Influence of inter-observer delineation variability on radiomics stability in different tumor sites

Radiomics for precision medicine: Current challenges, future prospects, and the proposal of a new framework

The advancement of artificial intelligence concurrent with the development of medical imaging techniques provided a unique opportunity to turn medical imaging from mostly qualitative, to further quantitative and mineable data that can be explored for the development of clinical decision support systems (cDSS). Radiomics, a method for the high throughput extraction of hand-crafted features from medical images, and deep learning -the data driven modeling techniques based on the principles of simplified brain neuron interactions, are the most researched quantitative imaging techniques. Many studies reported on the potential of such techniques in the context of cDSS. Such techniques could be highly appealing due to the reuse of existing data, automation of clinical workflows, minimal invasiveness, three-dimensional volumetric characterization, and the promise of high accuracy and reproducibility of results and cost-effectiveness. Nevertheless, there are several challenges that quantitative imaging techniques face, and need to be addressed before the translation to clinical use. These challenges include, but are not limited to, the explainability of the models, the reproducibility of the quantitative imaging features, and their sensitivity to variations in image acquisition and reconstruction parameters. In this narrative review, we report on the status of quantitative medical image analysis using radiomics and deep learning, the challenges the field is facing, propose a framework for robust radiomics analysis, and discuss future prospects.

Radiomics as a personalized medicine tool in lung cancer: Separating the hope from the hype

Radiomics has become a popular image analysis method in the last few years. Its key hypothesis is that medical images harbor biological, prognostic and predictive information that is not revealed upon visual inspection. In contrast to previous work with a priori defined imaging biomarkers, radiomics instead calculates image features at scale and uses statistical methods to identify those most strongly associated to outcome. This builds on years of research into computer aided diagnosis and pattern recognition. While the potential of radiomics to aid personalized medicine is widely recognized, several technical limitations exist which hinder biomarker translation. Aspects of the radiomic workflow lack repeatability or reproducibility under particular circumstances, which is a key requirement for the translation of imaging biomarkers into clinical practice. One of the most commonly studied uses of radiomics is for personalized medicine applications in Non-Small Cell Lung Cancer (NSCLC). In this review, we summarize reported methodological limitations in CT based radiomic analyses together with suggested solutions. We then evaluate the current NSCLC radiomics literature to assess the risk associated with accepting the published conclusions with respect to these limitations. We review different complementary scoring systems and initiatives that can be used to critically appraise data from radiomics studies. Wider awareness should improve the quality of ongoing and future radiomics studies and advance their potential as clinically relevant biomarkers for personalized medicine in patients with NSCLC.

Machine learning applications in imaging analysis for patients with pituitary tumors: a review of the current literature and future directions

PURPOSE

To provide an overview of fundamental concepts in machine learning (ML), review the literature on ML applications in imaging analysis of pituitary tumors for the last 10 years, and highlight the future directions on potential applications of ML for pituitary tumor patients.

METHOD

We presented an overview of the fundamental concepts in ML, its various stages used in healthcare, and highlighted the key components typically present in an imaging-based tumor analysis pipeline. A search was conducted across four databases (PubMed, Ovid, Embase, and Google Scholar) to gather research articles from the past 10 years (2009-2019) involving imaging related to pituitary tumor and ML. We grouped the studies by imaging modalities and analyzed the ML tasks in terms of the data inputs, reference standards, methodologies, and limitations.

RESULTS

Of the 16 studies included in our analysis, 10 appeared in 2018-2019. Most of the studies utilized retrospective data and followed a semi-automatic ML pipeline. The studies included use of magnetic resonance imaging (MRI), facial photographs, surgical microscopic video, spectrometry, and spectroscopy imaging. The objectives of the studies covered 14 distinct applications and majority of the studies addressed a binary classification problem. Only five of the 11 MRI-based studies had an external validation or a holdout set to test the performance of a final trained model.

CONCLUSION

Through our concise evaluation and comparison of the studies using the concepts presented, we highlight future directions so that potential ML applications using different imaging modalities can be developed to benefit the clinical care of pituitary tumor patients.

Current status and quality of radiomics studies in lymphoma: a systematic review

OBJECTIVES

To perform a systematic review regarding the developments in the field of radiomics in lymphoma. To evaluate the quality of included articles by the Quality Assessment of Diagnostic Accuracy Studies-2 (QUADAS-2), the phases classification criteria for image mining studies, and the radiomics quality scoring (RQS) tool.

