Recognition Of Lung Nodules On CT Scans

Computer-Aided Diagnosis is proposed for help radiologists increaseefficiency and accuracy when finding focuses by combining iconography,image processing, computer analysis, modeling and pattern recognition.This paper describes a Computer-Aided Diagnosis (CAD) system forautomatic pulmonary nodules detection on serial CT scans based onshape features. It also proposes two novel recognition methods which aremoment invariant features and3D thinning. The system recognizesnodules by3D geometric information through the process of interpolation,segmentation, suspicious area searching and recognition. Firstly, the serialCT images are interpolated to equal scales in each dimension, in order torecover the original3D shape of nodules. Secondly, pretreatment isimplemented to segment the lung parenchyma region, which is beneficialto the next step recognition with a simple distributed grey level image.The segmentation process includes maximum between-class variancemethod, OTSU algorithm, to select thresholds adaptively, connected-areajudgment and mathematical morphology such as the combine of openalgorithm, close algorithm, dilation and erosion. Thirdly, detect objectscalled regions of interest (ROIs) as potential nodules by threshold of graylevel decay period and region growing. Finally, distinguish ROIs to find realnodules using moment invariants. Furthermore, it deals with the whole CTserials as a group by3D thinning method. It gets good effect on regulartumors. The experimental results from CT scans data sets demonstratethat the proposed method yields a good performance of nodule detection.The system recognizes all the nodules of the data sets with a reasonable77.55%detection rate. The system could distinguish nodules from other tissues like bronchus and organs. It also provides friendly interface toallow free browsing among scans. In total, it’s an effective method forautomatic diagnosis.