Program Information
What Is the Most Optimal Mother Wavelet in Radiomic Analysis for the Survival Prediction of Lung Cancer Patients?
M Soufi1,2* , H Arimura3 , S Ohga3 , T Hirose1 , Y Umedu5 , H Honda3 , T Sasaki3 , (1) Graduate School of Medical Sciences, Kyushu University, Fukuoka, Fukuoka, (2) Research Fellow at Japan Society for the Promotion of Science, Tokyo (3) Faculty of Medical Sciences, Kyushu University, Fukuoka, Fukuoka, (4) Kyushu University Hospital, Fukuoka, Fukuoka, (5) Kyushu University Hospital, Fukuoka, Fukuoka
Presentations
SU-K-702-6 (Sunday, July 30, 2017) 4:00 PM - 6:00 PM Room: 702
Purpose: To investigate the most optimal mother wavelet in radiomic analysis for the survival prediction of lung cancer patients.
Methods: Planning computed-tomography (CT) images of 221 patients with adenocarcinoma, squamous cell carcinoma and large cell carcinoma were used for this study. The regions of interest (ROI) for tumors were extracted from the CT images by using boundary boxes of the gross tumor volumes (GTVs) as defined in the radiation treatment plans. The ROI images were decomposed into undecimated low- and high-frequency component images by using a discrete wavelet transformation. 486 radiomic features, which included the first-order statistics and texture features, were calculated in the original and wavelet-transformed images. The medians of the feature values were used for stratifying the patients into short- and long-survival groups. A univariate Cox proportional hazard regression model was employed for assessing the survival prediction of each radiomic feature. The P-value of the estimated model’s coefficients was used for comparison among the performances of the radiomic features. The P-values obtained from seven mother wavelets, i.e., Coiflets, Daubechies, symlet, Fejer-Korovkin, discrete Meyer, biorthogonal and reverse biorthogonal wavelets, were calculated for clarifying the optimal mother wavelet.
Results: The smallest P-values corresponded with features that were obtained by using Daubechies, symlet, biorthogonal and reverse biorthogonal mother wavelets, which have superior edge characterization property (P = 1.82×10⁻⁴). Coiflets, Fejer-Korovkin and discrete Meyer mother wavelets produced larger P-values (P = 2.08×10⁻³).
Conclusion: This study has shown that Daubechies, symlet, biorthogonal and reverse biorthogonal mother wavelets produce proper survival predictions in lung cancer patients.
Funding Support, Disclosures, and Conflict of Interest: This work was supported by Japan Society for the Promotion of Science Grant-in-Aid for JSPS fellows (16J04082).
Contact Email: