Computer-aided Detection And Diagnosis Of Pulmonary Nodules In Chest CT Scans

Lung cancer is the leading cause of cancer-related death worldwide.A primary way for improving lung cancer patients’ survival rates is to detect cancerous pulmonary nodules at an early stage.Currently,computed tomography(CT)is increasingly used as an effective imaging tool for detecting pulmonary nodules at early stage.However,each CT scan contains 150~500 slices with image slice thickness in a range 0.5~2mm,which can be a cumbersome,time-consuming task for radiologists to read and evaluate such large number of CT image data of each case.Thus,developing computer-aided detection and/or diagnosis(CAD)schemes to assist radiologists in reading and interpreting lung CT images has been attracting extensive research interest in the past two decades.In order to improve the performance of CAD scheme,this study has proposed and developed new algorithms and methods to build computer-aided detection(CADe)scheme and computer-aided diagnosis(CADx)scheme.First,in order to improve the performance of CADe system,this study develops two different CADe schemes to detect pulmonary nodules in CT scan.The descriptions of two schemes are as follows.?We propose a new CADe scheme for pulmonary nodule detection based on 3D dynamic self-adaptive template matching and Fisher linear discriminant analysis(FLDA)classifier.First,in order to segment lung volume,we use an OTSU algorithm and three-dimensional(3D)region growing algorithm to segment the CT scan.Next,we extract and filter the suspicious regions of interest(ROIs)by applying 3D dot filtering and thresholding method.Then,we roughly detect pulmonary nodule candidates by using 3D dynamic self-adaptive template matching.Finally,we apply a FLDA(Fisher’s linear discriminant analysis)classifier to reduce false positive nodules.By comparing with other previously reported CADe schemes,the results demonstrate that our new scheme can yield higher and more robust performance in detecting pulmonary nodules.?In this study,we present a novel scheme for automatic detection of pulmonary nodules in CT images based on a 3D tensor filtering algorithm and local image feature analysis.We first apply a series of preprocessing steps to segment the lung volume and generate the isotropic volumetric CT data.Next,a unique 3D tensor filtering approach and local image feature analysis are used to detect nodule candidates.Finally,a random forest classifier is trained to classify the detected candidates.Our detection results show that the use of 3D tensor filtering algorithm combined with local image feature analysis constitutes an effective approach to detect pulmonary nodules.Comparing the two methods proposed in this study,we found that the second CADe scheme generated higher performance than former one.Second,to investigate the effect of training data sets on the performance of CADx scheme,and further improve the performance of CADx scheme,we have explored two different studies.The details of the two studies are as follows.?This study aims to develop a CADx scheme for classification between malignant and benign lung nodules,and also assess whether CADx performance changes in detecting nodules associated with early and advanced stage lung cancer.By using image processing approaches and feature analyais methods such as,nodule segmentation,feature extraction and selection,SMOTE(Synthetic Minority Over-sampling Technique)oversampling,and machine learning classifier,we build three CADx scheme with three datasets involving:(1)all nodules,(2)benign and stage I malignant nodules,and(3)benign and stage III malignant nodules.This study demonstrates(1)feasibility of applying CADx scheme to accurately classify between benign and malignant lung nodules,and(2)a positive trend between CADx performance and cancer progression stage.Thus,in order to increase CADx performance in detecting subtle and early cancer,training dataset should include more diverse early stage cancer cases.?This study aims to develop multi-feature based CADx for lung cancer by using quantitative imaging(QI)features and serum biomarkers,and investigate the information-fusion method to improve the performance of CADx in classifying benign and malignant pulmonary nodules.First,we use QI features and serum biomarker features to build two CADx schemes with a support vector machine(SVM)classifier,respectively.Then,we train and test the SVM classifiers using the overall dataset with a Relief feature selection method,a synthetic minority over-sampling technique and a leave-one-case-out validation method.Finally,to further improve the performance of the CADx scheme,we use an information-fusion method to combine the prediction scores of the two schemes.This study demonstrates(1)feasibility of improving the CADx performance by combining QI feature and biomarker,and(2)QI features perform better than serum biomarkers in classifying between benign and malignant nodules,which may be considered as new “biomarkers”.Thus,in order to improve the CADx performance in classifying benign and malignant nodules,more attentions should be focused on developing effective QI features,and exploring optimal information-fusion models of different types of features.Last,after investigating the algorithms for CADe and CADx,we develop a CAD software for lung nodules in CT scan.All the algorithms proposed in our study have been applied in this software.This CAD software contains four parts namely,CT scan input module,image reader module,segmentation module and detection and analysis module.By using the image processing and analysis methods proposed in our previous studies,this software implement and realize the function of pulmonary detection and analysis.In summary,in order to develop new CADe and CADx schemes,we have proposed some novel algorithms to build the CAD schemes and further improved the performance of CAD scheme.By using these algorithms proposed in this paper,we have developed a CAD software for detecting lung nodules in CT scan.In the future study,we will investigate and test the performance of the proposed CAD scheme in clinical.