Program Information
CT Radiomics Analysis to Differentiate Between Benign Parenchymal Lesions and Malignant Pulmonary Nodules
S Tu1*, (1) Chang Gung University, Tao-yuan, Taiwan
Presentations
SU-I-GPD-J-11 (Sunday, July 30, 2017) 3:00 PM - 6:00 PM Room: Exhibit Hall
Purpose: We used the radiomics approach with thin-section CT images to differentiate nodules of different biological invasiveness between the benign parenchymal lesion and malignant pulmonary nodule of non-small cell lung cancer.
Methods: This study is retrospective and the approval of institutional review board was formally granted. 120 image sets of thin-section CT and their pathology reports were reviewed. These images were acquired from the GE BrightSpeed 16 scanner at Chang Gung Memorial Hospital and the slice-thickness of these images was 0.625 mm. The software program of Imaging Biomarker Explorer (IBEX) was used in the nodule delineation and quantitative radiomics analysis. These nodules were delineated by a research assistant and approved by a radiologist of 10 years. 275 image features were extracted from IBEX and the t-test with a p-value threshold of 0.05 was used to study the differentiation performance. A machine learning algorithm of support vector machine was implemented to study the classification performance between these two different classes of nodules. We used 10-fold cross-validation and receiver operating characteristic curve (ROC) to evaluate the accuracy of machine learning classification. The test of Spearman rank correlation was used to further study the dependency among different image features.
Results: 168 out of 275 features were found to allow differentiation between benign parenchymal lesions and malignant pulmonary nodules. These features included entropy (p=0.013868908), variance (p=0.0017074), surface area density (p=0.025637055), auto correlation (p=0.002501154), and contrast (p=0.004370289). The accuracy of machine learning classification was 0.70±0.01 with the 10-fold cross-validation. The area under the ROC curve was 0.75±0.04.
Conclusion: A benign parenchymal lesion is often diagnosed incorrectly as a malignant pulmonary nodule. Consequently, the patient may receive an aggressive and invasive treatment. Our CT radiomics work allowed differentiation between nodules of different classes and may support the radiologist with more accurate diagnosis.
Contact Email: