Search

Recently, hospitals and health care institutions have increasingly been addressing clinical decision support systems (CDSS), which can offer specific patient assessments or recommendations to physicians and health care professionals. It is very useful to develop CDSS which can help physicians to make meaningful and correct decisions by using existing data or image sets. Also, CDSS increases the diagnostic accuracy of diseases, provides significant facilities in precision medicine applications, increases operating efficiency of hospitals and reduces costs. In this context, the proposed project intends to create a model using pre-trained networks (i.e. VGG-16,) based on deep learning (DL) that can successfully predict the melanoma using dermoscopic images. The current study provides clinical support to physicians in the medical decision-making process for the diagnosis of melanoma.

Association rules mining is one of the uttermost applied techniques in data mining and artificial intelligence. Support and confidence are two basic measures employed in the evaluation of association rules. The rules obtained with these two values are often correct; however, they are not strong rules. Most of the rules, especially with a high support value, are misleading. For this reason, there are many interestingness measures proposed to achieve stronger rules. In this study it is aimed to establish strong association rules with variables in open sourced heart data set. In the current study, Apriori algorithm was used to obtain the rules. As a result of the analysis, only 55 confidence and support criteria were taken into consideration. For more powerful rules, certainty factor was used as one of the interestingness measure proposed in the literature, and it was concluded that only 26 of these rules were strong. As a result of the analysis of the findings obtained in the context of the research, it can be inferred that stronger rules can be obtained by using the certainty factor in association rules mining.

In this study, it is aimed to develop a new user-friendly web-based software to ensure the normal distribution of the new data set obtained by applying various mathematical transformations to data sets that do not meet the normal distribution from parametric hypothesis testing conditions. The web based software is developed by Shiny web framework, which is an open source R package. In the web-based software, arcsinh (x), Box-Cox, Exponential, Lambert W (h type), Lambert W (s type), Logarithmic, Square root, Yeo-Johnson data transformation methods are used. The software also allows us to decide on the best conversion for the data set to be transformed by calculating the Pearson P test statistic, which chooses on the best method in the methods used. This is done by selecting the conversion with the smallest Pearson P statistic. In order to evaluate the outputs of the software, a data set with three variables was derived with the Cauchy, gamma, Weibull distributions respectively and 500 samples were used for each distribution. The developed software is novel user-friendly web-based software that enables researchers to get more accurate results in their work and to obtain data transformation methods that provide parametric test assumptions known as more powerful tests.

In this study, it is aimed to develop a user-friendly open-source web-based software which enables the assignment of the subjects included in the scientific research to the groups with equal probability. An open source R package, Shiny, is used to develop the recommended web tool. In the developed software, one of the required equilibrium methods; random allocation rule, truncated binomial design, maximal procedure design, complete randomization methods; complete randomization design, blocking methods; permuted block randomization with random block constellation, the Hadamard randomization, adaptive methods; the big stick design, Efron's biased coin design, Wei's urn design, generalized biased coin design, Chen's biased coin design are included. A random allocation rule, one of the random assignment methods, is applied to a hypothetical data set where the sample size is 140 and the number of groups are two. As a result, in the first group, a random assignment was made in such a way that the number of samples are 70 and the number of samples in the second group are 70. According to the hypothetical data set findings, the developed easily assigns the subjects to the study groups by using random assignment methods. Therefore, it is stated that it easily solved a significant bias problem in scientific studies. In the following stages of the study, the scope of the software can be expanded with the addition of techniques comparing the results of random assignment methods.

This study intends to develop an easy-to-use software using deep learning algorithms classification of benign and malignant lung nodules with high accuracy. In modelling phase Keras library and convolutional neural networks model are used. To construct user interface, shiny web framework is utilized. Experimental findings showed that the convolutional neural network model has high classification performance metrics of benign and malignant lung nodules on computed tomography images. As future research, it will be aimed to develop new software for medical images of other types of definitely diagnosed diseases by using appropriate deep learning models. The developed software is available at: biostatapps.inonu.edu.tr/AKSY/

Deep learning is much preferred in image processing applications since it can give fast and important results. This research aims at developing an open source software for deep learning based-classification of dementia in magnetic resonance imaging scans. Keras (i.e., a deep-learning framework) is employed for constructing a deep learning based-model that could be discriminate between dementia patients and healthy individuals. The achieved findings demonstrate that the proposed system can be used to detect individuals with suspected dementia disease.

There are many assumptions that should be provided in regression and correlation analyses. If the assumptions are not met, the results of the analysis lead to errors. Multivariate regression analyses are performed by many software. However, many of this software are commercial and platform dependent. In this study, open source web-based software is developed to test the assumptions of simple/multiple linear regression, simple/multiple logistic regression analysis and correlation analysis, which are classification and regression techniques in machine learning. This software also provides a statistical interpretation of the results to researchers.