مشاريع طلاب خريف 2023 _ F23
Deep learning Model to diagnose retinal diseases using Optical Coherence Tomography images (OCT)
Background:
There are numerous eye conditions but the most two common retinal conditions are Age- Related Macular Degeneration (AMD), which the sharp, central vision and a leading cause of vision loss among people age 50 and aged, there are two types of AMD are wet AMD (Choroidal neovascularization (CNV)) and dry AMD (DRUSEN). Diabetic Macular Edema (DME), which is a complication of diabetes that can affect the fovea and it caused by fluid accumulation in the macula. thus, early discovery of conditions is a critical significance. The goal of this thesis is implementation of deep learning model used to detect four types of retinal cases (NORMAL, CNV, DME, and DRUSEN) by using Convolutional Neural Network (CNN) to Avoid manual diagnostic errors and help doctors make faster, and more accurate diagnoses.
Aim:
The goal of this thesis is implementation of deep learning model used to detect four types of retinal cases (NORMAL, CNV, DME, and DRUSEN) by using Convolutional Neural Network (CNN) to Avoid manual diagnostic errors and help doctors make faster, and more accurate diagnoses.
Materials and Methods:
To diagnose retinal disorders utilizing Optical Coherence Tomography (OCT) scans, this investigation developed a methodology based on image pre-processing and convolutional neural networks (CNNs) (a deep learning method). A publicly available OCT dataset containing four categories (normal, CNV, DME, drusen) was used. Data imbalance was addressed using oversampling techniques. Image pre-processing techniques including images resizing, noise reduction. The AlexNet architecture, a convolutional neural network (CNN), was utilized for image classification. The model comprised five convolutional layers, fully-connected layers, and dropout layers. Categorical cross-entropy loss function and Adam optimizer were employed. Overfitting was mitigated using techniques like learning rate reduction, early stopping, and model checkpointing. Training and validation metrics were logged for analysis.
Results:
Our deep learning model achieved high accuracy in classifying retinal diseases from OCT images. The model achieved on the test set an accuracy of 99.38%, with minimal misclassifications between categories. These results are comparable to the best performing models in previous studies (96.54% - 99.48%). The proposed model has low computing costs in comparison to other studies in the literature
Conclusion:
This study explored deep learning application to classify retinal diseases by using OCT images. We obtained a high test accuracy of 99.38% and this indicate the potential of deep learning models for aiding in the diagnosis of retinal diseases, potentially improving patient outcomes. Future work will focus on exploring hybrid algorithms for further accuracy improvement.
إعداد: الطالبة ريما عثمان ملحم
إشراف: الدكتور ينال أحمد القدسي
Deep learning Model to diagnose retinal diseases using Optical Coherence Tomography images (OCT)
تصميم لقاح افتراضي ضد جرثومة الملوية البوابية بمساعدة أدوات المعلوماتية المناعية
Designing a virtual multi-epitope vaccine against Helicobacter pylori using immunoinformatic tools
Helicobacter pylori is a gram-negative, spiral, microaerophilic bacterium that infects the stomachs of over 50% of the global human population. It is commonly acquired during childhood and, if left untreated, can persist chronically, leading to conditions such as chronic gastritis, peptic ulcer disease, gastric adenocarcinoma, and gastric B cell lymphoma. The current treatment approach involves proton-pump inhibitors and antibiotics, but it faces challenges like patient compliance, antibiotic resistance, and potential recurrence of infection. Developing an effective vaccine against H. pylori would offer significant advantages.
During this study a virtual vaccine against Helicobacter pylori was designed using immunoinformatics approaches targeting specific antigens known to be involved in the pathogenesis of the infection. The selected protective antigens include Vacuolating cytotoxin autotransporter (vacA), Neuraminyllactose-binding hemagglutinin (hpaA), Diaminopimelate decarboxylase (lysA), Alpha-(1,3)-fucosyltransferase FucT (fucT), Shikimate dehydrogenase (NADP(+)) (aroE), Outer inflammatory protein A (oipA), Outer membrane proteins(omp 6/18), Urease (ureB), and IceA2.
The epitopes underwent a thorough filtering process, including tests for antigenicity, toxicity, allergenicity, and cytokine inducibility, with the primary aim of identifying epitopes capable of triggering both T and B cell-mediated immune responses. To enhance vaccine immunogenicity, the final epitopes were fused with the Heat-labile enterotoxin B chain from E.coli (LTB) adjuvant using appropriate linkers, resulting in the development of a multi-epitope vaccine. The selected T cell epitope ensemble is expected to cover 87.09% of the global human population.
