Radiomics Signature Based On Magnetic Resonance Diffusion Weighted Imaging For Discriminating Between Benign And Malignant Pulmonary Lesions

Purpose1.To compare the value of single shot turbo spin-echo(TSE)and single shot echo-planar imaging(EPI)in lung cancer imaging with single-shot echo-planar imaging(DWI).2.To investigate the robustness and reproducibility of radiomics features in lung TSE-DWI and EPI-DWI.3.To develop and validate the diagnostic value of a radiomics signature based on diffusion weighted imaging(DWI)and apparent diffusion coefficient(ADC)maps in patients with pulmonary lesions.Methods1.TSE-DWI and EPI-DWI were performed twice in 15 cases of lung cancer.The Distortion ratio(DR),signal-to-noise ratio(SNR)and contrast-to-noise ratio(CNR)of the two methods were compared.Two groups of parameters were analyzed by bland-altman analysis.The coefficient of variation(CV)and intra-class correlation coefficient(ICC)were used to evaluate the repeatability and inter-observer consistency of the short-term retest parameters.2.Images of 30 patients underwent lung TSE-DWI and EPI-DWI were segmented by the two observers.A total of 396 radiomic features in 6 categories were extracted from per-patient images.Concordance correlation coefficients(CCC)and dynamic range(DR')were used to evaluate the robustness of the radiomics features between two repeated scans.The reproducibility of features within and between observers was assessed by ICC.3.A total of 153 patients who underwent lung DWI were retrospectively studied.The samples were balanced by using synthetic minority over-sampling technique and were randomly divided into a training cohort and a validation cohort according to a ratio of 7:3.ADC values(ADC800)were measured and radiomics features were extracted from the DWI and ADC maps.The maximum correlation and minimum redundancy,Lasso regression and logistic regression were used to screen the features and build radiomics signature.A nomogram including clinical features,rad-score and ADC values was established by multivariate logistic regression.The diagnostic efficacy and clinical value of different models were evaluated by ROC and decision curve.Results1.TSE-DWI has lower SNR,similar CNR and significantly smaller DR compared with EPI-DWI.The Bland-Altman analysis showed unacceptable limits of agreement between these two sequences.Test-retest repeatability was good for ADC and D of EPI-DWI(CV 14.11-16.60%and 17.08-19.53%)and excellent for ADC and D of TSE-DWI(CV 4.8%-6.19%and 6.05-8.71%),but relatively poor for perfusion fraction(f)and pseudo-diffusion coefficient(D*)(CV 25.95-27.70%and 56.92-71.84%for EPI,23.67-28.67%and 60.85-70.17%for TSE).For inter-observer agreement,both techniques were good to excellent in ADC and D(The lower limit of 95%confidence interval for ICC was almost all greater than 0.75),while D*and f had higher inter-observer variabilities with D*of TSE-DWI showing poorest reproducibility(ICC-0.27 to 0.12).2.In both TSE and EPI sequences,the number of stable features extracted based on DWI(CCC and DR>0.85)was higher than the corresponding ADC maps.The proportion of robust feature of TSE-DWI?EPI-DWI?TSE-ADC?EPI-ADC was 49.7%(197/396),42.7%(169/396),31.8%(126/396)and 37.4%(148/396),respectively.Of the 396 features,54(13.6%)of them showed great robustness(CCC and DR'?0.85)and interobserver and interobserver reproducibility(ICC?0.75)across all sequences.3.The AUC of differentiating benign and malignant pulmonary lesions was 0.694(95%CI 0.524-0.864)for clinical model,0.802(0.666-0.937)for ADC800,0.885(0.774-0.996)for ADC radiomic signature and 0.767(0.578-0.956)for DWI radiomic signature in the test group.The ADC and DWI combined radiomic model(AUC=0.670(0.467-0.873))was not superior to the single sequence radiomics model.The nomogram combined with clinical variate,ADC800 and rad-score had the best performance and were significantly better than clinical model and combined radiomic model.The decision curve shows the ADC radiomics model is superior to the ADC800 and the clinical model;the nomogram bring the most net benefit to patients in most cases.Conclusion1.Lung DWI or IVIM using TSE could provide distortion-free images and improve the test-retest robustness of ADC and D as compared to EPI-DWI,however,it might exert a negative effect on perfusion parameter D*.2.Each sequence has different number of robust features.Some features are robust and reproducible across different scans,observers and sequences.These robust and reproducible features lay a foundation for future diffusion radiomics studies.3.ADC radiomics signature has better application value in differentiating pulmonary benign and malignant lesions compared with ADC value and clinical model.The nomogram,including radiomics signature,ADC values,and clinical data showed better diagnostic performance and clinical value,providing a new idea for personalized and accurate diagnosis of lung cancer.