Y Ieko^1,2*, N Kadoya¹, K Abe^1,3, S Tanaka¹, H Takagi⁴, T Kanai⁵, K Ichiji⁶, T Yamamoto¹, H Ariga², K Jingu¹, (1) Department of Radiation Oncology, Tohoku University Graduate School of Medicine, Sendai, Japan, (2) Department of Radiation Oncology, Iwate Medical University School of Medicine, Iwate, Japan, (3) Department of Radiology, Tokyo Metropolitan Cancer and Infectious Diseases Center Komagome Hospital, Tokyo, Japan, (4) Department of Radiological Technology, School of Health Sciences, Faculty of Medicine, Tohoku University, Sendai, Japan, (5) Department of Radiation Oncology, Yamagata University Faculty of Medicine, Yamagata, Japan, (6) Department of Radiological Imaging and Informatics, Tohoku University Graduate School of Medicine, Sendai, Japan.

• 
  Y Ieko

(Sunday, 7/12/2020)   [Eastern Time (GMT-4)]

Room: AAPM ePoster Library

Purpose: purpose of this study was to evaluate the potential of radiomics features extracted from diagnostic CT images for predicting forced expiratory volume in one second (FEV1) measured using pulmonary function tests (PFT).

Methods: patients receiving thoracic radiation therapy were considered for this study. For each patient, pretreatment diagnostic breath-holding chest CT images and the FEV1 value were used, with the median interval between CT and PFT being 28 days. Diagnostic CT scans were acquired using a helical scan with an image resolution of 0.55–0.82 × 0.55–0.82 × 0.6–2.0 mm³ and a tube voltage of 100–120 kVp (Toshiba Aquilion ONE: n=19, Siemens Somatom Definition Flash: n=20, Siemens Somatom Sensation Cardiac: n=8, GE BrightSpeed: n=3). A total of 1,751 radiomics were extracted from the region of interest of the total lung using the IBEX software, including 188 first-order histogram-based features and 1,561 second-order texture-based features. First, we performed feature selection using the significant difference between the radiomic feature and the FEV1 value with a paired student t-test (p<0.05). Then, the rad score was reconstructed using the least-absolute-shrinkage-and-selection-operator (LASSO) logistic-regression model with the selected features (R package). To evaluate the prediction accuracy of the rad score, the Pearson correlation coefficient (r) between the predicted and measured FEV1 values was calculated.

Results: total of 308 features showed significant differences for FEV1 (p<0.05). The rad score with the LASSO model employed sixteen radiomic features: the Gray Level Co-occurrence Matrix (n=11), Intensity Direct (n=2), Neighbor Intensity Difference (n=2), and Gradient Orient Histogram (n=1). We found a strong correlation between rad score and FEV1 (r=0.83, p<0.01).

Conclusion: results showed a strong correlation between the measured and predicted FEV1, suggesting that the diagnosis of CT-based radiomic features could predict the parameters measured using PFT.

Funding Support, Disclosures, and Conflict of Interest: Grant-in-Aid for Scientific Research (C) (19K08116)

Keywords

Image Analysis, Lung, Feature Selection

Taxonomy

IM- CT: Radiomics

Contact Email