Yazar "Sehirli, E." seçeneğine göre listele
Listeleniyor 1 - 5 / 5
Sayfa Başına Sonuç
Sıralama seçenekleri
Öğe AN ALGORITHM TO DETECT THE RETINAL REGION OF INTEREST(Copernicus Gesellschaft Mbh, 2017) Sehirli, E.; Turan, M. K.; Demiral, E.Retina is one of the important layers of the eyes, which includes sensitive cells to colour and light and nerve fibers. Retina can be displayed by using some medical devices such as fundus camera, ophthalmoscope. Hence, some lesions like microaneurysm, haemorrhage, exudate with many diseases of the eye can be detected by looking at the images taken by devices. In computer vision and biomedical areas, studies to detect lesions of the eyes automatically have been done for a long time. In order to make automated detections, the concept of ROI may be utilized. ROI which stands for region of interest generally serves the purpose of focusing on particular targets. The main concentration of this paper is the algorithm to automatically detect retinal region of interest belonging to different retinal images on a software application. The algorithm consists of three stages such as pre-processing stage, detecting ROI on processed images and overlapping between input image and obtained ROI of the image.Öğe Detecting Face-Touch Hand Moves Using Smartwatch Inertial Sensors and Convolutional Neural Networks(Ismail Saritas, 2022) Alesmaeil, A.; Sehirli, E.As per World Health Organization (WHO), avoiding touching the face when people are in public or crowded places is an effective way to prevent respiratory viral infections. This recommendation has become more crucial with the current health crisis and the worldwide spread of COVID-19 pandemic. However, most face touches are done unconsciously, that is why it is difficult for people to monitor their hand moves and try to avoid touching the face all the time. Hand-worn wearable devices like smartwatches are equipped with multiple sensors that can be utilized to track hand moves automatically. This work proposes a smartwatch application that uses small, efficient, and end-to-end Convolutional Neural Networks (CNN) models to classify hand motion and identify Face-Touch moves. To train the models, a large dataset is collected for both left and right hands with over 28k training samples that represents multiple hand motion types, body positions, and hand orientations. The app provides real-time feedback and alerts the user with vibration and sound whenever attempting to touch the face. Achieved results show state of the art face-touch accuracy with average recall, precision, and F1-Score of 96.75%, 95.1%, 95.85% respectively, with low False Positives Rate (FPR) as 0.04%. By using efficient configurations and small models, the app achieves high efficiency and can run for long hours without significant impact on battery which makes it applicable on most off-the-shelf smartwatches. © 2022, Ismail Saritas. All rights reserved.Öğe The Efficiency of Ensemble Techniques in Predicting Thyroid Disorder: A Comparative Study(Institute of Electrical and Electronics Engineers Inc., 2022) Alsaadawi, M.; Sehirli, E.Data science is presently connected with a wide range of technical and scientific fields. Thyroid disorder is a widespread issue that affects a great variety of people. Hospitals report several forms of thyroid conditions. In this thesis, a thyroid disease prediction model has been created by classification and comparing traditional and Ensemble algorithms. A dataset including 1,250 records from the Iraqi people was utilized for the first-time using Ensemble methods. Stacking is one of the most effective Ensemble approaches for forecasting complicated structured data. Several metrics, including Accuracy, Precision, Sensitivity, Specificity, F-Score, and the Matthews correlation coefficient, were used to evaluate the performance of the prediction model. The experimental findings show that the proposed technique to optimize the detection of thyroid illnesses may be successfully implemented. The majority of Ensemble methods achieved 100 % accuracy with both the whole data set and the feature selection data set. In terms of precision and computational expense, the given findings outperform comparable models in their field. © 2022 IEEE.Öğe ESTIMATION OF POPULATION NUMBER VIA LIGHT ACTIVITIES ON NIGHT-TIME SATELLITE IMAGES(Copernicus Gesellschaft Mbh, 2017) Turan, M. K.; Yucer, E.; Sehirli, E.; Karas, I. R.Estimation and accurate assessment regarding population gets harder and harder day by day due to growth of world population in a fast manner. Estimating tendencies to settlements in cities and countries, socio-cultural development and population numbers is quite difficult. In addition to them, selection and analysis of parameters such as time, work-force and cost seems like another difficult issue. In this study, population number is guessed by evaluating light activities in Istanbul via night-time images of Turkey. By evaluating light activities between 2000 and 2010, average population per pixel is obtained. Hence, it is used to estimate population numbers in 2011, 2012 and 2013. Mean errors are concluded as 4.14% for 2011, 3.74% for 2012 and 3.04% for 2013 separately. As a result of developed thresholding method, mean error is concluded as 3.64% to estimate population number in Istanbul for next three years.Öğe A novel method for segmentation of QRS complex on ecg signals and classification of cardiovascular diseases via a hybrid model based on machine learning(Ismail Saritas, 2021) Sehirli, E.; Turan, M.K.Automated-detecting intelligent programs and methods are developing to find out diseases in medicine in recent years. Developing new methods and improving existing ones are currently ongoing research. One of the most important health problems is heart diseases for all people in the world. Electrocardiography (ECG) is a diagnosis tool that gives substantially functional information about heart and cardiac system. In this work, it is primarily aimed at developing an intelligent system based on ECG signal processing, analysis, and classification via a hybrid machine learning model. This work uses 837 ECG signal fragments that includes 7 different classes shared in MIT-BIH Arrhythmia database for one lead. The ECG signals are applied on a preprocessing to smooth signals and correct baselines. Q, R and S waves (QRS) complex on ECG signals are segmented based on k-means clustering and tracking local extrema points. Feature extraction and selection are then performed, and a dataset is created by calculating measurement parameters for each QRS points separately. Training sets and test sets based on 8-fold cross validation are generated. A hybrid model based on machine learning models including decision tree (DT), k-nearest neighbor (KNN), random forest (RF), naïve bayes (NB), linear discriminant analysis (LDA), support vector machines (SVM) and quadratic discriminant analysis (QDA) is developed to classify cardiovascular diseases (CVD) into 7 different classes such as normal sinus rhythm (NSR), atrial premature beat (APB), atrial fibrillation (AFIB), premature ventricular contraction (PVC), ventricular bigeminy (VB), left bundle branch block beat (LBBBB) and right bundle branch block beat (RBBBB). Sensitivity, specificity, accuracy, and Matthews correlation coefficient (MCC) of detection of QRS complex are obtained respectively as 94.75%, 95.96%, 95.57% and 0.90. Sensitivity, specificity, accuracy and MCC of classification of CVD classes are obtained respectively as 92.33%, 92.50%, 92.41%, 0.85. © 2021, Ismail Saritas. All rights reserved.
