Automated Tool for the Extraction of Healthy Sinus Area

Sinus diagnosis has proven to be a difficult task due to the complexity of the image obtained using the computed Tomography (CT). This makes image processing and analysis difficult and time consuming. The aim of this paper is to investigate and implement a technique for the extraction of healthy maxillary and ethmoid sinuses. The proposed algorithm involves the use of neural network (self-organising map) for image processing. The self organising map (SOM) was trained using sample image which is a cut out from the CT-image of an anonymous patient. SOM learns to classify input vector according to how they are grouped. This algorithm works by first automatically reducing the input image which is usually (1012x938) pixels to (452 x 500). The next step involves reassigning all pixels in the input image with the trained SOM thereby automatically thresholding the image. Next the resulting image is inverted to make the region of interest visible against a black background and using image processing technique it crops the unwanted background and finally displays the region of interest (ROI). Data used were supplied by the John Dalton research group of the Manchester Metropolitan University. The CT scans were of anonymous patient from Trafford general hospital. All analysis was done under the Matlab environment.. The technology is based on the traditional image processing technique and Artificial Neural Network (ANNs) self organising map (SOM). The result obtained shows that the proposed algorithm is promising in the sinus extraction with the possibility to extend to the diagnosis of abnormalities. Further work would be to extend the extraction system to cover all sinuses (Maxillary, Frontal,Ethmoid, Sphenoid) and also for diagnosis of abnormalities in the sinus. In future the system would also find its usefulness in the area of training junior doctors.