The Study Of Computer-aided Detection And Diagnosis Of Pulmonary Nodules In CT Images

Part 1.Computer-aided Detection of Pulmonary Nodules in CT Images Based on Sparse Non-negative Matrix Factorization Model Learning: Effect on Observer PerformancePurpose: To evaluate how computer-aided detection(CADe)affects observer performance in detecting lung nodules on computed tomography(CT)scans.Methods: Two hundred chest CT scans of healthy people and 80 patients' CT scans containing 96 lung nodules were retrospectively included.The CADe technique is based on sparse non-negative matrix factorization(NMF)model learning.A total of 100 CT scans randomly selected from the 200 CT scans of healthy people were used as training data.The remaining 100 CT scans of healthy people,intermixed with the 80 patients' CT scans using a randomization method,were used as the test data.Six observers,including two senior chest radiologists,two secondary chest radiologists and two junior radiology residents,were asked to find out the potential lung nodules on the CT scans,first without and subsequently with the assist of CADe scheme.Mc Nemar's test was used to compare observer sensitivity without and with CADe.Results: Of the 96 nodules contained within these scans,89(92.7%)nodules were correctly detected by the computer,with an average 0.09 FP(false positive)annotations per CT scan.With use of the CADe scheme,the average sensitivity improved from 87.3% to 96.9% for the 6 radiologists,from 77.6% to 94.8% for junior radiology residents,from 89.1% to 97.9% for secondary chest radiologists,and from 95.3% to 97.9% for senior chest radiologists.Sixty two(64.6%)nodules were detected by all the six observers.CADe depicted two nodules that were initially not found by any of the observers.Conclusion: Our study suggests that the CADe system can improve observer sensitivity for the detection of lung nodules in CT images.Part 2.Estimation of malignancy of pulmonary nodules at CT scans: effect of computer-aided diagnosis on diagnostic performance of radiologistsPurpose: To evaluate the effect of the computer-aided diagnosis(CADx)system we developed on the diagnostic performance of radiologists with different levels of experience for the estimation of the likelihood of malignancy of pulmonary nodules on computed tomographic(CT)scans.Materials and Methods: A total of 857 malignant nodules from 601 patients(299 males,302 females)and 426 benign nodules from 278 patients(170 males,108 females)were retrospectively collected from 4 hospitals.A total of 745 malignant nodules and 370 benign nodules were used as the training data of our CADx system based on Res Net(Residual Network).The remaining 112 malignant nodules and 56 benign nodules were used as the test data.The participants were two senior chest radiologists,two secondary chest radiologists and two junior radiology residents.The readers estimated the likelyhood of malignancy of pulmonary nodules first without and then with CADx output.Receiver operating characteristic(ROC)analysis was used to evaluate readers' diagnostic performance.Results: When a threshold level of 58% was used for estimate of the likelihood of malignancy,the sensitivity,specificity,and diagnostic accuracy values of our CADx scheme alone were 93.8%,83.9%,and 90.5%,respectively.For all 6 readers,the mean area under the ROC curve(Az)values without and with CADx system were 0.913 and 0.938,respectively.The difference of Az values without and with CADx system for each reader was significant.With CADx output,The readers made correct changes an average of 15.7 times and incorrect changes an average of 2 times.Conclusion: Our CADx system can effectively distinguish malignant nodules and benigh nodules.Use of our CADx system significantly improved the diagnostic performance of readers regardless of their experience level for assessment of the likelyhood of malignancy of pulmonary nodules.