Reliability improvement of computer aided diagnostic system using mutual information

Computer aided diagnosis (CAD) is one of the most important topic in recent years since the systems are able to provide a second reliable opinion to physicians and early diagnosis with these systems are possible. In this study we aim to construct a system for the detection of Alzheimer's disease (AD) using PET images from a database. The CAD system includes a database consisting of a 3D PET image for every query. Via using a similarity metric namely mutual information(MI), every query compares to all other query in database. According to their similarity results, a decision index is calculated. The decision index demonstrate presence or absence of AD. The system was developed and evaluated using two different databases extracted from Alzheimer's disease Neuroimaging Initiative (ADNI) database. All normal and Alzheimer's images are stored and ordered in database. First database consists of 259 normal and 138 AD patient whereas second database consists of 102 normal and 95 AD patient. Main difference of two database is registration. Images in second database are warped with talairach atlas. CAD performance was evaluated using Receiver Operating Characteristic analysis. For every query, a decision index was calculated. According to our results we observed that accuracy and speed of the CAD system is affected by certain parameters. The method proposed in the article is adequate to distinguish the disease. The mutual information method is very simple, applicable and fast enough to use in clinic area.