infections has become a critical challenge in wound management, with conventional treatments often proving ineffective. Electrospun nanofibers offer a promising solution by providing a versatile platform that combines antimicrobial efficacy with enhanced wound healing capabilities. Herein, this study pioneers an advanced electrospun nanofiber matrix, designed to revolutionize wound care by simultaneously addressing antibiotic resistance and promoting superior wound healing. The naofibrous matrix integrates gelatin and CMC, with TA serving as a crosslinker, significantly reinforcing both its structural integrity(up to 1.70±0.46 MPa) and biocompatibility (< 2%). The innovative synergy of these biopolymers endows the matrix with extraordinary antimicrobial potency, exhibiting unmatched efficacy against both pathogenic Gram-positive and Gram-negative bacterial strains—surpassing the capabilities of current commercial alternatives compromised by emerging resistance. This extraordinary performance stems from unmatched physicochemical properties of the developed matrix, featuring exceptional water uptake and a beneficial amorphous state, which together create an ideal environment for rapid, superior wound healing. In vitro assays, alongside excisional wound studies in mice, demonstrated that wounds treated with the fabricated matrix exhibited significantly accelerated and more effective healing compared to the control groups, achieving up to 95% wound closure across all bacterial species tested. To unravel the sophisticated interactions between matrix components and bacterial cells, in silico simulations were employed, providing a deep mechanistic understanding of its superior performance. The findings underscore the potential of this advanced electrospun matrix as a multifaceted platform for next-generation wound care, providing a robust solution to the escalating issue of antibiotic-resistant infections while fostering an optimal healing environment.
Immobilization of antimicrobial peptide on novel PCL-sodium alginate fiber to address resistant bacteria in surgical site infection
(Biomedical Engineering, 2023-11-06) Taufiq Hasan Aneem; Muhammad Tarik Arafat, Dr.
Surgical site infections (SSIs) caused by pathogenic bacteria lead to delayed wound healing, extended hospital stay, and thus, create a burden in healthcare. The overuse and misuse of antibioticsinmoderntimeshavecausedasurgeinSSIinahospitalsettingandcommonlyavailable antibiotics are proving to be ineffective against them. Antimicrobial peptides (AMPs) can be a potentialsolutiontopreventSSIandacceleratewoundhealingbecauseoftheirbroadspectrumof antimicrobial activities. Therefore, AMPs were extracted from natural sources and incorporated onto an engineered microfiber to be used as surgical suture. To accomplish this, two polymers of opposite characteristics — a hydrophilic polymer sodium alginate (SA) and a hydrophobic one polycaprolactone (PCL), were used to manufacture microfibers via a novel wet-spinning method. Wet-spinning of PCL and SA is difficult because of their immiscibilities and tendencies to create a complex gelation upon mixing. To circumvent this problem, PCL solution in acetone was dissolvedinaSAaqueoussolution.Thepolymersolutionsactedascoagulantsforeachotherwhen PCL molecules encapsulated SA molecules via hydrogen bonding. AMPs were then immobilized onto the fibers exploiting the self-polymerization behavior of dopamine. FTIR results confirmed the successful integration of both polymers and peptides. Conjugation between PCL and SA resulted in fiber with a smooth surface improving the crystallinity and mechanical strength of the fibers via hydrogen bond. Having an average diameter of 220 µm, the mechanical properties of the fiber complied with USP standards for suture of size 3-0. Due to the antibacterial activity of AMP, the microfibers were able to hinder the growth of Proteus spp., a pathogenic bacterium for at least 60 hours which was not the case when the antibiotic ceftazidime was used against it. AttachmentofProteusspp.onfibersurfacewasalsosignificantlyhinderedcomparedtoVicryl,a commercialsutureasfoundbythedirectcontactresponsestudy.Whensubjectedtoinvivostudy, accelerated wound healing was observed when the wound was closed using the engineered fiber comparedtocommercialsutureVicryl.Tracesofwoundswereobservedfromdermoscopicimages even after 14 days in all cases except for PCL-SA-pep. Histological results revealed the superior wound regeneration ability of the manufactured fiber. PCL-SA-pep fiber promoted faster re- epithelialization in just 3 days as characterized by a continual decrease in epithelial gap in the wound, did not elicit much inflammatory reactions, and reduced scar area. Neoangiogenesis and growthoffollicularstructureswerealsoobservedtobesignificantlyhigherinPCL-SA-peptreated wounds compared to Vicryl proving the higher wound healing capacity of the engineered microfiber.
