Physicians evaluate computed tomography (CT) of the brain to quantitatively and qualitatively identify various types of intracranial hematomas for patients with neurological emergencies. We propose a novel method that can perform this task in a totally automatic fashion, based on a multiresolution binary level set method. The skull regions are segmented in downsized images generated with a maximum filter. The intracranial regions are located using the average gray levels and connectivity. These regions compose the regions of interest (ROIs) for segmenting the hematoma from the normal brain. The gray levels of the voxels within these ROIs are generated with an averaging filter in a multiresolution fashion. After identifying the candidate hematoma voxels using adaptive thresholds and connectivity, binary level set algorithm is applied repeatedly until the original resolution is reached. We apply our method to non-volumetric non-contrast CT images of 15 surgically proven intracranial hematomas and the results were quantitatively evaluated by a human expert. The correlation coefficient between the volumes measured manually and automatically is 0.97. The overlap metrics ranged from 0.97 to 0.74, with an average of 0.88. The average precision and recall are 0.89 and 0.87, respectively. We use decision rules to classify these hematomas and were able to make correct diagnoses in all cases.
ASJC Scopus subject areas
- Radiology Nuclear Medicine and imaging
- Health Informatics
- Radiological and Ultrasound Technology
- Computer Graphics and Computer-Aided Design
- Computer Vision and Pattern Recognition