Image processing techniques for brain haemorrhage detection in head CT scans

Abstract

Medical imaging is an important tool used for obtaining visual information of the interior of the body. There exist various imaging modalities such as Magnetic Resource Imaging, Ultrasound and Computed Tomography (CT). The latter is used extensively for detection and diagnosis of brain haemorrhage. Computer Aided Detection/Diagnosis (CAD) systems are used by radiologists as a tool which helps them during the diagnosis phase. Increasingly, CAD is becoming a key component of routine clinical practice in several medical areas such as mammography and colonoscopy. Research regarding brain CAD systems has not progressed at the same pace as research in the mentioned areas. This provides a need as well as an opportunity to contribute to the research by developing a CAD system for brain haemorrhage detection and classification. This study applies image processing techniques to brain CT scans with the aim of creating a CAD system which detects fresh bleeds. The system also includes a basic classification that identifies the haemorrhage type. The system distinguishes between an intra-axial haemorrhage and an extra-axial haemorrhage with the only limitation being subarachnoid haemorrhage (SAH), which is not always properly classified due to its complex structure. The techniques implemented in this study include noise reduction methods through the use of filters, morphological operations and segmentation algorithms, such as thresholding and clustering. The designed CAD system was tested on 36 brain CT sets obtained from the general hospital in Malta. The results show that the system achieved a sensitivity of 94.4%, a specificity of 94.4%, a precision of 91.259% and a classification accuracy of 88.89%. The system performed the required operations in an average time of 0.26 seconds per slice.