Impact of hematological and morphological variations at circle of willis on cerebrovascular hemodynamics
(Department of Biomedical Engineering (BME), BUET, 2025-05-19) Jafrin Sultana; Tarik Arafat, Dr. Muhammad
Cerebrovascular diseases, including ischemic stroke and intracranial hemorrhage, remain leading global health burdens. The Circle of Willis (CoW), a critical arterial network for cerebral perfusion, is highly influenced by hematological factors, such as blood viscosity and hematocrit levels, as well as anatomical variations like hypoplasia, agenesis, and asymmetry. This study employs Computational Fluid Dynamics (CFD) simulations on patient-specific CoWs’ geometries to investigate the combined impact of hematocrit variation, stenosis morphology, and anatomical abnormalities on intracranial hemodynamics. Using ANSYS software and modeling blood as a non-Newtonian Carreau fluid under pulsatile flow, key hemodynamic indices as wall shear stress (WSS), time average wall shear stress, oscillatory shear index, velocity, translesional pressure ratio, pressure drop index, and stroke risk index, were computed to characterize cerebrovascular behavior under diverse physiological and pathological states. The influence of varying hematocrit levels on wall shear stress and cerebral perfusion is observed using computational fluid dynamics models in normal and aneurysmal geometries. Elevated hematocrit increased blood viscosity and WSS, while lower hematocrit led to reduced shear forces. The WSS-viscosity relationship was nonlinear: low WSS regions were linked to endothelial apoptosis and aneurysm formation, whereas high WSS areas correlated with increased rupture risk. Stenosis geometry and CoW integrity as key determinants of hemodynamic cooperation were identified through nine design of experiments framework. Irregular stenosis and anatomical incompleteness impaired collateral flow, particularly during occlusion, elevating ischemic susceptibility. Hemodynamic metrics indicated that local stress variations promoted atherogenesis and increased thromboembolic risk. Results emphasize the correlation among CoW morphology, hematological factors, and flow dynamics in cerebrovascular pathology. CFD modeling, integrated with clinical data, offers a robust platform for individualized risk assessment and therapeutic planning. Future research incorporating real-time imaging and AI-driven analysis may enhance predictive accuracy and improve stroke prevention strategies. Keywords: Stroke; Ischemic Stroke; Hematological and Morphological Variation; Circle of Willis; Cerebrovascular Hemodynamics; Computational Fluid Dynamics
Investigation of chitosan/starch- based polyphenolic acid crosslinked antibacterial microparticles for rapid hemostasis
(Department of Biomedical Engineering (BME), BUET, 2024-07-01) Tithy, Lamiya Hassan; Arafat, Dr. Muhammad Tarik
Hemorrhage is a major contributor to avoidable fatalities worldwide. It occurs when there is excessive blood loss due to accidents, battles, or surgical procedures. The preparation of rapid hemostatic materials along with strong bactericidal property and minimum erythrocyte lysis for deep hemorrhage still remains a challenge. Herein, a series of novel microparticle hemostats made with different polysaccharides such as chitosan, starch, cross linked with unoxidized tannic acid was prepared via hydrothermal treatment and facile ionic gelation method. Hemostats' comparative functional properties, such as adjustable antibacterial and erythrocyte compatibility upon various starch additions were evaluated. The chemical bonds between the constituent materials were verified by ATR-FTIR. The microparticles had excellent hemostatic property and could concentrate blood cells, blood proteins and clotting factors as well as could induce electrostatic interaction between positively charged chitosan and negatively charged blood cells. Blood clotting index (BCI), red blood cell (RBC) adhesion was performed to verify its hemostatic capability for in vitro study. Prothrombin time (PT) and activated partial thromboplastin time (aPTT) were employed to find out the composite’s hemostatic mechanism. For in vivo hemostatic study, mice liver laceration and rat tail amputation were performed to evaluate the composite’s efficacy in real injury scenario. The in vivo hemostatic study reveals that the developed hemostats for mouse liver laceration and rat tail amputation had clotting times (13 s and 38 s, respectively) and blood loss (51 mg and 62 mg, respectively) similar to those of CeloxTM. These studies suggest that the prepared microparticles can be employed in deep wound to halt bleeding. The erythrocyte adhesion test demonstrated that erythrocyte lysis can be lowered by modifying the antibacterial hemostats with different starches. Antibacterial efficacy of the hemostats remained intact against S. aureus (>90%), E. coli (>80%), and P. mirabilis bacteria upon starch modification. In comparison with CeloxTM, which does not have any antibacterial property against gram negative bacteria, the developed hemostats microparticles have broad-spectrum antibacterial activity. They also demonstrated high hemocompatibility (<3% hemolysis ratio), moderate cell viability (>81%), in vivo biodegradation, and angiogenesis indicating adequate biocompatibility and wound healing. These results suggest that the developed series of microparticle hemostats have a great promise as rapid hemostatic agent with strong antibacterial property to be used in emergency situations for preventing massive hemorrhage.
