PET-Based Radiogenomics Supports KEAP1/NFE2L2 Pathway Targeting for Non-Small Cell Lung Cancer Treated with Curative Radiotherapy

PURPOSE/OBJECTIVE(S)

Radiotherapy (RT) is one of the major treatment options for localized lung cancer. Either delivered in normo- or moderate/highly hypofractionated regimens, use of RT is increasing especially thanks to the development of stereotactic radiotherapy (SBRT). RT is associated with a higher risk of local relapse when compared with surgery but with a preferable toxicity profile. The KEAP1/NFE2L2 mutational status has been proven as significantly correlated with local relapse in patients treated with RT. Several (33) transcriptomic features were previously identified as dependent from the KEAP1/NFE2L2 mutational status. However, these genetic and transcriptomic tests are rarely performed because of their cost and lack of availability. Prediction of the KEAP1/NFE2L2 mutational status on non-invasive modalities such as imaging could help in further personalizing each therapeutic strategy.

MATERIALS/METHODS

Due to the small size of patients with both mutation status (MutKEAP1-NFE2L2) and PET/CT, a first model (RNASeq) predicting the mutation status (MutRNASeq) using the 33 previously identified transcriptomic features was developed on patients from the TCGA-LUSC, TCGA-LUAD, CPTAC-LSCC and CPTAC-LUAD cohorts (770 patients) and externally validated on the NSCLC-Radiogenomics cohort (117 patients). Narrowing the patients to those with an available PET/CT, a second model (RNAPET) was then built and internally validated to predict the previously MutRNASeq probability using PET/CT-extracted radiomics features. The RNAPET model was then validated on an external cohort of 151 patients treated with curative radiotherapy for a localized non-small cell lung cancer (VMAT cohort). For each model, features were combined using a neural network approach (Multilayer Perceptron) within a statistical software modeler. Performances were evaluated based on the ROC-features as well as decision curve analysis.

RESULTS

The RNASeq model resulted in a C-Index of 0.82, Sensitivity (Se) of 70.3% and Specificity (Sp) of 93.4% in the validation cohort. Regarding the PET/CT-based prediction on a training cohort of 101 patients, the retained RNAPET model resulted in an AUC of 0.90 (p < 0.001). With a probability threshold of 20% and applied to the testing cohort, the RNAPET model achieved a C-Index of 0.7 with respective Se/Sp of 60.0% and 80.9% for the prediction of the MutRNASeq. The same radiomics model was validated on the VMAT cohort as patients were significantly stratified based on their risk of locoregional (LR) relapse with a hazard ratio of 2.61 (p = 0.02).

CONCLUSION

Our three-step approach enables the prediction of the MutKEAP1-NFE2L2 using PET/CT-extracted radiomics features and efficiently classified patient at risk of LR relapse in an external cohort treated with radiotherapy. This first evidence should be further evaluated on larger cohorts, and implemented in LR risk prediction models.

Joint EANM/SNMMI guideline on radiomics in nuclear medicine : Jointly supported by the EANM Physics Committee and the SNMMI Physics, Instrumentation and Data Sciences Council

PURPOSE

The purpose of this guideline is to provide comprehensive information on best practices for robust radiomics analyses for both hand-crafted and deep learning-based approaches.

METHODS

In a cooperative effort between the EANM and SNMMI, we agreed upon current best practices and recommendations for relevant aspects of radiomics analyses, including study design, quality assurance, data collection, impact of acquisition and reconstruction, detection and segmentation, feature standardization and implementation, as well as appropriate modelling schemes, model evaluation, and interpretation. We also offer an outlook for future perspectives.

CONCLUSION

Radiomics is a very quickly evolving field of research. The present guideline focused on established findings as well as recommendations based on the state of the art. Though this guideline recognizes both hand-crafted and deep learning-based radiomics approaches, it primarily focuses on the former as this field is more mature. This guideline will be updated once more studies and results have contributed to improved consensus regarding the application of deep learning methods for radiomics. Although methodological recommendations in the present document are valid for most medical image modalities, we focus here on nuclear medicine, and specific recommendations when necessary are made for PET/CT, PET/MR, and quantitative SPECT.

Predicting response to radiotherapy of head and neck squamous cell carcinoma using radiomics from cone-beam CT images

INTRODUCTION

Radiotherapy (RT) for head and neck cancer is now guided by cone-beam computed tomography (CBCT). We aim to identify a CBCT radiomic signature predictive of progression to RT.

