Fusion of shallow and deep features from 18F-FDG PET/CT for predicting EGFR-sensitizing mutations in non-small cell lung cancer

Abstract

Background: Non-small cell lung cancer (NSCLC) patients with epidermal growth factor receptor-sensitizing (EGFR-sensitizing) mutations exhibit a positive response to tyrosine kinase inhibitors (TKIs). Given the limitations of current clinical predictive methods, it is critical to explore radiomics-based approaches. In this study, we leveraged deep-learning technology with multimodal radiomics data to more accurately predict EGFR-sensitizing mutations.

Methods: A total of 202 patients who underwent both flourine-18 fluorodeoxyglucose positron emission tomography/computed tomography (¹⁸F-FDG PET/CT) scans and EGFR sequencing prior to treatment were included in this study. Deep and shallow features were extracted by a residual neural network and the Python package PyRadiomics, respectively. We used least absolute shrinkage and selection operator (LASSO) regression to select predictive features and applied a support vector machine (SVM) to classify the EGFR-sensitive patients. Moreover, we compared predictive performance across different deep models and imaging modalities.

Results: In the classification of EGFR-sensitive mutations, the areas under the curve (AUCs) of ResNet-based deep-shallow features and only shallow features from different multidata were as follows: RES_TRAD, PET/CT vs. CT-only vs. PET-only: 0.94 vs. 0.89 vs. 0.92; and ONLY_TRAD, PET/CT vs. CT-only vs. PET-only: 0.68 vs. 0.50 vs. 0.38. Additionally, the receiver operating characteristic (ROC) curves of the model using both deep and shallow features were significantly different from those of the model built using only shallow features (P<0.05).

Conclusions: Our findings suggest that deep features significantly enhance the detection of EGFR-sensitizing mutations, especially those extracted with ResNet. Moreover, PET/CT images are more effective than CT-only and PET-only images in producing EGFR-sensitizing mutation-related signatures.

Keywords: Epidermal growth factor receptor-sensitizing mutation (EGFR-sensitizing mutation); deep learning; flourine-18 fluorodeoxyglucose positron emission tomography/computed tomography (18F-FDG PET/CT); non-small cell lung cancer (NSCLC); radiomics.

Figure 1

The workflow for NSCLC patients with PET/CT images and clinical data. NSCLC, non-small cell lung cancer; ¹⁸F-FDG PET/CT, flourine-18 fluorodeoxyglucose positron emission tomography/computed tomography; EGFR, epidermal growth factor receptor; TKI, tyrosine kinase inhibitor; ROI, region of interest; NSE, neuron-specific enolase; SCC, squamous cell carcinoma.

Figure 2

A systematic exposition of the radiomics workflow used in this study for EGFR-sensitizing mutation prediction in NSCLC patients. Model 1, the ONLY_TRAD model uses only traditional features; Model 2, the RES_TRAD model includes all traditional features and deep features extracted with the ResNet-101 network. PET/CT, positron emission tomography/computed tomography; LoG, Laplacian of Gaussian; LASSO, least absolute shrinkage and selection operator; ROC, receiver operating characteristic; AUC, area under the curve; EGFR, epidermal growth factor receptor; NSCLC, non-small cell lung cancer.

Figure 3

The most predictive features with the corresponding coefficients selected for constructing RES_TRAD model. CT_F denotes CT deep features; PET_F denotes PET deep features. PET, positron emission tomography; CT, computed tomography; LDHLE, large dependence high gray-level emphasis; HGLRE, high gray-level run emphasis.

Figure 4

Correlations for features extracted by VGG_TRAD. (A) Heatmap of the correlations between the CT features and labels. CT_F denotes the CT deep features; (B) heatmap of the correlations between the PET features and labels, where PET_F represents the PET deep features. LGLRE, low gray level run emphasis; PET, positron emission tomography; CT, computed tomography; MP, maximum probability; P, percentile.

Figure 5

ROC curves and AUC values for evaluating the predictive abilities of the models based on CT-only, PET-only, and PET/CT images. The results of the different multidata in the ONLY_TRAD model are shown as ONLY TRAD_PETCT_AUC, ONLY TRAD_CT_AUC, and ONLY TRAD_PET_AUC. The results of the different multidata in the RES_TRAD model are shown as RES + TRAD_PETCT_AUC, RES + TRAD _CT_AUC, and RES + TRAD _PET_AUC. The results of the different multidata in the VGG_TRAD model are shown as VGG + TRAD_PETCT_AUC, VGG + TRAD_CT_AUC, and VGG + TRAD_PET_AUC. ROC, receiver operating characteristic; AUC, area under the curve; PET, positron emission tomography; CT, computed tomography.

Figure 6

The decision curve analysis of three models. (A) The decision curve analysis of the RES_TRAD model; (B) the decision curve analysis of the ONLY_TRAD model; (C) the decision curve analysis of the VGG_TRAD model. RES_TRAD denotes the model with ResNet-deep and shallow features. ONLY_TRAD denotes the model with only shallow features. VGG_TRAD denotes the model with VGGdeep and shallow features. CT, computed tomography; PET, positron emission tomography.