Start Your Search
MA05 - Improving Outcomes in Locoregional NSCLC II (ID 901)
- Event: WCLC 2018
- Type: Mini Oral Abstract Session
- Track: Treatment of Locoregional Disease - NSCLC
- Presentations: 1
- Coordinates: 9/24/2018, 13:30 - 15:00, Room 105
MA05.11 - Radiomics Analysis Using SVM Predicts Mediastinal Lymph Nodes Status of Squamous Cell Lung Cancer by Pre-Treatment Chest CT Scan (ID 12033)
14:40 - 14:45 | Author(s): Quanzheng Li
Assessment of mediastinal lymph nodes (N2 station) is essential in staging patients with Non-small-cell lung cancer (NSCLC), for patients with preoperative confirmed N2 status should follow neoadjuvant therapy before surgery, and occult N2 status should be avoided. There are several invasive and non-invasive exams available for preoperative N staging, like EBUS-TBNA and PET-CT scan. Chest CT scan was the basic examination of every patient, while only the length of minor axis could be used to predict lymph node involvement, and the potential value of CT might be underestimated. In this study we aimed to explore the value of radiomics analysis with machine learning in differentiating N2 from N1/N0 subjects using pre-treatment chest CT.
Ninety-three patients with squamous cell lung cancer, who underwent pre-treatment CT scans were included in this study. By use of Laplacian of Gaussian (LoG) filter and matrix based radiomics models (e.g. gray-level co-occurrence matrix), comprehensive radiomics features were extracted from the regions of interest which were manually delineated on primary tumors. We performed radiomics analysis using support vector machine (SVM) to test texture and heterogeneity features derived from pre-treatment CT images as indicators for the staging of lymph node metastasis, especially N2. The gold standard of N staging is confirmed pathologically after systematic mediastinal lymphadenectomy (N2 subjects=31).4c3880bb027f159e801041b1021e88e8 Result
For the performance evaluation of single image feature, there are 16 features able to differentiate N2 subjects from others (N0 and N1) with p value <0.05. Furthermore, SVM training and classification were performed using 5-feature combinations as inputs. With feature selection, the best performance of N2 prediction is 83% accuracy with 87% sensitivity and 81% specificity.
Radiomics analysis using SVM training can successfully predict N staging by pre-treatment chest CT scan for NSCLC patients, which could diminish the odds of occult N2 status and provide unique information preoperatively for treatment planning.6f8b794f3246b0c1e1780bb4d4d5dc53