Μελετη Και Αναπτυξη Αλγοριθμων Για Την Αναλυση Βρογχοσκοπικων Εικονων Σε Πραγματικο Χρονο.

Abstract

Autofluorescence (AF) bronchoscopy is an established method to detect dysplasia and carcinoma in situ (CIS). For this reason "SOTIRIA" Hospital in Athens, Greece uses the Karl Storz D-light system visualizing malignant and pre-malignant changes as early as possible. This system induces autofluorescence in bronchial submucosa. The thickened epithelium near the tumorous changes causes the light to be dimmer there than in healthy regions of the bronchial mucosa. However, in early tumor stages and clinically difficult cases the visualization is not that obvious even with the previous system. With the help of a PC Intel P-4, using S-Video cable to connect it with the bronchoscope and a 128 MB DDR ATI video graphics card with frame grabber software, we analyze the color images we capture and segment the suspicious regions of early cancer. Previously, Nikos Apostolou [1] has used statistical methods based on co-occurrence matrices, signal processing methods based on Gabor models and conversion algorithms between device dependent color spaces. In this paper further investigation and research were done to the direction of detecting suspicious regions of early cancer in bronchoscopic images. Our job was to develop Matlab® functions that detect malicious regions (CIS) in either autofluorescence images (blue light) or traditional bronchoscopic images (white light). For the blue-light images we used SOBEL filters, unsharpness filters , the newly developed SUSAN algorithm and the very promising Log Gabor filter bank. For the white-light images we implemented methods for converting them to different colorspaces(HIS and YP and and applying Gabor filters.. We believe that we reduce the error made by the naked eye to a significant percentage. The platform we implement is not only for the clinical work but also for the educational one related with the doctors in pneumonology speciality. Healthcare providers may include such a platform as software with the bronchoscope devices in sale meeting the up-to date clinical needs and improving the quality of life for patients with pneumonological pathology.