Tıp Fakültesi / Faculty of Medicine
Permanent URI for this collectionhttps://hdl.handle.net/11727/1403
Browse
7 results
Search Results
Item Evaluation of Machine Learning Algorithms for Renin-Angiotensin-Aldosterone System Inhibitors Associated Renal Adverse Event Prediction(2023) Guven, Alper Tuna; Ozdede, Murat; Sener, Yusuf Ziya; Yildirim, Ali Osman; Altintop, Sabri Engin; Yesilyurt, Berkay; Uyaroglu, Oguz Abdullah; Tanriover, Mine Durusu; 0000-0002-6310-4240; 37217407Background: Renin-angiotensin-aldosterone system inhibitors (RAASi) are commonly used medications. Renal adverse events associated with RAASi are hyperkalemia and acute kidney injury. We aimed to evaluate the performance of machine learning (ML) algorithms in order to define event associated features and predict RAASi associated renal adverse events.Materials and Methods: Data of patients recruited from five internal medicine and cardiology outpatient clinics were evaluated retrospectively. Clinical, laboratory, and medication data were acquired via electronic medical records. Dataset balancing and feature selection for machine learning algorithms were performed. Random forest (RF), k-nearest neighbor (kNN), naive Bayes (NB), extreme gradient boosting (xGB), support vector machine (SVM), neural network (NN), and logistic regression (LR) were used to create a prediction model.Results: 409 patients were included, and 50 renal adverse events occurred. The most important features predicting the renal adverse events were the index K and glucose levels, as well as having uncontrolled diabetes mellitus. Thiazides reduced RAASi associated hyperkalemia. kNN, RF, xGB and NN algorithms have the highest and similar AUC (> 98%), recall (> 94%), specifity (> 97%), precision (> 92%), accuracy (> 96%) and F1 statistics (> 94%) performance metrics for prediction.Conclusion: RAASi associated renal adverse events can be predicted prior to medication initiation by machine learning algorithms. Further prospective studies with large patient numbers are needed to create scoring systems as well as for their validation.Item Utilizing Deep Convolutional Generative Adversarial Networks for Automatic Segmentation of Gliomas: An Artificial Intelligence Study(2022) Aydogan Duman, Ebru; Sagiroglu, Seref; Celtikci, Pinar; Demirezen, Mustafa Umut; Borcek, Alp Ozgun; Emmez, Hakan; Celtikci, Emrah; 34542897AIM: To describe a deep convolutional generative adversarial networks (DCGAN) model which learns normal brain MRI from normal subjects than finds distortions such as a glioma from a test subject while performing a segmentation at the same time. MATERIAL and METHODS: MRIs of 300 healthy subjects were employed as training set. Additionally, test data were consisting anonymized T2-weigted MRIs of 27 healthy subjects and 27 HGG patients. Consecutive axial T2-weigted MRI slices of every subject were extracted and resized to 364x448 pixel resolution. The generative model produced random normal synthetic images and used these images for calculating residual loss to measure visual similarity between input MRIs and generated MRIs. RESULTS: The model correctly detected anomalies on 24 of 27 HGG patients' MRIs and marked them as abnormal. Besides, 25 of 27 healthy subjects' MRIs in the test dataset detected correctly as healthy MRI. The accuracy, precision, recall, and AUC were 0.907, 0.892, 0.923, and 0.907, respectively. CONCLUSION: Our proposed model demonstrates acceptable results can be achieved only by training with normal subject MRIs via using DCGAN model. This model is unique because it learns only from normal MRIs and it is able to find any abnormality which is different than the normal pattern.Item Importance of Systematic Right Ventricular Assessment in Cardiac Resynchronization Therapy Candidates: A Machine Learning Approach(2021) Galli, Elena; Le Rolle, Virginie; Smiseth, Otto A.; Duchenne, Jurgen; Aalen, John M.; Larsen, Camilla K.; Sade, Elif A.; Hubert, Arnaud; Anilkumar, Smitha; Penicka, Martin; Linde, Cecilia; Leclercq, Christophe; Hernandez, Alfredo; Voigt, Jens-Uwe; Donal, Erwan; 33422667Background: Despite all having systolic heart failure and broad QRS intervals, patients screened for cardiac resynchronization therapy (CRT) are highly heterogeneous, and it remains extremely challenging to predict the impact of CRT devices on left ventricular function and outcomes. The aim of this study was to evaluate the relative impact of clinical, electrocardiographic, and