Furthermore, docking studies were conducted to assess the interaction between the vaccine and Toll-like receptor 2 (TLR2) / Toll-like receptor 4 (TLR4), demonstrating significant affinity, consistency, and stability. This research aimed to design an effective subunit vaccine
capable of eliciting both humoral and cellular immune responses, with the objective of addressing antibiotic-resistant Helicobacter pylori infections.
إعداد: الطالبة منال غسان العلاوي
إشراف: الدكتور آية طوير
Designing a virtual multi-epitope vaccine against Helicobacter pylori using immunoinformatic tools
تحليل التنوع البيولوجي لميكروبيوم الأمعاء البشرية في الأفراد الأصحاء باستخدام أدوات المعلوماتية الحيوية
Biodiversity Analysis of the Human Gut Microbiome in Healthy Individuals Using
Bioinformatics Tools
Background and Aim: The human gut microbiome plays a crucial role in maintaining health and preventing disease. This study aims to provide a comprehensive analysis of the gut microbiome composition in healthy individuals using integrated bioinformatics tools.
Methods: We analyzed gut microbiota samples from ten healthy individuals using UGENE software, employing taxonomy tools such as Kraken, DIAMOND, and MetaPhlan. Alpha diversity indices, including the Shannon and Simpson diversity indices, were calculated. Correlation analyses were performed to explore relationships between microbial diversity, age, and geographic living conditions.
Results: The gut microbiome showed significant diversity across samples. Alpha diversity indices indicated high microbial diversity, which tended to decrease with age. Individuals living in mountainous regions exhibited lower diversity than those in villages.
Conclusion: This study highlights the complex diversity of the human gut microbiome and its variation with age and geographic location. The presence of both indigenous microbiota and pathobionts genera can lead to possible dysbiosis within the gut ecosystem. High microbial diversity is associated with better health outcomes, emphasizing its importance in maintaining gut health. Future research should aim to further elucidate the functional roles of these microbial communities.
إعداد: الطالبة تيما مهند السبيعي
إشراف: الدكتور عبد القادر عبادي
Biodiversity Analysis of the Human Gut Microbiome in Healthy Individuals Using Bioinformatics Tools
In silico identification of hub-genes in attributing the development, prognosis, and diagnosis of Pancreatic Ductal Adenocarcinoma using Weighted gene correlation network analysis of lncRNA and interactome analysis of their products.
Pancreatic Ductal Adenocarcinoma (PDAC) stands out as one of the most lethal forms of cancer, exhibiting a mere 11% 5-year survival rate when considering all stages collectively. The challenging prognosis is largely a result of delayed diagnosis owing to the highly aggressive nature of the tumor and the absence of early specific symptoms. Treatment of PDAC is further hindered by resistance to targeted therapies, immunotherapy, chemotherapy, and radiation, necessitating novel strategies to overcome this resistance. The present study endeavors to address these challenges by conducting an integrated bioinformatics analysis to elucidate the molecular mechanisms, identify diagnostic markers, and pinpoint therapeutic targets for PDAC. A considerable portion of the human genome is transcribed into long noncoding RNAs (lncRNAs) with the number of identified lncRNAs steadily increasing, although only a few have been functionally characterized. Notably, the functions of lncRNAs are closely linked to their subcellular localization, with cytoplasmic lncRNAs influencing mRNA stability, translation, and serving as miRNA sponges, while nuclear-retained lncRNAs have been implicated in transcriptional regulation, chromosome scaffolding, alternative splicing modulation, and chromatin remodeling. These diverse roles of lncRNAs play vital regulatory functions in both normal cellular processes and disease states. The analysis leveraged a normalized RNA-seq dataset from the GEO database, specifically the dataset with the accession number GSE133684 encompassing complete PDAC patient samples and full healthy controls. From this analysis, 451 differentially expressed genes (DEGs) were pinpointed. Subsequent Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway analyses were executed to shed light on the biological processes and crucial pathways associated with PDAC. Following this, a protein-protein interaction (PPI) network was established, leading to the identification of 11 hub genes and five key modules. Notably, all identified hub genes, including SUZ12, EZH2, YWHAZ, CTNNA1, SMARCA5, H2AC6, H2BC11, H2BC12, H2BC21, H3-3B, and H3C12, exhibited overexpression in PDAC patients. The validation of these hub genes was carried out in-silico using expression profiles and reference datasets and published studies. Functional enrichment analysis underscored the pivotal roles of these key genes in various biological processes, encompassing cytokine signaling in the immune system, cell cycle regulation, and apoptosis-related mechanisms such as DNA damage response and oxidative activities. Moreover, the study predicted and validated 6 lncRNA transcripts responsible for regulating 25 miRNAs and 7 key transcription factors known to govern a significant portion of the hub genes linked to PDAC. These lncRNAs are (ENST00000641741, ENST00000637240, ENST00000434887, ENST00000500800, ENST00000509873, ENST00000564694). This comprehensive bioinformatics analysis not only provides valuable insights into the pathogenesis of PDAC but also identifies potential biomarkers for clinical management and unveils novel drug targets. These findings advance our comprehension of PDAC and present promising prospects for further progress in diagnostics and therapeutic interventions.
