Uma abordagem por level set usando estimação não paramétrica com base na janela de parzen para segmentação automática de regiões de interesse em imagens médicas.

Abstract

This thesis proposes a level set approach using a nonparametric estimation based on the parzen window to automatically segment regions of interest in medical images to aid medical diagnosis called the FLog Parzen Level Set (FPLS). The FPLS method combines the level set method using a nonparametric estimation approach based on parzen window and clustering with Floor of Log to segment the region of interest in medical images. In this thesis, the Floor of Log (FLog) intersection with PDI methods is responsible for initializing the seed point in the region of interest to determine the zero level set. This combination provides the system greater robustness in determining the zero level set that will increase the efficiency in segmenting the region of interest in the shortest possible time. As initialization becomes more stable, the growth and refinement of the region using probabilistic estimates with the parzen window within the level set produces a more accurate outcome. The proposed method was applied and evaluated together with others in the literature in three applications: segmentation of regions with stroke on CT images of the skull, segmentation of the lungs on CT images of the chest, and segmentation of skin lesions on digital images. The results demonstrate the efficiency of FPLS in systems that assist in medical diagnosis. The speed of the method proposed in the segmentation and processing of information for medical diagnosis is promising, presenting efficient and reliable results. It achieves comparable results and, often, is superior to other methods with low computational costs. Therefore, considering the general analysis of the results obtained by the proposed approach, we can conclude that the FPLS is promising in the segmentation of regions of interest. The method showed promise in analyzing healthy, unhealthy, and anomalies patients’ images. Therefore, the FPLS can be integrated into medical diagnosis aid systems, either for diagnosis or for the follow-up of diseases throughout treatment, in a more targeted way, for the analysis of stroke in CT images of the skull, of lungs on chest CT images, and of skin lesions on digital images.