Data analysis in MR images and imputing missing data: An application for dementia diseases

Within the scope of this doctoral thesis, strategic data science approaches such as processing incomplete data, feature analysis, evaluation of valuable or ineffective features, and imputation of missing data, which are essential problems in computer science, are discussed through the application of dementia diseases. Distorted medical images or other missing information in the neuroimaging field can lead to erroneous decisions in the disease diagnosis process. Therefore, analyzing and preprocessing the features, the difficulties brought to data science by missing data, especially clinical test scores, and scale studies that indirectly question the reliability of the findings become important in the decision support phase. In this doctoral thesis, sliced brain scans evaluated during clinical tests were first processed with auxiliary software tools and virtual models were generated. The features were extracted from the scans via the proposed "parallel three-dimensional brain modeling and feature extraction algorithm" at a 95.57 speedup factor. Manipulation of extracted features was facilitated through the proposed "FreeSurfer: data preprocessing and manipulation software tool." Effective features that detailly preprocessed and achieved up to 76.5% performance in 4-class classification were analyzed. The critical missing values on the data were completed with low error rates over different error metrics with the proposed "dementia-related user-based collaborative filtering" method, and a "reliability scale" that medical doctors can use for decision support during clinical tests was defined at a 95% confidence level.