METHODS

We searched for eligible articles in the MEDLINE/PubMed and EMBASE databases using the terms "radiomics", "texture" and "lymphoma". The included studies were divided into two categories: diagnosis-, therapy response- and outcome-related studies. The diagnosis-related studies were evaluated using the QUADAS-2; all studies were evaluated using the phases classification criteria for image mining studies and the RQS tool by two reviewers.

RESULTS

Forty-five studies were included; thirteen papers (28.9%) focused on the differential diagnosis of primary central nervous system lymphoma (PCNSL) and glioblastoma (GBM). Thirty-two (71.1%) studies were classified as discovery science according to the phase classification criteria for image mining studies. The mean RQS score of all studies was 14.2% (ranging from 0.0 to 40.3%), and 23 studies (51.1%) were given a score of < 10%.

CONCLUSION

The radiomics features could serve as diagnostic and prognostic indicators in lymphoma. However, the current conclusions should be interpreted with caution due to the suboptimal quality of the studies. In order to introduce radiomics into lymphoma clinical settings, the lesion segmentation and selection, the influence of the pathological pattern and the extraction of multiple modalities and multiple time points features need to be further studied.

KEY POINTS

• The radiomics approach may provide useful information for diagnosis, prediction of the therapy response, and outcome of lymphoma. • The quality of published radiomics studies in lymphoma has been suboptimal to date. • More studies are needed to examine lesion selection and segmentation, the influence of pathological patterns, and the extraction of multiple modalities and multiple time point features.

Deep learning-based radiomic features for improving neoadjuvant chemoradiation response prediction in locally advanced rectal cancer

Radiomic features achieve promising results in cancer diagnosis, treatment response prediction, and survival prediction. Our goal is to compare the handcrafted (explicitly designed) and deep learning (DL)-based radiomic features extracted from pre-treatment diffusion-weighted magnetic resonance images (DWIs) for predicting neoadjuvant chemoradiation treatment (nCRT) response in patients with locally advanced rectal cancer (LARC). 43 Patients receiving nCRT were included. All patients underwent DWIs before nCRT and total mesorectal excision surgery 6-12 weeks after completion of nCRT. Gross tumor volume (GTV) contours were drawn by an experienced radiation oncologist on DWIs. The patient-cohort was split into the responder group (n = 22) and the non-responder group (n = 21) based on the post-nCRT response assessed by postoperative pathology, MRI or colonoscopy. Handcrafted and DL-based features were extracted from the apparent diffusion coefficient (ADC) map of the DWI using conventional computer-aided diagnosis methods and a pre-trained convolution neural network, respectively. Least absolute shrinkage and selection operator (LASSO)-logistic regression models were constructed using extracted features for predicting treatment response. The model performance was evaluated with repeated 20 times stratified 4-fold cross-validation using receiver operating characteristic (ROC) curves and compared using the corrected paired t-test. The model built with handcrafted features achieved the mean area under the ROC curve (AUC) of 0.64, while the one built with DL-based features yielded the mean AUC of 0.73. The corrected paired t-test on AUC showed P-value < 0.05. DL-based features extracted from pre-treatment DWIs achieved significantly better classification performance compared with handcrafted features for predicting nCRT response in patients with LARC.

Deep learning-based segmentation of breast masses in dedicated breast CT imaging: Radiomic feature stability between radiologists and artificial intelligence

A deep learning (DL) network for 2D-based breast mass segmentation in unenhanced dedicated breast CT images was developed and validated, and its robustness in radiomic feature stability and diagnostic performance compared to manual annotations of multiple radiologists was investigated. 93 mass-like lesions were extensively augmented and used to train the network (n = 58 masses), which was then tested (n = 35 masses) against manual ground truth of a qualified breast radiologist with experience in breast CT imaging using the Conformity coefficient (with a value equal to 1 indicating a perfect performance). Stability and diagnostic power of 672 radiomic descriptors were investigated between the computerized segmentation, and 4 radiologists' annotations for the 35 test set cases. Feature stability and diagnostic performance in the discrimination between benign and malignant cases were quantified using intraclass correlation (ICC) and multivariate analysis of variance (MANOVA), performed for each segmentation case (4 radiologists and DL algorithm). DL-based segmentation resulted in a Conformity of 0.85 ± 0.06 against the annotated ground truth. For the stability analysis, although modest agreement was found among the four annotations performed by radiologists (Conformity 0.78 ± 0.03), over 90% of all radiomic features were found to be stable (ICC>0.75) across multiple segmentations. All MANOVA analyses were statistically significant (p ≤ 0.05), with all dimensions equal to 1, and Wilks' lambda ≤0.35. In conclusion, DL-based mass segmentation in dedicated breast CT images can achieve high segmentation performance, and demonstrated to provide stable radiomic descriptors with comparable discriminative power in the classification of benign and malignant tumors to expert radiologist annotation.

