Research On Automatic Detection Methods For Colonic Polyps Based On CT Images

Computed tomographic colonography(CTC)is a computed tomography(CT)image-based screening technique for colonic polyps.Compared with optical colonoscopy,CTC has the advantages of non-invasiveness,low cost,high patient acceptance,and no risks of complications.However,CTC is a time-consuming procedure that can easily cause visual fatigue to doctors,and its detection results are susceptible to the subjective opinons of doctors.Therefore,computer-aided detection(CAD)technique is applied to CTC(CTC CAD thereafter)to address these issues.The CTC CAD technique locates colonic polyps in the CT images automatically,which reduces the polyp detection time and increases the polyp detection rate.In this thesis,algorithm research is carried out for polyp candidate detection,feature extraction,and data sampling.A new CTC CAD system is also proposed to assist doctors in accurately locating colonic polyps during colon cancer screening.The main research contributions of this thesis are as follows:(1)This thesis proposes a shape-based polyp candidate detection method.This method computes the shape index and multiscale sphere enhancement filter in the geodesic distance field and initially segments the suspected lesion regions by using the shape index images and sphere-enhanced images.Then,the polyp candidates are obtained after the suspected lesion regions and their inner boundaries are adjusted by using the conditional dilation and convex conditional dilation,respectively.Experimental results indicate that the proposed candidate detection method can detect more polyps and yield fewer false positives than other candidate detection method.(2)The diverse 3D radiomic features are used for polyp detection.This thesis sufficiently mines the information contained in the images and computes numerous 3D radiomic features from different feature images.These radiomic features are rich and diverse,which can not only greatly expand the feature library in the CTC CAD field but also comprehensively and quantitatively describe the shape and texture characteristics of polyp candidates.Experimental results indicate that the used 3D radiomic features can effectively distinguish true positives from false positives and have a better ability to identify polyps than the conventional CTC CAD features.(3)This thesis proposes a hybrid sampling method to deal with the imbalanced training set.The hybrid sampling method firstly removes the incorrect true positive samples and redundant false positive samples from the training set,then standardizes the training set,and finally oversamples the training set to make the sample number of both classes equal.Experimental results indicate that the proposed hybrid sampling method can better improve the detection performance of CAD system than other sampling methods.(4)This thesis proposes a new CTC CAD system.Firstly,the oral contrast agent-enhanced CT images are isotropicly interpolated,and the colonic lumen and colonic wall are segmented from the isotropic CT dataset.Then,polyp candidates are detected and segmented by using the shape-based polyp candidate detection method.Subsequently,the abnormal CT values in the colon wall are linearly corrected,and numerous radiomic features are calculated to describe the polyp candidates.Finally,the training samples are balanced by using the hybrid sampling method,and Random Forests classifier is used to classify the polyp candidates.Experimental results indicate that the proposed CAD system has good detection performance and can obtain higher sensitivity and lower false positive rate than other CTC CAD system.In sum,this thesis studies the relevant algorithms involved in the automatic detection of colonic polyps and proposes a new CTC CAD system.This system achieves a high sensitivity of 98.8% and a low false positive rate of average 1.2 false positives per CT volumetric dataset for the clinically significant polyps(i.e.,the polyps not less than 5 mm in diameter)of 672 patients.The research results show that the proposed CTC CAD system performs well and can be used as a non-invasive and effective means to improve the doctors' diagnostic accuracy in colon cancer screening.