Two major treatment strategies employed in non-small cell lung cancer, NSCLC, are tyrosine kinase inhibitors, TKIs, and immune checkpoint inhibitors, ICIs. The choice of strategy is based on heterogeneous biomarkers that can dynamically change during therapy. Thus, there is a compelling need to identify comprehensive biomarkers that can be used longitudinally to help guide therapy choice. Herein, we report a ¹⁸F-FDG-PET/CT-based deep learning model, which demonstrates high accuracy in EGFR mutation status prediction across patient cohorts from different institutions. A deep learning score (EGFR-DLS) was significantly and positively associated with longer progression free survival (PFS) in patients treated with EGFR-TKIs, while EGFR-DLS is significantly and negatively associated with higher durable clinical benefit, reduced hyperprogression, and longer PFS among patients treated with ICIs. Thus, the EGFR-DLS provides a non-invasive method for precise quantification of EGFR mutation status in NSCLC patients, which is promising to identify NSCLC patients sensitive to EGFR-TKI or ICI-treatments.

Although cancer often is referred to as “a disease of the genes,” it is indisputable that the (epi)genetic properties of individual cancer cells are highly variable, even within the same tumor. Hence, preexisting resistant clones will emerge and proliferate after therapeutic selection that targets sensitive clones. Herein, the authors propose that quantitative image analytics, known as “radiomics,” can be used to quantify and characterize this heterogeneity. Virtually every patient with cancer is imaged radiologically. Radiomics is predicated on the beliefs that these images reflect underlying pathophysiologies, and that they can be converted into mineable data for improved diagnosis, prognosis, prediction, and therapy monitoring. In the last decade, the radiomics of cancer has grown from a few laboratories to a worldwide enterprise. During this growth, radiomics has established a convention, wherein a large set of annotated image features (1‐2000 features) are extracted from segmented regions of interest and used to build classifier models to separate individual patients into their appropriate class (eg, indolent vs aggressive disease). An extension of this conventional radiomics is the application of “deep learning,” wherein convolutional neural networks can be used to detect the most informative regions and features without human intervention. A further extension of radiomics involves automatically segmenting informative subregions (“habitats”) within tumors, which can be linked to underlying tumor pathophysiology. The goal of the radiomics enterprise is to provide informed decision support for the practice of precision oncology.

Background: Currently, only a fraction of patients with non-small cell lung cancer (NSCLC) treated with immune checkpoint inhibitors (ICIs) experience a durable clinical benefit (DCB). According to NCCN guidelines, Programmed death-ligand 1 (PD-L1) expression status determined by immunohistochemistry (IHC) of biopsies is the only clinically approved companion biomarker to trigger the use of ICI therapy. Based on prior work showing a relationship between quantitative imaging and gene expression, we hypothesize that quantitative imaging (radiomics) can provide an alternative surrogate for PD-L1 expression status in clinical decision support. Methods: ¹⁸F-FDG-PET/CT images and clinical data were curated from 697 patients with NSCLC from three institutions and these were analyzed using a small-residual-convolutional-network (SResCNN) to develop a deeply learned score (DLS) to predict the PD-L1 expression status. This developed model was further used to predict DCB, progression-free survival (PFS), and overall survival (OS) in two retrospective and one prospective test cohorts of ICI-treated patients with advanced stage NSCLC. Results: The PD-L1 DLS significantly discriminated between PD-L1 positive and negative patients (area under receiver operating characteristics curve ≥0.82 in the training, validation, and two external test cohorts). Importantly, the DLS was indistinguishable from IHC-derived PD-L1 status in predicting PFS and OS, suggesting the utility of DLS as a surrogate for IHC. A score generated by combining the DLS with clinical characteristics was able to accurately (C-indexes of 0.70–0.87) predict DCB, PFS, and OS in retrospective training, prospective testing and external validation cohorts. Conclusion: Hence, we propose DLS as a surrogate or substitute for IHC-determined PD-L1 measurement to guide individual pretherapy decisions pending in larger prospective trials.

The aim of the study is to assess the staging value of the tumor heterogeneity characterized by texture features and other commonly used semi-quantitative indices extracted from ¹⁸ F-FDG PET images of cervical cancer (CC) patients. Forty-two patients suffering CC at different stages were enrolled in this study. Firstly, we proposed a new tumor segmentation method by combining the intensity and gradient field information in a level set framework. Secondly, fifty-four 3D texture features were studied besides of SUVs (SUVmax, SUVmean, SUVpeak) and metabolic tumor volume (MTV). Through correlation analysis, receiver-operating-characteristic (ROC) curves analysis, some independent indices showed statistically significant differences between the early stage (ES, stages I and II) and the advanced stage (AS, stages III and IV). Then the tumors represented by those independent indices could be automatically classified into ES and AS, and the most discriminative feature could be chosen. Finally, the robustness of the optimal index with respect to sampling schemes and the quality of the PET images were validated. Using the proposed segmentation method, the dice similarity coefficient and Hausdorff distance were 91.78 ± 1.66% and 7.94 ± 1.99 mm, respectively. According to the correlation analysis, all the fifty-eight indices could be divided into 20 groups. Six independent indices were selected for their highest areas under the ROC curves (AUROC), and showed significant differences between ES and AS ( P < 0.05). Through automatic classification with the support vector machine (SVM) Classifier, run percentage (RP) was the most discriminative index with the higher accuracy (88.10%) and larger AUROC (0.88). The Pearson correlation of RP under different sampling schemes is 0.9991 ± 0.0011. RP is a highly stable feature and well correlated with tumor stage in CC, which suggests it could differentiate ES and AS with high accuracy.