Quantitative CT texture analysis in predicting PD-L1 expression in locally advanced or metastatic NSCLC patients

Abstract

Purpose The assessment of Programmed death-ligand 1 (PD-L1) expression has become a game changer in the treatment of patients with advanced non-small cell lung cancer (NSCLC). We aimed to investigate the ability of Radiomics applied to computed tomography (CT) in predicting PD-L1 expression in patients with advanced NSCLC. Methods By applying texture analysis, we retrospectively analyzed 72 patients with advanced NSCLC. The datasets were randomly split into a training cohort (2/3) and a validation cohort (1/3). Forty radiomic features were extracted by manually drawing tumor volumes of interest (VOIs) on baseline contrast-enhanced CT. After selecting features on the training cohort, two predictive models were created using binary logistic regression, one for PD-L1 values ≥ 50% and the other for values between 1 and 49%. The two models were analyzed with ROC curves and tested in the validation cohort. Results The Radiomic Score (Rad-Score) for PD-L1 values ≥ 50%, which consisted of Skewness and Low Gray-Level Zone Emphasis (GLZLM_LGZE), presented a cut-off value of − 0.745 with an area under the curve (AUC) of 0.811 and 0.789 in the training and validation cohort, respectively. The Rad-Score for PD-L1 values between 1 and 49% consisted of Sphericity, Skewness, Conv_Q3 and Gray Level Non-Uniformity (GLZLM_GLNU), showing a cut-off value of 0.111 with AUC of 0.763 and 0.806 in the two population, respectively. Conclusion Rad-Scores obtained from CT texture analysis could be useful for predicting PD-L1 expression and guiding the therapeutic choice in patients with advanced NSCLC.

Ability of Delta Radiomics to Predict a Complete Pathological Response in Patients with Loco-Regional Rectal Cancer Addressed to Neoadjuvant Chemo-Radiation and Surgery

“…Many authors have already investigated the possible correlation of imaging techniques, especially MRI (volumetric analysis and TA) with patient outcomes; however, many of them mainly used T2 and DWI maps [ 23 , 31 , 76 , 77 , 78 , 79 , 80 , 81 , 82 , 83 , 84 ]. Conversely, CT imaging is used in different pathologies [ 85 , 86 , 87 , 88 , 89 , 90 , 91 ]. The results obtained in the present study show that TA applied to ADC maps could help to assess clinical and pathological response in LARC.…”

Correlation between PD-L1 expression and radiomic features in early-stage lung adenocarcinomas manifesting as ground-glass nodules

“…Several studies have also attempted to establish CT-based PD-L1 expression prediction models in advanced NSCLC, but yielded inconsistent results. Bracci et al created two radiomic models based on 48 texture features: one model determining whether TPS ≥1% achieved AUC values of 0.763 and 0.806 in the training cohort (n=48) and the validation cohort (n=24), and the other for TPS≥50% got AUC values of 0.811 and 0.789 respectively ( 25 ). Sun et al built a radiomic model for PD-L1 expression≥50% on a much larger data set (390 patients, 200 texture features), and achieved the similar predictive effect (AUC: 0786 and 0.807) ( 42 ).…”

Multi-Omics Approaches for the Prediction of Clinical Endpoints after Immunotherapy in Non-Small Cell Lung Cancer: A Comprehensive Review

“…Several radiomics models were able to predict the level of expression of PD-L1 with reasonable performances [ 84 ]. For instance, using a small retrospective overall cohort of 72 patients and combining two features extracted from pre-treatment CTs, the model reached an AUC of 0.79 for the prediction of PD-L1 values ≥ 50% in the validation cohort [ 85 ]. Combining three different features resulted in the prediction of PD-L1 values between 1 and 49% with an AUC of 0.81 in the validation cohort.…”