Browsing by Author "Yildirim, Pinar"

Now showing 1 - 7 of 7

Citation Count: 18
Chronic Kidney Disease Prediction on Imbalanced Data by Multilayer Perceptron
(Ieee, 2017) Yıldırım, Pınar; Bilgisayar Mühendisliği / Computer Engineering
Imbalanced data is an important problem for medical data analysis. Medical datasets are often not balanced in their class labels. The traditional classifiers can be seriously affected by the imbalanced class distribution in the data. This is because they aim to optimize the overall accuracy without considering the relative distribution of each class. This study searches the effect of class imbalance in training data when developing neural network classifier for medical decision making on chronic kidney disease. Neural networks are widely used in a number of applications including data mining and decision systems. Back propagation networks are a popular type of neural networks that can be trained to recognize different patterns. The importance of these networks was considered and a comparative study of some sampling algorithms was performed based on multilayer perceptron with different learning rate values for the prediction of chronic kidney disease. This study reveals that sampling algorithms can improve the performance of classification algorithms and learning rate is a crucial parameter which can significantly effect on multilayer perceptron.
Citation Count: 5
Clustering Inflammatory Markers with Sociodemographic and Clinical Characteristics of Patients with Diabetes Type 2 Can Support Family Physicians' Clinical Reasoning by Reducing Patients' Complexity
(Mdpi, 2021) Yıldırım, Pınar; Yildirim, Pinar; Babic, Frantisek; Sahinovic, Ines; Wittlinger, Thomas; Martinovic, Ivo; Majnaric, Ljiljana Trtica; Bilgisayar Mühendisliği / Computer Engineering
Diabetes mellitus type 2 (DM2) is a complex disease associated with chronic inflammation, end-organ damage, and multiple comorbidities. Initiatives are emerging for a more personalized approach in managing DM2 patients. We hypothesized that by clustering inflammatory markers with variables indicating the sociodemographic and clinical contexts of patients with DM2, we could gain insights into the hidden phenotypes and the underlying pathophysiological backgrounds thereof. We applied the k-means algorithm and a total of 30 variables in a group of 174 primary care (PC) patients with DM2 aged 50 years and above and of both genders. We included some emerging markers of inflammation, specifically, neutrophil-to-lymphocyte ratio (NLR) and the cytokines IL-17A and IL-37. Multiple regression models were used to assess associations of inflammatory markers with other variables. Overall, we observed that the cytokines were more variable than the marker NLR. The set of inflammatory markers was needed to indicate the capacity of patients in the clusters for inflammatory cell recruitment from the circulation to the tissues, and subsequently for the progression of end-organ damage and vascular complications. The hypothalamus-pituitary-thyroid hormonal axis, in addition to the cytokine IL-37, may have a suppressive, inflammation-regulatory role. These results can help PC physicians with their clinical reasoning by reducing the complexity of diabetic patients.
Citation Count: 0
Clustering of Phentermine HCL Drug from Online Patient Medication Reviews
(Elsevier, 2019) Yıldırım, Pınar; Kaya, Alkan; Kaya, Ayşe Demet; Bilgisayar Mühendisliği / Computer Engineering; Tıbbi Mikrobiyoloji / Medical Microbiology
In this paper, a study to reveal hidden knowledge in the online patient medication reviews for phentermine HCL is presented. Phentermine HCL is used most frequently in the treatment of obesity. Obesity is a complex health disorder that affects huge amount of people. In recent years, the number of overweight people in industrialized countries has increased significantly and people who are obese are at a much higher risk for serious medical conditions such as high blood pressure, heart attack, stroke and diabetes. Considering the importance of the medication of obesity, knowledge discovery from online patient reviews is performed. Some information technologies and data mining techniques are used to discover some hidden knowledge between patient information and side effects in these reviews. Our results can give new ideas to medical researchers and pharmaceutical industry for drug safety. (C) 2019 The Authors. Published by Elsevier B.V. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/) Peer-review under responsibility of the Conference Program Chairs.