Stable and discriminating radiomic predictor of recurrence in early stage non-small cell lung cancer: Multi-site study

OBJECTIVES

To evaluate whether combining stability and discriminability criteria in building radiomic classifiers will improve the prognosis of cancer recurrence in early stage non-small cell lung cancer on non-contrast computer tomography (CT).

MATERIALS AND METHODS

CT scans of 610 patients with early stage (IA, IB, IIA) NSCLC from four independent cohorts were evaluated. A total of 350 patients from Cleveland Clinic Foundation and University of Pennsylvania were divided into two equal sets for training (D₁) and validation set (D₂). 80 patients from The Cancer Genome Atlas Lung Adenocarcinoma and Squamous Cell Carcinoma and 195 patients from The Cancer Imaging Archive, were used as independent second (D₃) and third (D₄) validation sets. A linear discriminant analysis (LDA) classifier was built based on the most stable and discriminate features. In addition, a radiomic risk score (RRS) was generated by using least absolute shrinkage and selection operator, Cox regression model to predict time to progression (TTP) following surgery.

RESULTS

A feature selection strategy focusing on both feature discriminability and stability resulted in the classifier having a higher discriminability on validation datasets compared to the discriminability alone criteria in discriminating cancer recurrence (D₂, AUC of 0.75 vs. 0.65; D₃, 0.74 vs. 0.62; D₄, 0.76 vs. 0.63). The RRS generated by most stable-discriminating features was significantly associated with TTP compared to discriminating alone criteria (HR = 1.66, C-index of 0.72 vs. HR = 1.04, C-index of 0.62).

CONCLUSION

Accounting for both stability and discriminability yielded a more generalizable classifier for predicting cancer recurrence and TTP in early stage NSCLC.

[18F] FDG Positron Emission Tomography (PET) Tumor and Penumbra Imaging Features Predict Recurrence in Non-Small Cell Lung Cancer

We identified computational imaging features on 18F-fluorodeoxyglucose positron emission tomography (PET) that predict recurrence/progression in non–small cell lung cancer (NSCLC). We retrospectively identified 291 patients with NSCLC from 2 prospectively acquired cohorts (training, n = 145; validation, n = 146). We contoured the metabolic tumor volume (MTV) on all pretreatment PET images and added a 3-dimensional penumbra region that extended outward 1 cm from the tumor surface. We generated 512 radiomics features, selected 435 features based on robustness to contour variations, and then applied randomized sparse regression (LASSO) to identify features that predicted time to recurrence in the training cohort. We built Cox proportional hazards models in the training cohort and independently evaluated the models in the validation cohort. Two features including stage and a MTV plus penumbra texture feature were selected by LASSO. Both features were significant univariate predictors, with stage being the best predictor (hazard ratio [HR] = 2.15 [95% confidence interval (CI): 1.56–2.95], P < .001). However, adding the MTV plus penumbra texture feature to stage significantly improved prediction (P = .006). This multivariate model was a significant predictor of time to recurrence in the training cohort (concordance = 0.74 [95% CI: 0.66–0.81], P < .001) that was validated in a separate validation cohort (concordance = 0.74 [95% CI: 0.67–0.81], P < .001). A combined radiomics and clinical model improved NSCLC recurrence prediction. FDG PET radiomic features may be useful biomarkers for lung cancer prognosis and add clinical utility for risk stratification.

