Cerebral hemorrhage segmentation with energy functional based on anatomy theory

Abstract

In the computer-aided diagnosis of brain injury, the segmentation of lesions can provide help for subsequent diagnosis and treatment. For many CT images of cerebral hemorrhage, there are irregular blood clots, whose spatial location, shape, and area are changeable. Besides, the blood clots are sometimes connected with the skull (called a cranial connection), which can add to the difficulty and complexity of segmentation. In this paper, the anatomy and imaging features are used to analyze blood clots at cranial connection and disconnection in CT images. Then, a novel energy functional based on the imaging and anatomy features is proposed, which combines shape and area constraints of a cerebral hemorrhage. The proposed method is verified on the actual CT images of 42 patients with cerebral hemorrhage in the Second Affiliated Hospital of Dalian Medical University, which obtains good segmentation results. It is proved that the proposed method has good performance under the evaluation indicators of Dice coefficient, Jaccard coefficient, and recall.