작성자 M.I.D
[논문] THE CUT-OFF VALUES FOR AUTO-DETECTION OF LUNG CANCER IN CHEST RAD…
본문
THE CUT-OFF VALUES FOR AUTO-DETECTION OF LUNG CANCER IN CHEST RADIOGRAPHY: AN EXAMPLE USING AN UNSUPERVISED METHOD
Biomedical Engineering : Applications, Basis and Communications vol.24, No.06, pp. 525-536 (2012)
Accepted 17 May 2012
Published: 22 November 2012
Seokyoon Choi, Onseok Lee and Mingi Kim
Aim: For automatic recognition of lung cancer, a previous step to detect nodules automatically is needed. The aim is to propose a feature descriptor to detect the various types of nodules automatically and evaluate its performance. Materials and Methods: For experimental imaging, control group sub-image (n = 50), benign group sub-image (n = 32), malignant group sub-image (n = 38) were used, and feature of image was extracted through TIA and PCA. An excellent method was selected by using the following ROC curve analysis. Results: The experiment showed the result of the best area under the ROC curve (AUC) was shown when using TIA feature descriptor. The value at this time was 95.9 in the control group-benign group and 95.1 in the control group-malignant group. When using TIA feature descriptor, the cut-off value results were shown to be 116.4 in the control group-benign group (sensitivity: 88.9%, specificity: 90.3%), and 149.1 in the control group-malignant group (Sensitivity: 86.4%, Specificity: 92.1%). Conclusion: The proposed method can be applied to chest radiography for the automatic detection of nodules for the automatic diagnosis of lung cancer, and excellent results can be obtained.
Biomedical Engineering : Applications, Basis and Communications vol.24, No.06, pp. 525-536 (2012)
Accepted 17 May 2012
Published: 22 November 2012
Seokyoon Choi, Onseok Lee and Mingi Kim
Aim: For automatic recognition of lung cancer, a previous step to detect nodules automatically is needed. The aim is to propose a feature descriptor to detect the various types of nodules automatically and evaluate its performance. Materials and Methods: For experimental imaging, control group sub-image (n = 50), benign group sub-image (n = 32), malignant group sub-image (n = 38) were used, and feature of image was extracted through TIA and PCA. An excellent method was selected by using the following ROC curve analysis. Results: The experiment showed the result of the best area under the ROC curve (AUC) was shown when using TIA feature descriptor. The value at this time was 95.9 in the control group-benign group and 95.1 in the control group-malignant group. When using TIA feature descriptor, the cut-off value results were shown to be 116.4 in the control group-benign group (sensitivity: 88.9%, specificity: 90.3%), and 149.1 in the control group-malignant group (Sensitivity: 86.4%, Specificity: 92.1%). Conclusion: The proposed method can be applied to chest radiography for the automatic detection of nodules for the automatic diagnosis of lung cancer, and excellent results can be obtained.