echocardiographic data on the left ventricular remodeling and prognosis of CRT candidates by the application of machine learning approaches. Methods: One hundred ninety-three patients with systolic heart failure receiving CRT according to current recommendations were prospectively included in this multicenter study. A combination of the Boruta algorithm and random forest methods was used to identify features predicting both CRT volumetric response and prognosis. Model performance was tested using the area under the receiver operating characteristic curve. The k-medoid method was also applied to identify clusters of phenotypically similar patients. Results: From 28 clinical, electrocardiographic, and echocardiographic variables, 16 features were predictive of CRT response, and 11 features were predictive of prognosis. Among the predictors of CRT response, eight variables (50%) pertained to right ventricular size or function. Tricuspid annular plane systolic excursion was the main feature associated with prognosis. The selected features were associated with particularly good prediction of both CRT response (area under the curve, 0.81; 95% CI, 0.74-0.87) and outcomes (area under the curve, 0.84; 95% CI, 0.75-0.93). An unsupervised machine learning approach allowed the identification of two phenogroups of patients who differed significantly in clinical variables and parameters of biventricular size and right ventricular function. The two phenogroups had significantly different prognosis (hazard ratio, 4.70; 95% CI, 2.1-10.0; P < .0001; log-rank P < .0001). Conclusions: Machine learning can reliably identify clinical and echocardiographic features associated with CRT response and prognosis. The evaluation of both right ventricular size and functional parameters has pivotal importance for the risk stratification of CRT candidates and should be systematically performed in patients undergoing CRT. (J Am Soc Echocardiogr 2021;34:494-502.)Item Characterization of Responder Profiles for Cardiac Resynchronization Therapy through Unsupervised Clustering of Clinical and Strain Data(2021) Gallard, Alban; Bidaut, Auriane; Hubert, Arnaud; Sade, Elif; Marechaux, Sylvestre; Sitges, Martha; Separovic-Hanzevacki, Jadranka; Le Rolle, Virginie; Galli, Elena; Hernandez, Alfredo; 33524492Background: The mechanisms of improvement of left ventricular (LV) function with cardiac resynchronization therapy (CRT) are not yet elucidated. The aim of this study was to characterize CRT responder profiles through clustering analysis, on the basis of clinical and echocardiographic preimplantation data, integrating automatic quantification of longitudinal strain signals. Methods: This was a multicenter observational study of 250 patients with chronic heart failure evaluated before CRT device implantation and followed up to 4 years. Clinical, electrocardiographic, and echocardiographic data were collected. Regional longitudinal strain signals were also analyzed with custom-made algorithms in addition to existing approaches, including myocardial work indices. Response was defined as a decrease of $15% in LV end-systolic volume. Death and hospitalization for heart failure at 4 years were considered adverse events. Seventy features were analyzed using a clustering approach (k-means clustering). Results: Five clusters were identified, with response rates between 50% in cluster 1 and 92.7% in cluster 5. These five clusters differed mainly by the characteristics of LV mechanics, evaluated using strain integrals. There was a significant difference in event-free survival at 4 years between cluster 1 and the other clusters. The quantitative analysis of strain curves, especially in the lateral wall, was more discriminative than apical rocking, septal flash, or myocardial work in most phenogroups. Conclusions: Five clusters are described, defining groups of below-average to excellent responders to CRT. These clusters demonstrate the complexity of LV mechanics and prediction of response to CRT. Automatic quantitative analysis of longitudinal strain curves appears to be a promising tool to improve the understanding of LV mechanics, patient characterization, and selection for CRT.Item A Novel Deep Learning Algorithm for the Automatic Detection of High-Grade Gliomas on T2-Weighted Magnetic Resonance I mages: A Preliminary Machine Learning Study(2020) Atici, Mehmet Ali; Sagiroglu, Seref; Celtikci, Pinar; Ucar, Murat; Borcek, Alp