إعداد: الطالب عبد الرحمن محمد بشار البابا
إشراف: الدكتورة لمى يوسف
Identification and genetic characterization of Sars-CoV-2 in Syria (2023) by using Oxford Nanopore Technology
Background:
The COVID-19 virus represented a difficult stage in modern history that the world has gone through since the beginning of its spread in Wuhan province in China on December 12, 2019, and quickly turned into an epidemic in March 2020, leaving millions of victims and injured. Vaccines and monoclonal antibodies have contributed to reducing the number of deaths and alleviating the severity of the symptoms until the WHO announced the end of the public health emergency in an important plan towards the end of the Corona pandemic https://www.who.int/. The first Syrian case of COVID-19 was reported in March 2020. over time, Measures like lockdowns, social distancing, and vaccination drives have been enforced in Syria to manage virus spread and safeguard the populace despite the difficult health and living conditions. The Omicron variant (B.1.1.529) has become the dominant type by sublineages and the only VOC circulating until now.
Methods:
we used Oxford Nanopore Technology to detect COVID-19, generating consensus sequences data for 24 samples that consist of the base for analyzing data and extracting results using many bioinformatics tools that varied between processing data, studying mutations, and phylogenetic trees. determine the genetic lineage for samples in Syria using three platforms Pangolin, Nextclade, and GISAID database.
Results:
out of a hundred samples only 24 samples were sequenced after giving high CT values. Through tracking viral variants in Syria, the XBB recombinant Omicron variant circulated at time Jun, July, and August. XBB.142.1(75%) in 18 sample was designed as Syrian lineage, XBB.2.3.11(16.67 %) in 4 samples, XBB.1.9.1(4.17%) and FL.12 (4.17%). the clades were 22F, 23E, and 23D. The clade in GISAID is GRA unless one sample GR. 19 samples from this study were submitted to GISAID (Global Initiated on Sharing All Influenza Data). We apply a genome-based survey for mutations and evaluate the genetic diversity and the extent of its relationship with the changes recorded in the world through the phylogenetic tree with neighboring countries showing the relation in genetics and the prevalence.
Conclusion:
This is the first study that has provided epidemiological descriptive and valuable insights into the genetic variation in the Syrian population, hoping to introduce valuable analysis used in more studies to evaluate the vaccines and medicines in our country.
إعداد: الطالبة شذا عبد العظيم جحجاح
إشراف: الدكتور حيان حسن
Deep Learning-based System for Automated Classification and Detection of Lesions in Dental Panoramic Radiography
The field of dentistry is highly dependent on the analysis of dental radiographs to guide diagnosis of disease. The integration of AI applications based on deep learning in analyzing panoramic radiographs aims to offer highly accurate, efficient, and automated detection of dental lesions, aiding dentists in diagnosing these lesions while reducing their time and risk of misdiagnosis. These applications enhance the accuracy and allow for early intervention, ultimately improving patient outcomes in dental care. The aim of this project is to develop a deep learning-based system for the automated classification and detection of lesions in dental panoramic radiography. Utilizing the Tufts Dental Database (TDD), which comprises 1000 dental panoramic images, a classification step using traditional CNN model was implemented to classify X-Ray images into normal and abnormal classes, then an object detection system was developed using Faster R-CNN model, to detect abnormal dental lesions. The dataset was augmented to enhance the model's robustness and performance, then split into training, validation, and testing sets in a 70-15-15 ratio, respectively. After classifying the images as abnormal by CNN, the Faster RCNN model provides bounding boxes around detected lesions in abnormal regions. The system showed promising results for the classification step with accuracy of 84.56 and for the detection stage with a precision of 82.35%, recall of 75%, F1 score of 78.5% and IoU of 55%.