Leveraging frequency domain attributes for multi-modal medical image segmentation using convolutional neural networks
(Department of Biomedical Engineering, BUET, 2025-05-28) Shams Nafisa Ali; Hasan Al Banna, Dr. Taufiq
Recent advances in deep learning have significantly enhanced medical image segmen- tation. As medical data becomes increasingly diverse and complex, the need for ar- chitectures that can generalize across modalities and anatomical structures has grown paramount. While CNNs, Transformers, and their hybrid architectures have addressed issues such as limited receptive fields and redundant feature representations, most mod- els remain confined to the spatial domain—overlooking the frequency domain’s rich structural cues. Some recent studies have explored spectral information at the feature level; however, frequency-domain integration at the supervision level remains largely untapped. To this end, we propose Phi-SegNet, a CNN-based architecture that incor- porates phase-aware cues at both architectural and optimization levels. The network integrates Bi-Feature Mask Former (BFMF) modules that blend neighboring encoder features to reduce semantic gaps, and Reverse Fourier Attention (RFA) blocks that re- fine decoder outputs using phase-regularized embeddings. A dedicated phase-aware loss aligns these embeddings with structural priors, forming a closed feedback loop that emphasizes boundary precision. Evaluated on five public datasets spanning ultra- sound, X-ray, histopathology, MRI, and colonoscopy, Phi-SegNet consistently achieves state-of-the-art performance, particularly excelling in fine-grained boundary segmen- tation tasks. On average, across these five datasets, Phi-SegNet achieves a relative improvement of 1.54% ± 1.26% in IoU and 1.10% ± 0.69% in F1-score over the next best-performing model for each dataset. Additionally, under generalized training using a unified dataset comprising all five modalities, as well as in cross-dataset generaliza- tion scenarios involving unseen datasets from the known domain, Phi-SegNet exhibits robust and superior performance—highlighting its adaptability and modality-agnostic design. These findings demonstrate the potential of leveraging spectral priors in both learning and supervision, offering a new direction toward generalized, universal, and anatomically precise segmentation frameworks.
Multimodal feature fusion based thoracic disease classification framework combining medical data and chest x-ray images
(Department of Biomedical Engineering(BME), BUET, 2023-06-19) Nusrat Binta Nizam; Al Banna, Dr. Taufiq Hasan
Chest X-rays are commonly used in clinical settings to diagnose thoracic diseases, es- pecially in low-resource settings. However, interpreting these images can be challeng- ing, particularly in resource-constrained environment. Current AI-based methods focus solely on the X-ray images without considering relevant clinical information. To effec- tively assist with limited resources, it is important for a computerized system to gen- erate decisions relevant to those of radiologists. This requires incorporating pertinent clinical details, such as medical history, symptoms, and demographic information, into image-based computerized systems to enhance their performance. The development of AI-based systems faces two main challenges: the limited availability of comprehensive medical image datasets suitable for machine learning and the difficulty in reproducing the advanced reasoning abilities of experienced radiologists, who have undergone ex- tensive training and accumulated expertise. In this work, at first an unimodal anatomy aware network is proposed which provided about 11% relative improvement in mean square error (MSE) compared to existing methods when evaluated on a dataset for pre- dicting the severity of COVID-19 pneumonia. This model also exhibits promising re- sults on an unseen clinical evaluation dataset which provides evidence of the efficacy of anatomy-aware architecture for predicting the severity of COVID-19 disease. Addi- tionally, this thesis proposes a multimodal feature fusion framework to improve disease classification by combining medical data and image information. Existing approaches rely on textual information, lacking anatomical details. An advanced multimodal fea- ture fusion-based approach is needed to enhance disease classification accuracy. In this study, a comparison of incorporating clinical information demonstrates the substantial value of patient indication data (i.e., medical history, demographics, symptoms) in dis- ease classification. Incorporating such information enables computer-aided systems to function more closely to radiologists. The proposed feature fusion-based framework, ResVCBERT and DenseVCBERT exhibit a significant improvement in accuracy com- pared to baseline architectures, even when there are errors in the textual information. The proposed DenseVCBERT provided significant improvement with an accuracy of about 88.44% using the OpenI dataset of radiological reports and chest X-rays. Includ- ing anatomical information in deep learning models through feature fusion enhances the accuracy of AI-based frameworks, as demonstrated in the analysis of COVID-19 pneu- monia severity prediction. This approach aids disease diagnosis and severity prediction, benefiting radiologists in developed and underdeveloped nations.