Citation Count: 0
Discovery of the Similarities for Parasites
(Ieee, 2020) Yıldırım, Pınar; Ceken, Kagan; Bilgisayar Mühendisliği / Computer Engineering
In this paper we report on a study for discovering hidden patterns in commonly seen parasites by using abstracts from MEDLINE database. Parasites affect millions of people in the world and cause tremendous morbidity and mortality. Diagnosing parasites can be difficult because some symptoms and related to gene-proteins can be common to some of them. We utilize a web based biomedical text mining tool to find symptoms and gene-proteins. After selecting the most common symptoms and gene-proteins, we create two datasets with the frequencies of symptoms and gene-proteins for each parasite. For this work we selected the k-means algorithm for clustering analysis and apply it on the datasets. In addition, we compared different algorithms to observe the performance of k-means. Clustering analysis generated different types of groups of parasites. Although the results are not 100% certain, they can make positive contributions to medical researchers and experts for the diagnosis of parasites.
Citation Count: 11
Disease-disease relationships for rheumatic diseases Web-based biomedical textmining and knowledge discovery to assist medical decision making
(Ieee, 2012) Yıldırım, Pınar; Simonic, Klaus-Martin; Yildirim, Pinar; Bilgisayar Mühendisliği / Computer Engineering
The MEDLINE database (Medical Literature Analysis and Retrieval System Online) contains an enormously increasing volume of biomedical articles. There is urgent need for techniques which enable the discovery, the extraction, the integration and the use of hidden knowledge in those articles. Text mining aims at developing technologies to help cope with the interpretation of these large volumes of publications. Co-occurrence analysis is a technique applied in text mining and the methodologies and statistical models are used to evaluate the significance of the relationship between entities such as disease names, drug names, and keywords in titles, abstracts or even entire publications. In this paper we present a method and an evaluation on knowledge discovery of disease-disease relationships for rheumatic diseases. This has huge medical relevance, since rheumatic diseases affect hundreds of millions of people worldwide and lead to substantial loss of functioning and mobility. In this study, we interviewed medical experts and searched the ACR (American College of Rheumatology) web site in order to select the most observed rheumatic diseases to explore disease-disease relationships. We used a web based text-mining tool to find disease names and their co-occurrence frequencies in MEDLINE articles for each disease. After finding disease names and frequencies, we normalized the names by interviewing medical experts and by utilizing biomedical resources. Frequencies are normally a good indicator of the relevance of a concept but they tend to overestimate the importance of common concepts. We also used Pointwise Mutual Information (PMI) measure to discover the strength of a relationship. PMI provides an indication of how more often the query and concept co-occur than expected by change. After finding PMI values for each disease, we ranked these values and frequencies together. The results reveal hidden knowledge in articles regarding rheumatic diseases indexed by MEDLINE, thereby exposing relationships that can provide important additional information for medical experts and researchers for medical decision-making.
Citation Count: 14
Knowledge discovery of drug data on the example of adverse reaction prediction
(Bmc, 2014) Yıldırım, Pınar; Majnaric, Ljiljana; Ekmekci, Ozgur Ilyas; Holzinger, Andreas; Bilgisayar Mühendisliği / Computer Engineering
Background: Antibiotics are the widely prescribed drugs for children and most likely to be related with adverse reactions. Record on adverse reactions and allergies from antibiotics considerably affect the prescription choices. We consider this a biomedical decision-making problem and explore hidden knowledge in survey results on data extracted from a big data pool of health records of children, from the Health Center of Osijek, Eastern Croatia. Results: We applied and evaluated a k-means algorithm to the dataset to generate some clusters which have similar features. Our results highlight that some type of antibiotics form different clusters, which insight is most helpful for the clinician to support better decision-making. Conclusions: Medical professionals can investigate the clusters which our study revealed, thus gaining useful knowledge and insight into this data for their clinical studies.
Citation Count: 12
Pattern classification with imbalanced and multiclass data for the prediction of albendazole adverse event outcomes
(Elsevier Science Bv, 2016) Yıldırım, Pınar; Bilgisayar Mühendisliği / Computer Engineering
Class imbalance problem is one of the important problems for classification studies in data mining. In this study, a comparative analysis of some sampling methods was performed based on the evaluation of four classification algorithms for the prediction of albendazole adverse events outcomes. Albendazole is one of the main medications used for the treatment of a variety of parasitic worm infestations. The dataset was created from the public release of the FDA's FAERS database. Four sampling algorithms were used to analyze the dataset and their performance was evaluated by using four classifiers. Among the algorithms, ID3 with resample algorithm has higher accuracy results than the others after the application of sampling methods. This study supported that sampling methods are capable to improve the performance of learning algorithms. (C) 2016 The Authors. Published by Elsevier B.V.