إعداد: الطالبة كنانة محمد حجازي
إشراف: الدكتور ياسر خضرا
Respiratory Diseases Detection and Classification Based on Respiratory Voices Using Artificial Intelligence Methods
Background: Respiratory illnesses are a significant global health issue, accounting for a large proportion of smokers, addiction, air pollution, increased CO2, occupational hazards like chemical fumes, and common allergens. Despite advancements in detection and treatment, respiratory diseases remain a leading cause of lung cancer and deaths globally.
Deep learning and transformers, branches of artificial intelligence, have emerged as valuable tools for predicting respiratory diseases (RD) including COPD, asthma, and URTI, for early detection and treatment. Harnessing the power of DL and transformers in respiratory issues prediction holds promise for advancing personalized medicine and optimizing treatment strategies.
Aim of Study: The aim of this study is to find high accuracy and sensitive algorithms capable of predicting RD based on respiratory sounds. The study considers respiratory sounds to reduce harmful and surgical diagnosis for quickly intervention in the patient's treatment protocol for less mortality cases.
Material and Methods: This study utilized the respiratory sound database; The data is an open-source dataset that was compiled by two collaborative research teams based in Portugal and Greece. It includes a total of 920 recordings obtained from 126 subjects, up to 6898 respiratory cycles. The data contains patient ID, age, sex, diagnosis, acquisition tool, and the chest location.
The features were extracted and the dataset was split into training and testing sets for model development and evaluation (DL and transformer). Data preprocessing techniques were applied and Python libraries facilitated data manipulation and analysis.
Results: The accuracy of the algorithms varied, RNN model achieved the lowest accuracy of 87.68%, while transformer achieved better results up to 90% for MFCC feature. The best accuracy shown by CNN model which reached to 97.5%. The findings highlight the superior performance of CNN and the lowest performance of RNN model as the CNN is the most common model used for audio signal.
Conclusion: The study underscores the importance of selecting appropriate classification algorithms for predicting RD. The CNN model and Transformer offer promising results, while RNN may not be the most effective choice. These findings contribute to the field of respiratory system prognosis and provide insights for improving personalized treatment strategies.
إعداد: الطالبة رزان محمد معتز دخان
إشراف: الدكتور رؤوف حمدان
Chronic Kidney Disease Early Prediction Using Machine Learning.
In the human body, the kidneys, play the important role of filtering wastes and toxic bodies from the blood. Chronic kidney disease (CKD) is a condition in which the human kidneys are damaged and unable to filter the blood in a proper way. It is a nontransmissible disease that causes mortality of large numbers worldwide and is very expensive to properly detect and diagnose, therefore, CKD patients often reach its chronic stages, especially in countries with limited resources. Furthermore, CKD is a silent killer due to the lack of physical symptoms at the initial stage, but a steady loss of glomerular filtration rate (GFR) occurs over a period of time longer than three months. CKD is a fatal disease if left undetected as it leads to renal failure, in the worst cases. However, the early diagnosis of CDK can significantly reduce the mortality rate. Moreover, if CKD is predicted early and correctly, it results in an increased probability of successful treatment and prolongs the patient’s life. The advances in ML, in addition to predictive analytics, provide promising results which in turn prove the capability of prediction in CKD and beyond. The utilization of ML methods in nephrology enables the building of ML models to better detect the at-risk patients of CKD especially in primary care settings. The current study carries out a prediction-based method that helps in early detecting of CKD patients at the early stage. In this study, we utilize on of the boosting method, XGBoost to achieve a higher prediction accuracy for CKD. Various preprocessing steps are employed to achieve better prediction performance, along with suitable hyperparameter tuning and feature selection. We assessed the degree of importance of each feature in the dataset leading to CKD. The performance of the model was evaluated with accuracy. It attained 98 % accuracy for training and testing sets. The way the research was done leads to the conclusion that recent improvements in machine learning, along with the help of predictive modeling, make for an interesting way to find new solutions that can then be used to test the accuracy of prediction in the field of kidney disease and beyond.
إعداد: الطالبة سارة محمد خالد النقطه
إشراف: الدكتور ينال أحمد القدسي
Chronic Kidney Disease Early Prediction Using Machine Learning.