Research On Lesion Detecting Based On Gastroscope Image

Gastroscope technique has been widely used in the detection of gastric disease.However, identifying gastric disease from a large number of gastroscope images by nakedeye is time-consuming. To solve this problem, this paper proposes two computer-aideddetection methods based on two different ideas.This paper puts forward a lesion detection method of electronic gastroscope imagesbased on superpixel segmentation. This method applies image segmentation algorithmbased on the theory of superpixels to overcome the trait of outline dims and serious noisepollution in gastroscope images, and realizes the image segmentation with speedy andaccurate. The partitioned regions of uniform size and single pixels lay the foundation forfeature extraction and classification. Though the method improves the recognition rate ofsingle region, the method has not considered the region correlation. For this reason, thispaper proposes another lesion detection method based on Markov Random Field(MRF).This method applies the theory of MRF to analysis the relationship of each region, makesthe detection process no longer isolated. The detection process of each region has strongcoupling with the application of MRF, so that the identification accuracy is increasedfurther.In terms of lesion detection accuracy, lesion detection method based on superpixelsegmentation accuracy (AUC) can reach0.91588, and in the optimal conditions, thesensitivity is91.88%, specificity is80.36%. The lesion detection method based on MRFperforms well with the sensitivity of93.15%and specificity of78.52%. Although the latterhas a modest improvement in accuracy test method, but it needs longer testing time. Fromthe result of experiment, both reasonable image segmentation in the early detection and thecorrection of detection classifier score in the late detection can achieve a betterperformance than the existing detection methods.