Impact of contouring variability on oncological PET radiomics features in the lung

Radiomics features extracted from oncological PET images are currently under intense scrutiny within the context of risk stratification for a variety of cancers. However, the lack of robustness assessment poses problems for their application across institutions and for broader patient populations. The objective of the current study was to examine the extent to which radiomics parameters from oncological PET vary in response to manual contouring variability in lung cancer. Imaging data employed in the study consisted of 26 PET scans with lesions in the lung being created through the use of an anthropomorphic phantom in conjunction with Monte Carlo simulations. From each of the simulated lesions, 25 radiomics features related to the gray-level co-occurrence matrices (GLCOM), gray-level size zone matrices (GLSZM), and gray-level neighborhood difference matrices (GLNDM) were extracted from ground truth contour and from manual contours provided by 10 raters in regard to four intensity discretization schemes with number of gray levels of 32, 64, 128, and 256, respectively. The impact of interrater variability in tumor delineation upon the agreement between raters on radiomics features was examined via interclass correlation and leave-p-out assessment. Only weak and moderate correlations were found between segmentation accuracy as measured by the Dice coefficient and percent feature error from ground truth for the vast majority of the features being examined. GLNDM-based texture parameters emerged as the top performing category of radiomcs features in terms of robustness against contouring variability for discretization schemes engaging number of gray levels of 32, 64, and 128 while GLCOM-based parameters stood out for discretization scheme engaging 256 gray levels. How and to what extent interrater reliability of radiomics features vary in response to the number of raters were largely feature-dependent. It was concluded that impact of contouring variability on PET-based radiomics features is present to varying degrees and could be experienced as a barrier to convey PET-based radiomics research to clinical relevance.

The effects of volume of interest delineation on MRI-based radiomics analysis: evaluation with two disease groups

BACKGROUND

Manual delineation of volume of interest (VOI) is widely used in current radiomics analysis, suffering from high variability. The tolerance of delineation differences and possible influence on each step of radiomics analysis are not clear, requiring quantitative assessment. The purpose of our study was to investigate the effects of delineation of VOIs on radiomics analysis for the preoperative prediction of metastasis in nasopharyngeal carcinoma (NPC) and sentinel lymph node (SLN) metastasis in breast cancer.

METHODS

This study retrospectively enrolled two datasets (NPC group: 238 cases; SLN group: 146 cases). Three operations, namely, erosion, smoothing, and dilation, were implemented on the VOIs accurately delineated by radiologists to generate diverse VOI variations. Then, we extracted 2068 radiomics features and evaluated the effects of VOI differences on feature values by the intra-class correlation coefficient (ICC). Feature selection was conducted by Maximum Relevance Minimum Redundancy combined with 0.632+ bootstrap algorithms. The prediction performance of radiomics models with random forest classifier were tested on an independent validation cohort by the area under the receive operating characteristic curve (AUC).

RESULTS

The larger the VOIs changed, the fewer features with high ICCs. Under any variation, SLN group showed fewer features with ICC ≥ 0.9 compared with NPC group. Not more than 15% top-predictive features identical to the accurate VOIs were observed across feature selection. The differences of AUCs of models derived from VOIs across smoothing or dilation with 3 pixels were not statistically significant compared with the accurate VOIs (p > 0.05) except for T2-weighted fat suppression images (smoothing: 0.845 vs. 0.725, p = 0.001; dilation: 0.800 vs. 0.725, p = 0.042). Dilation with 5 and 7 pixels contributed to remarkable AUCs in SLN group but the opposite in NPC group. The radiomics models did not perform well when tested by data from other delineations.

CONCLUSIONS

Differences in delineation of VOIs affected radiomics analysis, related to specific disease and MRI sequences. Differences from smooth delineation or expansion with 3 pixels width around the tumors or lesions were acceptable. The delineation for radiomics analysis should follow a predefined and unified standard.

Homological radiomics analysis for prognostic prediction in lung cancer patients

PURPOSE

This study explored a novel homological analysis method for prognostic prediction in lung cancer patients.

MATERIALS AND METHODS

The potential of homology-based radiomic features (HFs) was investigated by comparing HFs to conventional wavelet-based radiomic features (WFs) and combined radiomic features consisting of HFs and WFs (HWFs), using training (n = 135) and validation (n = 70) datasets, and Kaplan-Meier analysis. A total of 13,824 HFs were derived through homology-based texture analysis using Betti numbers, which represent the topologically invariant morphological characteristics of lung cancer. The prognostic potential of HFs was evaluated using statistically significant differences (p-values, log-rank test) to compare the survival curves of high- and low-risk patients. Those patients were stratified into high- and low-risk groups using the medians of the radiomic scores of signatures constructed with an elastic-net-regularized Cox proportional hazard model. Furthermore, deep learning (DL) based on AlexNet was utilized to compare HFs by stratifying patients into the two groups using a network that was pre-trained with over one million natural images from an ImageNet database.