Natural polysaccharide modified olive oil based nanoemulsion as a controlled delivery vehicle for topical formulations
(Department of Biomedical Engineering (BME), BUET, 2024-06-23) Kobra., Khadijatul; Arafat, Dr. Muhammad Tarik
Nanoemulsions facilitate efficient drug delivery through the skin, overcoming the limitations of traditional topical formulations. However, their poor stability, low viscosity, and rapid drug release restrict their application in topical delivery. Nanoemulsions with enhanced stability, modified rheology, and increased viscosity, such as shear-thinning behavior, are well-accepted by patients due to their dosage integrity, controlled drug release properties, and ease of application. For oil-in-water emulsions, these challenges can be addressed by optimizing nanoemulsion formulation and modifying the rheological behavior of the continuous phase with biopolymers without significantly impacting the emulsion properties. Therefore, in this study, olive oil nanoemulsion was formulated and modified with xanthan gum and gum acacia to explore as a potential controlled topical delivery vehicle. Oil-in-water nanoemulsion formulated with the optimized composition of olive oil, tween 80, and water was used as the drug carrier and further modified with natural polysaccharides. The effect of gum on nanoemulsion different physiochemical characteristics, stability, rheology, drug release, and encapsulation efficiency were investigated. Results showed that the developed nanoemulsion behaved as low viscosity Newtonian fluid and released 100% drug within 6 h. Modification with xanthan and gum acacia had significantly improved formulation viscosity, drug encapsulation efficiency (> 85%), and controlled drug release by up to 40% with a release pattern following Korsmeyer–Peppas model. Additionally, xanthan gum modified formulation exhibited shear thinning rheology by forming an extended network in the continuous phase, whereas gum acacia modified formulation behaved as Newtonian fluid at high shear rate (> 200 s-1). Furthermore, xanthan gum modified formulations had improved zeta potential, stability, monodispersity, and hemocompatibility and showed high antibacterial activity against S. aureus than gum acacia modified formulations. These results indicate the higher potential of xanthan gum modified formulation as a topical delivery vehicle. In addition, the skin irritation test confirmed the safety of the developed formulations for topical use. Finally, an in vivo skin penetration test was conducted to assess the distribution and penetration of ciprofloxacin in various layers of the skin. The study found that the highest concentration of the drug was present in the stratum corneum, with drug concentration decreasing in the deeper layers of the skin.
Towards organ separation from projection radiographs using a semi-supervised deep learning-based technique
(Department of Biomedical Engineering (BME), BUET, 2024-03-25) Kawsar Ahmed, Md.; Banna, Dr. Taufiq Hasan Al
The objective of this thesis is to develop a novel deep-learning method to separate organ-specific tissue images from projection radiographs, facilitating improved disease diagnosis by providing focused diagnostic information alongside conventional radiographs. This study proposes OrGAN, a generative adversarial network based model to translate chest X-rays into lung tissue images. It consists of a U-Net generator modified with a domain classifier and gradient reversal layer for domain adaptation, and a CNN discriminator. OrGAN was trained on 779 paired synthetic X-ray/lung images generated from CT data, alongside 15,000 unpaired real X-ray images from the VinDr-CXR dataset. Qualitative evaluation involved radiologist assessment of visual quality, anatomical accuracy and diagnostic utility on 10 test cases. Quantitative assessment compared lung disease classification performance using deep learning models trained on generated lung tissues, segmented lungs, and original X-rays across two independent datasets - VinDr-CXR (6 labels) and COVIDx-CXR-4 (COVID vs normal). OrGAN achieved high performance in generating lung tissue images on synthetic X-ray data (PSNR 28.8dB, SSIM 0.944 on test set). Qualitative evaluation by 5 radiologists indicated that the generated images enhanced visibility of lung features, preserved diagnostic information, and complemented X-rays for improved diagnosis. In quantitative tests, the DenseNet model trained on lung tissues outperformed those trained on segmented lungs, and actual X-ray, showing statistically significant (p-value < 0.05) improvements in overall F1-score, and sensitivity for a multilabel (6 labels) disease classification task on the VinDr dataset. On COVIDx, both of the two classifier models (DenseNet and ResNet) achieved higher overall F1-scores for a multiclass (2 classes) disease classification task, compared to using original X-rays or segmented lungs. The proposed OrGAN model effectively separates interpretable lung tissue images from X-ray projections. The generated images retain diagnostic fidelity, offer additional diagnostic insights to complement conventional X-rays, and can improve performance of deep learning models for precise disease diagnosis. Thus, the study proposes a potential pathway for further exploration into organ-specific tissue image separation from projection radiographs.