[18F]FDG PET immunotherapy radiomics signature (iRADIOMICS) predicts response of non-small-cell lung cancer patients treated with pembrolizumab
, , , , , , , , , , et
29 juil. 2020
À propos de cet article
Catégorie d'article: Research Article
Publié en ligne: 29 juil. 2020
Pages: 285 - 294
Reçu: 30 avr. 2020
Accepté: 05 juin 2020
DOI: https://doi.org/10.2478/raon-2020-0042
Mots clés
© 2020 Damijan Valentinuzzi, Martina Vrankar, Nina Boc, Valentina Ahac, Ziga Zupancic, Mojca Unk, Katja Skalic, Ivana Zagar, Andrej Studen, Urban Simoncic, Jens Eickhoff, Robert Jeraj, published by Sciendo
This work is licensed under the Creative Commons Attribution-NonCommercial-NoDerivatives 3.0 License.
Figure 1

Figure 2

Figure 3
![Receiver operating characteristic (ROC) curve analysis. Blue: baseline iRADIOMICS multivariate logistic model (independent variables: Small Run Emphasis [SRE], Difference Entropy), yellow: baseline iRADIOMICS univariate logistic model (independent variable: SRE), grey: month 1 iRECIST univariate logistic model (independent variable: iRECIST response category), red: month 4 iRECIST univariate logistic model (independent variable: iRECIST response category). For each model, area under curve (AUC) and 95% confidence interval (CI) are reported.](https://sciendo-parsed.s3.eu-central-1.amazonaws.com/647355c34e662f30ba53aa2a/j_raon-2020-0042_fig_003.jpg?X-Amz-Algorithm=AWS4-HMAC-SHA256&X-Amz-Content-Sha256=UNSIGNED-PAYLOAD&X-Amz-Credential=AKIA6AP2G7AKOUXAVR44%2F20250909%2Feu-central-1%2Fs3%2Faws4_request&X-Amz-Date=20250909T030107Z&X-Amz-Expires=3600&X-Amz-Signature=73c5d53076decd27cf1ebc1e1b4dfcf3f709441d6a69a30e37717b88207280a2&X-Amz-SignedHeaders=host&x-amz-checksum-mode=ENABLED&x-id=GetObject)
Patient demographic and clinical data_ The data is presented for all patients, responders (overall survival [OS] > 14_9 months), and non-responders (OS < 14_9 months)_ The reported p-value is the result of Mann-Whitney U-test (MWU) (continuous variables) and Fisher exact test (categorical variables) comparing differences between responders and non-responders
Characteristic | All patients median (range) | Responders (OS > 14.9 months) median (range) | Non-responders (OS < 14.9 months) median (range) | p-value |
---|---|---|---|---|
Number of patients | 30 | 16 | 14 | |
Age [years] | 65 (46–77) | 67 (48–76) | 61 (46–77) | 0.298 |
PD-L1 TPS [%] | 75 (3–100) | 77.5 (3–100) | 75 (10–100) | 0.933 |
Female | 15 | 9 | 6 | |
Male | 15 | 7 | 8 | |
Adenocarcinoma | 17 | 8 | 9 | |
Squamous cell carcinoma | 8 | 4 | 4 | |
Other | 5 | 4 | 1 | |
Never | 1 | 0 | 1 | |
Former > 3 years ago | 12 | 7 | 5 | |
Former < 3 years ago | 5 | 3 | 2 | |
Until current disease | 8 | 3 | 5 | |
Current smoker | 4 | 3 | 1 | |
0 | 8 | 2 | 6 | |
1 | 18 | 12 | 6 | |
2 | 4 | 2 | 2 | |
1st | 15 | 10 | 5 | |
2nd | 13 | 4 | 9 | |
3rd | 2 | 2 | 0 | |
No | 24 | 12 | 12 | |
Yes | 6 | 4 | 2 |
Baseline radiomics features of primary tumours – Mann-Whitney U-test (MWU) and receiver operating characteristic (ROC) curve analysis_ Patients were dichotomized into 2 groups: responders (OS > 14_9 months) and non-responders (OS < 14_9 months)_ For each radiomics feature median value, range, p-value of MWU, and the area under the ROC curve (AUC) with the corresponding 95% confidence interval (CI), are reported_ See also Figure 1
Feature | Responders (OS > 14.9 months) median (range) | Non-responders (OS < 14.9 months) median (range) | p-value | AUC (95% CI) |
---|---|---|---|---|
Volume [cm3] | 27.9 (2.64–351) | 44.4 (7.81–792) | 0.098 | 0.69 (0.49–0.89) |
SUVmax [g/ml] | 20.6 (5.21–32.1) | 15.6 (9.54–37.0) | 0.185 | 0.65 (0.43–0.87) |
Sum entropy | 3.69 (3.53–3.77) | 3.7 (3.54–3.76) | 0.387 | 0.60 (0.38–0.82) |
Difference entropy | 2.98 (2.74–3.07) | 2.89 (2.74–3.06) | 0.080 | 0.70 (0.49–0.90) |
Baseline radiomics features of primary tumours – univariate and multivariate Cox proportional hazards regression analysis (Cox PH)_ For each radiomics feature, the hazard ratio (HR), corresponding 95% confidence interval (CI), and p-value of univariate analysis are reported_ The 2-variable multivariate regression model was chosen based on the Akaike information criterion (AIC)_ In order to achieve comparable HRs, all radiomics features were normalized into z-scores
Feature | Univariate HR (95% CI) | Univariate p-value | Multivariate HR (95% CI) | Multivariate p-value |
---|---|---|---|---|
SUVmax | 0.77 (0.46–1.3) | 0.320 | ||
Sum Entropy | 0.96 (0.60–1.5) | 0.860 | ||
Entropy-GLCM | 1.4 (0.82–2.3) | 0.230 | ||