Ozgun; Emmez, Hakan; Celtikci, Emrah; 0000-0002-1655-6957; 31608975AIM: To propose a convolutional neural network (CNN) for the automatic detection of high-grade gliomas (HGGs) on T2-weighted magnetic resonance imaging (MRI) scans. MATERIAL and METHODS: A total of 3580 images obtained from 179 individuals were used for training and validation. After random rotation and vertical flip, training data was augmented by factor of 10 in each iteration. In order to increase data processing time, every single image converted into a Jpeg image which has a resolution of 320x320. Accuracy, precision and recall rates were calculated after training of the algorithm. RESULTS: Following training, CNN achieved acceptable performance ratios of 0.854 to 0.944 for accuracy, 0.812 to 0.980 for precision and 0.738 to 0.907 for recall. Also, CNN was able to detect HGG cases even though there is no apparent mass lesion in the given image. CONCLUSION: Our preliminary findings demonstrate; currently proposed CNN model achieves acceptable performance results for the automatic detection of HGGs on T2-weighted images.Item Diagnosis of Attention Deficit Hyperactivity Disorder with combined time and frequency features(2020) Altinkaynak, Miray; Dolu, Nazan; Guven, Aysegul; Pektas, Ferhat; Ozmen, Sevgi; Demirci, Esra; Izzetoglu, Meltem; 0000-0002-3104-7587; AAG-4494-2019The aim of this study was to build a machine learning model to discriminate Attention Deficit Hyperactivity Disorder (ADHD) patients and healthy controls using information from both time and frequency analysis of Event Related Potentials (ERP) obtained from Electroencephalography (EEG) signals while participants performed an auditory oddball task. The study included 23 unmedicated ADHD patients and 23 healthy controls. The EEG signal was analyzed in time domain by nonlinear brain dynamics and morphological features, and in time-frequency domain with wavelet coefficients. Selected features were applied to various machine learning techniques including; Multilayer Perceptron, Naive Bayes, Support Vector Machines, k-nearest neighbor, Adaptive Boosting, Logistic Regression and Random Forest to classify ADHD patients and healthy controls. Longer P300 latencies and smaller P300 amplitudes were observed in ADHD patients relative to controls. In fractal dimension calculation relative to the control group, the ADHD group demonstrated reduced complexity. In addition, certain wavelet coefficients provided significantly different values in both groups. Combining these extracted features, our results indicated that Multilayer Perceptron method provided the best classification with an accuracy rate of 91.3% and a high level of reliability of concurrence (Kappa = 0.82). The results showed that combining time and frequency domain features can be a useful and discriminative for diagnostic purposes in ADHD. The study presents a supporting diagnostic tool that uses EEG signal processing and machine learning algorithms. The findings would be helpful in the objective diagnosis of ADHD. (C) 2020 Nalecz Institute of Biocybernetics and Biomedical Engineering of the Polish Academy of Sciences. Published by Elsevier B.V. All rights reserved.Item Saldırı tespit sistemlerinde genetik algoritma kullanarak nitelik seçimi ve çoklu sınıflandırıcı füzyonu(2018) Özgür, A.; Erdem, H.With the improvements in information systems, intrusion detection systems (IDS) become more important. IDS can be thought as a classification problem. An important step of classification applications is feature selection step. Nowadays, to improve accuracy of classifiers, it is recommended to use classifier fusion instead of single classifiers. This study proposes to use genetic algorithms for both feature selection and weight selection for classifier fusion in IDS. This proposed system called GA-NS-AB, has been applied to NSL-KDD dataset. Number of classifiers used in fusion changes between 2 and 8. Following classifiers have been used: Adaboost, Decision Tree, Logistic Regression, Naive Bayes, Random Forests, Gradient Boosting, K-Nearest Neighbor, and Neural Networks Multi-Layer Perceptron. The results of the proposed method have been compared with simple voting and probability voting fusion methods and single classifiers. In addition, GA-NS-AB is also compared with previous results. GA-NS-AB is a high accuracy classifier fusion that reduces test and training time. © 2018 Gazi Universitesi Muhendislik-Mimarlik. All rights reserved.