MATERIAL AND METHODS

A cohort of 93 patients was split into training (n = 60) and testing (n = 33) sets. A total of 88 features were extracted from the gross tumor volume (GTV) on each CBCT. Receiver operating characteristic (ROC) curves were used to determine the power of each feature at each week of treatment to predict progression to radio(chemo)therapy. Only features with AUC > 0.65 at each week were pre-selected. Absolute differences were calculated between features from each weekly CBCT and baseline CBCT1 images. The smallest detectable change (C = 1.96 × SD, SD being the standard deviation of differences between feature values calculated on CBCT1 and CBCTn) with its confidence interval (95% confidence interval [CI]) was determined for each feature. The features for which the change was larger than C for at least 5% of patients were then selected. A radiomics-based model was built at the time-point that showed the highest AUC and compared with models relying on clinical variables.

RESULTS

Seven features had an AUC > 0.65 at each week, and six exhibited a change larger than the predefined CI 95%. After exclusion of inter-correlated features, only one parameter remains, Coarseness. Among clinical variable, only hemoglobin value was significant. AUC for predicting the treatment response were 0.78 (p = .006), 0.85 (p < .001), and 0.99 (p < .001) for clinical, CBCT4-radiomics (Coarseness) and clinical + radiomics based models respectively. The mean AUC of this last model on a 5-fold cross-validation was 0.80 (±0.09). On the testing cohort, the best prediction was given by the combined model (balanced accuracy [BAcc] 0.67 , p < .001).

CONCLUSIONS

We described a feature selection methodology for delta-radiomics that is able to select reproducible features which are informative due to their change during treatment. A selected delta radiomics feature may improve clinical-based prediction models.

Harmonization strategies for multicenter radiomics investigations

Carrying out large multicenter studies is one of the key goals to be achieved towards a faster transfer of the radiomics approach in the clinical setting. This requires large-scale radiomics data analysis, hence the need for integrating radiomic features extracted from images acquired in different centers. This is challenging as radiomic features exhibit variable sensitivity to differences in scanner model, acquisition protocols and reconstruction settings, which is similar to the so-called 'batch-effects' in genomics studies. In this review we discuss existing methods to perform data integration with the aid of reducing the unwanted variation associated with batch effects. We also discuss the future potential role of deep learning methods in providing solutions for addressing radiomic multicentre studies.

Radiomics analysis of 3D dose distributions to predict toxicity of radiotherapy for lung cancer

PURPOSE

(Chemo)-radiotherapy (RT) is the gold standard treatment for patients with locally advanced lung cancer non accessible for surgery. However, current toxicity prediction models rely on clinical and dose volume histograms (DVHs) and remain unsufficient. The goal of this work is to investigate the added predictive value of the radiomics approach applied to dose maps regarding acute and late toxicities in both the lungs and esophagus.

METHODS

Acute and late toxicities scored using the CTCAE v4.0 were retrospectively collected on patients treated with RT in our institution. Radiomic features were extracted from 3D dose maps considering Gy values as grey-levels in images. DVH and usual clinical factors were also considered. Three toxicity prediction models (clinical only, clinical + DVH and combined, i.e., including clinical + DVH + radiomics) were incrementally trained using a neural network on 70% of the patients for prediction of grade ≥2 acute and late pulmonary toxicities (APT/LPT) and grade ≥2 acute esophageal toxicities (AET). After bootstrapping (n = 1000), optimal cut-off values were determined based on the Youden Index. The trained models were then evaluated in the remaining 30% of patients using balanced accuracy (BAcc).

RESULTS

167 patients were treated from 2015 to 2018: 78% non small-cell lung cancers, 14% small-cell lung cancers and 8% other histology with a median age at treatment of 66 years. Respectively, 22.2%, 16.8% and 30.0% experienced APT, LPT and AET. In the training set (n = 117), the corresponding BAcc for clinical only/clinical + DVH/combined were 0.68/0.79/0.92, 0.66/0.77/0.87 and 0.68/0.73/0.84. In the testing evaluation (n = 50), these trained models obtained a corresponding BAcc of 0.69/0.69/0.92, 0.76/0.80/0.89 and 0.58/0.73/0.72.

CONCLUSION

In patients with a lung cancer treated with RT, radiomic features extracted from 3D dose maps seem to surpass usual models based on clinical factors and DVHs for the prediction of APT and LPT.

