One of the serious public health problems is the Traumatic Brain Injury, also known as silent epidemic, affecting millions every year. Management of these patients essentially involves neuroimaging and noncontrast CT scans are the first choice amongst doctors. Significant anatomical changes identified on the neuroimages and volumetric assessment of haemorrhages and haematomas are of critical importance for assessing the patients’ condition for targeted therapeutic and/or surgical interventions. Manual demarcation and annotation by experts is still considered gold standard, however, the interpretation of neuroimages is fraught with inter-observer variability and is considered ’Achilles heel’ amongst radiologists. Errors and variability can be attributed to factors such as poor perception, inaccurate deduction, incomplete knowledge or the quality of the image and only a third of doctors confidently report the findings. The applicability of computer aided dianosis in segmenting the apposite regions and giving ’second opinion’ has been positively appraised to assist the radiologists, however, results of the approaches vary due to parameters of algorithms and manual intervention required from doctors and this presents a gap for automated segmentation and estimation of measurements of noncontrast brain CT scans. The Pattern Driven, Content Aware Active Contours (PDCAAC) Framework developed in this thesis provides robust and efficient segmentation of significant anatomical landmarks, estimations of their sizes and correlation to CT rating to assist the radiologists in establishing the diagnosis and prognosis more confidently. The integration of clinical profile of the patient into image segmentation algorithms has significantly improved their performance by highlighting characteristics of the region of interest. The modified active contour method in the PDCAAC framework achieves Jaccard Similarity Index (JI) of 0.87, which is a significant improvement over the existing methods of active contours achieving JI of 0.807 with Simple Linear Iterative Clustering and Distance Regularized Level Set Evolution. The Intraclass Correlation Coefficient of intracranial measurements is >0.97 compared with radiologists. Automatic seeding of the initial seed curve within the region of interest is incorporated into the method which is a novel approach and alleviates limitation of existing methods. The proposed PDCAAC framework can be construed as a contribution towards research to formulate correlations between image features and clinical variables encompassing normal development, ageing, pathological and traumatic cases propitious to improve management of such patients. Establishing prognosis usually entails survival but the focus can also be extended to functional outcomes, residual disability and quality of life issues.
| Identifer | oai:union.ndltd.org:bl.uk/oai:ethos.bl.uk:681067 |
| Date | January 2015 |
| Creators | Qureshi, Adnan Nabeel Abid |
| Publisher | University of Bedfordshire |
| Source Sets | Ethos UK |
| Detected Language | English |
| Type | Electronic Thesis or Dissertation |
| Source | http://hdl.handle.net/10547/601083 |
Page generated in 0.002 seconds