Early diagnosis and classification system for brain masses

This study consists of 2 main sections. The first part is a biomedical computer-based pre-diagnosis system as a decision support system that helps physicians to automatically detect and interpret brain masses with specific localization and parameters (Glioblastoma, Gliom, Meningioma) with an interface-supported system. The first part is divided into 2 sub-parts within itself. It is the realization of the interpretation and estimation process made after mass detection with classical Artificial Intelligence methods and advanced Deep Learning methods, which have been very trendy in recent years. Secondly, with using Convolutional Neural Network (CNN), AlexNet and ResNet-50 models and the classification estimation process is performed with the latest version algorithm. The second part is a biomedical diagnosis and scoring system for early detection, scoring and interpretation of the condition of stroke which is included in the group of brain masses. From Magnetic Resonance (MRI) and Computed Tomography (CT) radiological images, after image processing, for the next step, following the feature extraction process, one of the classical Artifical Intelligence methods, Support Vector Machine, k-Neareast Neighbor and Adaboost methods, the pre-diagnosis prediction and interpretation steps of the patients' mass iamges are performed. Classification processes are performed with more advanced and trending methods without the need for feature extraciton with CNN, AlexNet and ResNet-50 models, which are among the Deep Learning models. The second part is the early detection and interpretation of stroke using the image processing and pattern recognition methods and a computer-based automated stroke scoring method. In the experimental stage, 300 patient images were used for each of Glioblastoma, Glioma and Meningioma brain masses, and 200 patient images were used for stroke diagnosis and scoring parts. The results obtained after ROC analysis and accuracy / performance test are 82% from the first part of the classification process with classical methods, 87% from the classification processes with AlexNet and %95 from ResNet-50 advanced models, respectively.