Extended development of a Computer Aided Detection (CAD) system for brain bleed in CT / Muhammad Illyas Abdul Muhji

• Main Author: Muhammad Illyas, Abdul Muhji
• Format: Thesis
• Published: 2018
• Subjects: R Medicine (General)
• Online Access: http://studentsrepo.um.edu.my/11558/4/illyas.pdf; http://studentsrepo.um.edu.my/11558/

Computer aided detection Computer aided detection (CAD) is a tool developed to assist radiologist interpretations from diagnostic modalities to decrease observational oversights or false negative rates. With CAD, radiologists able to use the computer output as second opinion, where the final decisions is still made by a human. One possible CAD usage is the detection of brain haemorrhage or in general terms brain bleed make it potentially useful in detection of brain bleed bleeds as a first-line screening tool, particular in emergency cases which occur outside regular working hours. Previous study by Leong show the useful of CAD system but her study is not fully automated. Thus the main objective of this study is to implement an automatic algorithm to previous algorithm to make it fully automatic system. In this study, 227 volumes of brain CT images were used to develop and validate the CAD. The new develop algorithm is set to register image of brain patient in order to determine the rotation angle for brain realignment. The new rotation angle obtained is compare with previous study to evaluate the differences. Then the new rotation angle is used to detect bleeding in the patient and evaluate the performance of the new algorithm. The algorithm for bleed detection consist of image processing to separate brain from the skull, rotation and realignment of the brain for mid-sagittal plane determination which used in bleeding detection. Final output of the algorithm summarise the bleeding detection for all patient in an excel file. The result obtained in this study show statistically difference in rotation angle between the new and previous study. Overall performance for new algorithm gives a sensitivity, specificity and accuracy for training 74%, 68.1% and 70.6% respectively and 84.6%, 67.1% and 73.4% respectively for the validation set. To be concluded, the new algorithm decreased the performance of the previous CAD system slightly but for a transition from manual to fully automatic, the iv decrease is not very big. Having said that the new algorithm still needs improvements so that the CAD able to perform better.