Performance comparison of modified ComBat for harmonization of radiomic features for multicenter studies

Multicenter studies are needed to demonstrate the clinical potential value of radiomics as a prognostic tool. However, variability in scanner models, acquisition protocols and reconstruction settings are unavoidable and radiomic features are notoriously sensitive to these factors, which hinders pooling them in a statistical analysis. A statistical harmonization method called ComBat was developed to deal with the "batch effect" in gene expression microarray data and was used in radiomics studies to deal with the "center-effect". Our goal was to evaluate modifications in ComBat allowing for more flexibility in choosing a reference and improving robustness of the estimation. Two modified ComBat versions were evaluated: M-ComBat allows to transform all features distributions to a chosen reference, instead of the overall mean, providing more flexibility. B-ComBat adds bootstrap and Monte Carlo for improved robustness in the estimation. BM-ComBat combines both modifications. The four versions were compared regarding their ability to harmonize features in a multicenter context in two different clinical datasets. The first contains 119 locally advanced cervical cancer patients from 3 centers, with magnetic resonance imaging and positron emission tomography imaging. In that case ComBat was applied with 3 labels corresponding to each center. The second one contains 98 locally advanced laryngeal cancer patients from 5 centers with contrast-enhanced computed tomography. In that specific case, because imaging settings were highly heterogeneous even within each of the five centers, unsupervised clustering was used to determine two labels for applying ComBat. The impact of each harmonization was evaluated through three different machine learning pipelines for the modelling step in predicting the clinical outcomes, across two performance metrics (balanced accuracy and Matthews correlation coefficient). Before harmonization, almost all radiomic features had significantly different distributions between labels. These differences were successfully removed with all ComBat versions. The predictive ability of the radiomic models was always improved with harmonization and the improved ComBat provided the best results. This was observed consistently in both datasets, through all machine learning pipelines and performance metrics. The proposed modifications allow for more flexibility and robustness in the estimation. They also slightly but consistently improve the predictive power of resulting radiomic models.

[Prognostic value of the metabolically active tumour volume]

PURPOSE

The purpose of this study was to assess the prognostic value of different parameters on pretreatment fluorodeoxyglucose [((18)F)-FDG] positron emission tomography-computed tomography (PET-CT) in patients with localized oesophageal cancer.

PATIENTS AND METHOD

We retrospectively reviewed 83 cases of localised oesophageal cancer treated in our institution. Patients were treated with curative intent and have received chemoradiotherapy alone or followed by surgery. Different prognostic parameters were correlated to survival: cancer-related factors, patient-related factors and parameters derived from PET-CT (maximum standardized uptake value [SUV max], metabolically active tumor volume either measured with an automatic segmentation software ["fuzzy locally adaptive bayesian": MATVFLAB] or with an adaptive threshold method [MATVseuil] and total lesion glycolysis [TLGFLAB and TLGseuil]).

RESULTS

The median follow-up was 21.8 months (range: 0.16-104). The median overall survival was 22 months (95% confidence interval [95%CI]: 15.2-28.9). There were 67 deaths: 49 associated with cancer and 18 from intercurrent causes. None of the tested factors was significant on overall survival. In univariate analysis, the following three factors affected the specific survival: MATVFLAB (P=0.025), TLGFLAB (P=0.04) and TLGseuil (P=0.04). In multivariate analysis, only MATVFLAB had a significant impact on specific survival (P=0.049): MATVFLAB<18 cm(3): 31.2 months (95%CI: 21.7-not reached) and MATVFLAB>18 cm(3): 20 months (95%CI: 11.1-228.9).

CONCLUSION

The metabolically active tumour volume measured with the automatic segmentation software FLAB on baseline PET-CT was a significant prognostic factor, which should be tested on a larger cohort.

Hypoxia imaging with [18F]-FMISO-PET for guided dose escalation with intensity-modulated radiotherapy in head-and-neck cancers

BACKGROUND AND PURPOSE

Positron emission tomography (PET) with [(18)F]-fluoromisonidazole ([(18)F]-FMISO) provides a non-invasive assessment of hypoxia. The aim of this study is to assess the feasibility of a dose escalation with volumetric modulated arc therapy (VMAT) guided by [(18)F]-FMISO-PET for head-and-neck cancers (HNC).

PATIENTS AND METHODS

Ten patients with inoperable stages III-IV HNC underwent [(18)F]-FMISO-PET before radiotherapy. Hypoxic target volumes (HTV) were segmented automatically by using the fuzzy locally adaptive Bayesian method. Retrospectively, two VMAT plans were generated delivering 70 Gy to the gross tumour volume (GTV) defined on computed tomography simulation or 79.8 Gy to the HTV. A dosimetric comparison was performed, based on calculations of tumour control probability (TCP), normal tissue complication probability (NTCP) for the parotid glands and uncomplicated tumour control probability (UTCP).

