Automatic regions detection in CT images based on Haralick textures

The identification of anomalies (such as bone fractures or tendonitis in muscles and soft tissues) through image processing and analysis techniques in Computed Tomography (CT) images is today of great importance to assist doctors and health professionals in making accurate diagnoses. The extraction of relevant information from the CT image is characterized by the calculation of gray level input image attributes. Statistical moments (SM) are calculated using the gray level distribution of an image and are therefore generally calculated from that image's histogram. These characteristics provide a statistical description of the relationship between different gray levels in the CT image. Haralick proposed a methodology for describing textures based on second order statistics, where characteristics are derived from co-occurrence matrices, which are constructed by counting different combinations of gray levels in an image according to certain directions. In this work, it is intended to automatically identify and extract regions in CT images based on textures as an aid for a quick and accurate diagnosis. CT images are first pre-processed for noise reduction and image enhancement, followed by the application of Haralick textures to segment and detect zones of interest. Classifiers trained on the Haralick invariant features showed good accuracy and performance. Despite the presence of low contrast and noise in some images, the proposed algorithms present promising results in the segmentation and automatic identification of regions of tomographic images, being an important contribution to support health professionals in the characterization of anomalies and their extension. Good results are expected for the next step of this work in the detection and segmentation of anomalies in CT images.