Finding multimodality pathology connection of molecular structures using nano-biomechanical breast images

Finding Multimodality Pathology Connection of Molecular Structures Using Nano-Biomechanical Breast Images In this thesis, it is purposed to make stronger early diagnosis in breast cancer by evaluating the mammography, breast cell histopathology and nanobiomechanic images taken from Atomic Force Microscopy together. Afterwards, the specifications of 23 Gray- Level Syngenetic Matrix (GDEOM) are used for these 3 types of images. These specifications are measured in the different angles (θ=0°,45°,90°, ve 135). For obtaining optimum specification from these specifications, Minimum Redundancy/ Maximum Relation (mFMİ) and Principal Component Analysis (TBA) are used. As classifier, K-Nearest Neighbor (KEYK),Least Squares Support Vector System (EKKDVS) and new developed Arithmetic Average- Standard Deviation-Coefficient of a variance- Standard Error Maximum/Sort Minimum (ASMSM) algorithm is used. For the early diagnosis of the breast cancer, methods are developed. Total 540 images are used for the methods in this thesis. By using developed methods and by combining these images with mFMİ_KEYK method and mammography, histopathology, for AKM images by combining mammography-histopathology images respectively %100, %100, %100 and %100 accuracy diagnosis rates are found. With the method of mFMİ_EKKDVS, respectively %100, %100, %100 and %100; with mFMİ_ASMSM method, respectively %76.67, %82.22, %92.22 and %100; with mFMİ_ASMSM method %76.67, %82.22, %92.22 and %100 ; with TBA_KEYK method respectively %76.67, %71.11, %75.56 and %85.56; with TBA_EKKDVS method respectively %92.22, %97.78, %92.22 and %96.56; with TBA_ASMSM method respectively %95.56, %92.22, %94.44 and %100 accuracy diagnosis rates are achieved. Besides, by benefitting from the method of Area Under the Roc Curve (REAKA) which shows the interclass connections and used in the multiple classification, accurate positive rates are found between two classes. In addition, as the result of the analyses made for AKM images, it is observed that malignant AKM images are bigger than the benign AKM images with the normal surface roughness and its particle volume is smaller. Kew Words : Mammography, Histopathology, Nanobiomechanic Images, Minimum Redundancy /Maximum Relation, Principal Component Analysis, Nearest Neighbor, Least Squares Support Vector System