RESULTS

The mean hypoxic fraction, defined as the ratio between the HTV and the GTV, was 0.18. The mean average dose for both parotids was 22.7 Gy and 25.5 Gy without and with dose escalation respectively. FMISO-guided dose escalation led to a mean increase of TCP, NTCP for both parotids and UTCP by 18.1, 4.6 and 8% respectively.

CONCLUSION

A dose escalation up to 79.8 Gy guided by [(18)F]-FMISO-PET with VMAT seems feasible with improvement of TCP and without excessive increase of NTCP for parotids.

Potential of [18F]-fluoromisonidazole positron-emission tomography for radiotherapy planning in head and neck squamous cell carcinomas

BACKGROUND AND PURPOSE

Positron-emission tomography (PET) with [(18)F]-fluoromisonidazole (FMISO) permits consideration of radiotherapy dose escalation to hypoxic volumes in head and neck cancers (HNC). However, the definition of FMISO volumes remains problematic. The aims of this study are to confirm that delayed acquisition at 4 h is most appropriate for FMISO-PET imaging and to assess different methods of volume segmentation.

PATIENTS AND METHODS

A total of 15 HNC patients underwent several FMISO-PET/computed tomography (CT) acquisitions 2, 3 and 4 h after FMISO injection. Three automatic methods of PET image segmentation were tested: fixed threshold, adaptive threshold based on the ratio between tumour-derived and background activities (R(T/B)) and the fuzzy locally adaptive Bayesian (FLAB) method. The hypoxic fraction (HF), which is defined as the ratio between the FMISO and CT volumes, was also calculated.

RESULTS

The R(T/B) for images acquired at 2, 3 and 4 h differed significantly, with mean values of 2.5 (1.7-2.9), 3 (2-4.5) and 3.4 (2.3-6.1), respectively. The mean tumour volume, as defined manually using CT images, was 39.1 ml (1.2-116 ml). After 4 h, the mean FMISO volumes were 18.9 (0.1-81), 9.5 (0.9-33.1) and 12.5 ml (0.9-38.4 ml) with fixed threshold, adaptive threshold and the FLAB method, respectively; median HF values were 0.47 (0.1-1.93), 0.25 (0.11-0.75) and 0.35 (0.14-1.05), respectively. FMISO volumes were significantly different.

CONCLUSION

The best contrast is obtained at the 4-hour acquisition time. Large discrepancies were found between the three tested methods of volume segmentation.

HER2-overexpressing breast cancer: FDG uptake after two cycles of chemotherapy predicts the outcome of neoadjuvant treatment

Background:

Pathologic complete response (pCR) to neoadjuvant treatment (NAT) is associated with improved survival of patients with HER2+ breast cancer. We investigated the ability of interim positron emission tomography (PET) regarding early prediction of pathology outcomes.

Methods:

During 61 months, consecutive patients with locally advanced or large HER2+ breast cancer patients without distant metastases were included. All patients received NAT with four cycles of epirubicin+cyclophosphamide, followed by four cycles of docetaxel+trastuzumab. ¹⁸F-fluorodeoxyglucose (¹⁸F-FDG)-PET/computed tomography (CT) was performed at baseline (PET₁) and after two cycles of chemotherapy (PET₂). Maximum standardised uptake values were measured in the primary tumour as well as in the axillary lymph nodes. The correlation between pathologic response and SUV parameters (SUV_max at PET₁, PET₂ and ΔSUV_max) was examined with the t-test. The predictive performance regarding the identification of non-responders was evaluated using receiver operating characteristics (ROC) analysis.

Results:

Thirty women were prospectively included and 60 PET/CT examination performed. At baseline, 22 patients had PET+ axilla and in nine of them ¹⁸F-FDG uptake was higher than in the primary tumour. At surgery, 14 patients (47%) showed residual tumour (non-pCR), whereas 16 (53%) reached pCR. Best prediction was obtained when considering the absolute residual SUV_max value at PET₂ (AUC=0.91) vs 0.67 for SUV_max at PET₁ and 0.86 for ΔSUV_max. The risk of non-pCR was 92.3% in patients with any site of residual uptake >3 at PET₂, no matter whether in breast or axilla, vs 11.8% in patients with uptake ⩽3 (P=0.0001). The sensitivity, specificity, PPV, NPV and overall accuracy of this cutoff were, respectively: 85.7%, 93.8%, 92.3%, 88.2% and 90%.

Conclusion:

The level of residual ¹⁸F-FDG uptake after two cycles of chemotherapy predicts residual disease at completion of NAT with chemotherapy+trastuzumab with high accuracy. Because many innovative therapeutic strategies are now available (e.g., addition of a second HER2-directed therapy or an antiangiogenic), early prediction of poor response is critical.