Altınbaş University Institutional Repository
Not a member yet
5805 research outputs found
Sort by
Detecting myocardial infraction in ECG waveforms using YOLOv8
No full textThe detection of myocardial infarction (MI) using advanced imaging and analysis techniques is a pivotal advance- ment in medical diagnostics. This study employs YOLOv8, a sophisticated deep learning algorithm, for the identification of MI from the electrocardiogram (ECG) images, focusing specifically on the critical ST elevation segment associated with MI. Unlike traditional methods that rely on manual interpretation, our approach automates the detection process, significantly reducing the time and potential for human error. By fine-tuning YOLOv8 on a curated dataset and employing a loss-modified model, we achieved an average accuracy of 96% and a mean Average Precision (mAP) of 0.99 at a 50% confidence threshold, with a Recall of 97.7%. To make this technology accessible to medical staff and patients without technical expertise, we developed a user-friendly GUI that simplifies the detection process by clicking a button. This research not only highlights the potential of applying deep learning in cardiology but also emphasizes the importance of user-centered design in medical technology, promising significant improvements in clinical diagnosis and patient care
Using parenteral fish oil from birth may improve short and long term outcome in preterm infants
No full textBackground: Lipids are a crucial component of parenteral nutrition in its entirety. This study aimed to compare the short—and long-term outcomes of lipid solutions containing fish oil and standard parenteral lipid solutions in total parenteral nutrition of premature neonates. Methods: Preterm infants weighing less than 1500 grams or 32 weeks gestation admitted to Dr. Behçet Uz Children's Diseases and Surgery Training and Research Hospital Neonatal Intensive Care Unit between January 2018 and January 2020 were included in this retrospective cross-sectional study. Infants receiving only 3 g/kg/day Clinoleic were enrolled in Group 2; infants receiving 1 g/kg/day Omegaven plus 2 g/kg/day Clinoleic were enrolled in Group 1. Weight at discharge, laboratory data (liver enzymes, bilirubin levels), and long-term results were compared. Results: The study comprised 70 newborns, and 35 of them were in Group 1. Group 1 had a considerably lower direct bilirubin level (d bil p valueu; 1st week p=0.03, 2nd week p=0.87, 3rd week p=0.02). The omegaven group had a considerably higher weight upon discharge (p=0.02). Long-term neurodevelopmental results did not differ across the groups (MDI score p=0.33, PDI score p=0.11, NDI positivity p=1.00, MDP p=0.81). Conclusion: We demonstrated that the weight at discharge and laboratory measurements improved with Omegaven support. Early exposure to very high levels of oxidative stress may cause infants to employ their antioxidant system as a parenteral support system
Deep fake image detection based on deep learning using a hybrid CNN-LSTM with machine learning architectures as classifier
Get PDFIn the digital age, the proliferation of deep fake technology has emerged as a significant
threat, undermining trust and authenticity across various domains, including media, politics,
and security. This thesis presents a new deep learning architecture for finding deep fake
images. It uses the benefits of transfer learning in a framework that combines a
Convolutional Neural Network (CNN) and a Long Short-Term Memory (LSTM) network.
Our hybrid model combines the spatial feature extraction skills of CNNs with the temporal
sequence analysis skills of LSTMs. The goal is to make deep fake detection more accurate
and reliable. The model was developed using the Python programming language and trained
on two substantial datasets: the DFDC dataset and the Ciplab dataset, comprising a balanced
collection of 19,148 authentic and 19,148 deep fake images. During preprocessing, all
images were standardized to a resolution of 128 × 128 pixels to optimize training efficiency.
The training process employed the binary cross-entropy loss function to minimize
classification errors iteratively. Our experiments show that our hybrid CNN-LSTM model is
very stable. It got amazing results with accuracies of 97.32% on the DFDC dataset and
98.24% on the Ciplab dataset, with error rates as low as 0.15% and 0.26%, respectively.
These results underscore the model's ability to effectively identify deep fake images while
maintaining a low false positive rate. The model's success is attributed to its ability to
leverage CNNs for detecting visual anomalies indicative of manipulative artifacts and
LSTMs for capturing temporal dependencies within the image sequences. In conclusion, this thesis makes a significant contribution to the field of deep fake detection by presenting a
unique hybrid CNN-LSTM architecture that achieves remarkable accuracy and robustness.
