The Feature-Based Microscopic Image Segmentation for Thyroid Tissue

Thyroid diseases are prevalent among endocrine diseases. The microscopic image of thyroid tissue is the necessary and important material for investigating thyroid functional mechanism and diseases. A computerized system has been developed in this study to characterize the textured image features of the microscopic image in typical thyroid tissues, and then the compositions in the heterogeneous thyroid tissue image were classified and quantified. Seven image features were implemented to characterize the histological structure representation for tissue types including blood cells, colloid, fibrosis tissue, follicular cell. The statistical discriminant analysis was implemented for classification to determine which features discriminate between two or more occurring classes (types of tissues). The microscopic image was divided to be contiguous grid images. The image features of each grid image were evaluated. Multiple discriminant analysis was used to classify each grid image into the appropriated tissue type and Markov random fields were then employed to modify the results. 100 random selected clinical image samples were employed in the training and testing procedures for the evaluation of system performance. The results show that accuracy of the system is about 96%.