The suggested model combines the strengths of CNNs and LSTMs to improve detection
performance and reliability. It is a flexible and effective way to deal with the problems
caused by deep fake technology
Detection of epileptic seizure using EEG signals analysis based on deep learning techniques
No full textThe brain neurons' electrical activities represented by Electroencephalogram (EEG) signals are the most common data for diagnosing Epilepsy seizure, which is considered a chronic nervous disorder that cannot be controlled medically using surgical operation or medications with more than 40 % of Epilepsy seizure case. With the progress and development of artificial intelligence and deep learning techniques, it becomes possible to detect these seizures over the observation of the non-stationary-dynamic EEG signals, which contain important information about the mental state of patients. This paper provides a concerted deep machine learning model consisting of two simultaneous techniques detecting the activity of epileptic seizures using EEG signals. The time-frequency image of EEG waves and EEG raw waves are used as input components for the convolution neural network (CNN) and recurrent neural network (RNN) with long- and short-term memory (LSTM). Two processing signal methods have been used, Short-Time Fourier Transform (STFT) and Continuous Wavelet Transformation (CWT), have been used for generating spectrogram and scalogram images with sizes of 77 × 75 and 32 × 32, respectively. The experimental results showed a detection accuracy of 99.57 %, 99.57 % using CWT Scalograms, and 99.26 %, 97.12 % using STFT spectrograms as CNN input for the Bonn University dataset and the CHB-MIT dataset, respectively. Thus, the proposed models provide the ability to detect epileptic seizures with high success compared to previous studies
ENHANCING PREDICTIVE PERFORMANCE IN COVID-19 HEALTHCARE DATASETS: A CASE STUDY BASED ON HYPER ADASYN OVER-SAMPLING AND GENETIC FEATURE SELECTION
No full textPredictive analytics is paramount in the health industry, where it finds its wide application, in that it helps increase the forecast's accuracy level based on big data. Most of the time, there is a tendency toward the imbalance of the datasets in healthcare. In this study, two COVID-19 datasets from Kaggle were used as a case study of dataset imbalance. In such scenarios of imbalanced datasets like COVID-19, conventional sampling methods like ADASYN (Adaptive Synthetic Sampling Approach for Imbalanced Learning) tend to yield only modest accuracy levels. To address another problem like finding the optimal features, this study proposes a novel approach that combines oversampling techniques with genetic feature selection (GFs) using laboratory data. This innovative method aims to construct machine -learning clinical prediction models for the identification of COVID-19 infected patients, leveraging two widely recognized datasets by using hyper ADASYN over -sampling and genetic feature selection, stands out for its unprecedented precision in identifying relevant features crucial for accurate predictions. Unlike the traditional approach, it can solve the class imbalance problem and tune the feature space to bring about a dramatic increase in accuracy, precision, recall, and overall predictive performance by using our hypermodel. Our approach significantly enhanced the performance of the classifier, and the Random Forest (RF) model with n trees classifies accurately to the limit of 99%, with precision 99%, recall 99%, and F1 -score 99% for each of the datasets. Decision Tree (DT) model achieved 92% with all metrics for Dataset I, and 95% with all metrics for Dataset II. Multilayer Perceptron (MLP) achieved 99% with all metrics, respectively, for both datasets. Gradient Boosting (XGB) achieved 97% for all metrics with dataset I and 98% with all metrics for dataset II. These results underscore the efficacy of our proposed method in balancing COVID-19 datasets and enhancing predictive accuracy
Sınıf II Bölüm 1 Maloklüzyonlara Tek Fazlı ve İki Fazlı Tedavi Yaklaşımları
No full textOrtodontide erken teşhis ve tedavinin önemi büyüktür. Sınıf II bölüm 1 maloklüzyonların görülme sıklığının fazla olmasın- dan dolayı farklı tedavi seçenekleri farklı dönemlerde tercih edilebilmektedir. Erken tedavinin sonraki sabit tedavinin süre- sini ve kompleksitesini, çekim ve hastaların cerrahi ihtiyacını azaltması literatürde tartışma konusu olmaktadır. Sınıf II bölüm 1 maloklüzyonların tek veya iki fazlı tedavilerinde tercih edilen yöntemin zamanlaması; hastanın tedaviye hazır hissetmesi, zaman, maliyet, hastanın ebeveynlerinin beklentileri gibi bazı faktörlere göre şekillenmektedir. Bu faktörler her hastaya özel belirlenmeli ve hastanın lehine olmalıdır. Literatürde hasta ko- operasyonu düşükse tek fazlı tedavi sonuçlarının tatmin edici olmayacağı ve bu yüzden tedaviye erken dönemde başlan- masının daha makul olacağını savunanlar ile beraber; tek fazlı Sınıf II maloklüzyon tedavisinin iki fazlı tedaviden daha iyi sonuçlar elde ettiğini savunan görüşler de mevcuttur. Bu ça- lışmanın amacı literatürde Sınıf II bölüm 1 maloklüzyonun iki fazlı ve tek fazlı tedavilerinin etkinliğini ve tedavi yöntemlerin zamanlamasını değerlendirmektir
Türkiye's relations with/within NATO: The ontological security dilemmas of Türkiye's transatlantic identity
No full textSince 1952, Türkiye's Transatlantic identity has been largely shaped by its NATO membership and interactions with the Allies. During the Cold War, Türkiye aligned its national interests and security concerns with NATO, making its membership a central element of its identity. However, starting in the 2000s, Türkiye shifted toward a more independent foreign policy, influenced by its new activism and regional ambitions, particularly in the Middle East. This shift has caused tensions with NATO, especially considering Türkiye's military interventions in Syria, its purchase of the S-400 missile defense system from Russia, and its reactions to the latest enlargement of NATO. This paper examines Türkiye's evolving Transatlantic identity through the lens of ontological security, according to which states act in order to provide continuity in their identities (Türkiye's Transatlantic identity) and to prevent any instabilities to feel ontologically secure, which then might create security dilemmas between the protective security cocoons (Türkiye's NATO membership) and the crises with other actors (Türkiye's clashes with/within NATO). In other words, this paper aims to understand how Türkiye's Transatlantic identity is still valued, desired, and reconstructed, despite the security dilemmas of Türkiye's contentions with/within the NATO Alliance, from the ontological security perspective
Innovative fluorescent polymers in niosomal carriers: a novel approach to enhancing cancer therapy and imaging
No full textCancer is anticipated to become the pioneer reason of disease-related deaths worldwide in the next two decades, underscoring the urgent need for personalized and adaptive treatment strategies. These strategies are crucial due to the high variability in drug efficacy and the tendency of cancer cells to develop resistance. This study investigates the potential of theranostic nanotechnology using three innovative fluorescent polymers (FP-1, FP-2, and FP-3) encapsulated in niosomal carriers, combining therapy (chemotherapy and radiotherapy) with fluorescence imaging. These cargoes are assessed for their cytotoxic effects across three cancer cell lines (A549, MCF-7, and HOb), with further analysis to determine their capacity to augment the effects of radiotherapy using a Linear Accelerator (LINAC) at specific doses. Fluorescence microscopy is utilized to verify their uptake and localization in cancerous versus healthy cell lines. The results confirmed that these niosomal cargoes not only improved the antiproliferative effects of radiotherapy but also demonstrate the practical application of fluorescent polymers in in vitro imaging. This dual function underscores the importance of dose optimization to maximize therapeutic benefits while minimizing adverse effects, thereby enhancing the overall efficacy of cancer treatments
Optimizing MIMO-OFDM communication systems: A comparative analysis of LMMSE and LZF equalizers under varying antenna configurations
No full textThe primary goal of this research is to examine the effectiveness of the linear minimum
mean square error (LMMSE) and linear zero forcing (LZF) as linear channel equalizers for
MIMO-OFDM communication systems. A cyclic prefix is added to each transmitted
OFDM symbol as part of a MIMO-OFDM-based system to totally suppress both intercarrier interference (ICI) and inter-symbol interference (ISI). The cyclic prefix is often
longer than or equal to the channel length, although occasionally it might be shorter due to
unanticipated channel behaviour. We suggest examining how the OFDM system performs
when the number of antennas and the type of equalizer are changed. as in case I&II.
Evaluation of the examined equalizers' performance in terms of Bit Error Rate (BER) for
2*2 and 3*3 systems is done using MATLAB Monte Carlo simulations
Foam-based antibacterial hydrogel composed of carboxymethyl cellulose/polyvinyl alcohol/cerium oxide nanoparticles for potential wound dressing
No full textFoam-based wound dressing materials produced by dispersing gas phases in a polymeric material are soft, adapt to the body shape, and allow the absorption of wound exudate due to their porous structure. Most of these formulations are based on synthetic substances such as polyurethane. However, biopolymers have entered the field as a new player thanks to their biocompatible and sustainable nature. Incorporating biopolymers in formulations is gaining interest in scientific literature, and we extend this approach by adding antibacterial cerium oxide nanoparticles to biopolymer formulation. We introduce a novel biopolymer composite of carboxymethyl cellulose (CMC), polyvinyl alcohol (PVA), and cerium oxide nanoparticles (CeO2 NPs), namely PVA-CMC@CeO2. This mixture was first foamed and then cross-linked with sodium tetraborate solution, followed by a freeze-thaw process. After the novel material's spectroscopic, structural, and morphological characterization, we investigated its swelling, drug-delivery, antibacterial, and biodegradability properties PVA-CMC@CeO2 dressing effectively inhibits Escherichia coli (E. coli) and Staphylococcus aureus (S. aureus) growth and delivers the antibiotic drug silver sulfadiazine for up to 6 h. The antibacterial properties, good swelling, and drug release profile of the blend material show promising potential in wound care applications