RESULTS

For the training dataset, the p-values between the two survival curves were 6.7 × 10^-6 (HF), 5.9 × 10^-3 (WF), 7.4 × 10^-6 (HWF), and 1.1 × 10^-3 (DL). The p-values for the validation dataset were 3.4 × 10^-5 (HF), 6.7 × 10^-1 (WF), 1.7 × 10^-7 (HWF), and 1.2 × 10^-1 (DL).

CONCLUSION

This study demonstrates the excellent potential of HFs for prognostic prediction in lung cancer patients.

Diagnostic performance and inter-operator variability of apparent diffusion coefficient analysis for differentiating pleomorphic adenoma and carcinoma ex pleomorphic adenoma: comparing one-point measurement and whole-tumor measurement including radiomics approach

BACKGROUND AND PURPOSE

The purpose of this study was to compare the diagnostic performance between apparent diffusion coefficient (ADC) analysis of one-point measurement and whole-tumor measurement, including radiomics for differentiating pleomorphic adenoma (PA) from carcinoma ex pleomorphic adenoma (CXPA), and to evaluate the impact of inter-operator segmentation variability.

MATERIALS AND METHODS

One hundred and fifteen patients with PA and 22 with CXPA were included. Four radiologists with different experience independently placed one-point and whole-tumor ROIs and a radiomics-predictive model was constructed from the extracted imaging features. We calculated the area under the receiver-operator characteristic curve (AUC) for the diagnostic performance of imaging features and the radiomics-predictive model.

RESULTS

AUCs of the imaging features from whole-tumor varied between readers (0.50-0.89). The most experienced radiologist (Reader 1) produced significantly high AUCs than less experienced radiologists (Reader 3 and 4; P = 0.01 and 0.009). AUCs were higher for the radiomics-predictive model (0.82-0.87) than for one-point (0.66-0.79) in all readers.

CONCLUSION

Some imaging features of whole-tumor and radiomics-predictive model had higher diagnostic performance than one-point. The diagnostic performance of imaging features from whole-tumor alone varied depending on operator experience. Operator experience appears less likely to affect diagnostic performance in the radiomics-predictive model.

Reproducibility of CT Radiomic Features within the Same Patient: Influence of Radiation Dose and CT Reconstruction Settings

Background Results of recent phantom studies show that variation in CT acquisition parameters and reconstruction techniques may make radiomic features largely nonreproduceable and of limited use for prognostic clinical studies. Purpose To investigate the effect of CT radiation dose and reconstruction settings on the reproducibility of radiomic features, as well as to identify correction factors for mitigating these sources of variability. Materials and Methods This was a secondary analysis of a prospective study of metastatic liver lesions in patients who underwent staging with single-energy dual-source contrast material-enhanced staging CT between September 2011 and April 2012. Technique parameters were altered, resulting in 28 CT data sets per patient that included different dose levels, section thicknesses, kernels, and reconstruction algorithm settings. By using a training data set (n = 76), reproducible intensity, shape, and texture radiomic features (reproducibility threshold, R2 ≥ 0.95) were selected and correction factors were calculated by using a linear model to convert each radiomic feature to its estimated value in a reference technique. By using a test data set (n = 75), the reproducibility of hierarchical clustering based on 106 radiomic features measured with different CT techniques was assessed. Results Data in 78 patients (mean age, 60 years ± 10; 33 women) with 151 liver lesions were included. The percentage of radiomic features deemed reproducible for any variation of the different technical parameters was 11% (12 of 106). Of all technical parameters, reconstructed section thickness had the largest impact on the reproducibility of radiomic features (12.3% [13 of 106]) if only one technical parameter was changed while all other technical parameters were kept constant. The results of the hierarchical cluster analysis showed improved clustering reproducibility when reproducible radiomic features with dedicated correction factors were used (ρ = 0.39-0.71 vs ρ = 0.14-0.47). Conclusion Most radiomic features are highly affected by CT acquisition and reconstruction settings, to the point of being nonreproducible. Selecting reproducible radiomic features along with study-specific correction factors offers improved clustering reproducibility. © RSNA, 2019 Online supplemental material is available for this article. See also the editorial by Sosna in this issue.

Discovery of stable and prognostic CT-based radiomic features independent of contrast administration and dimensionality in oesophageal cancer

The aim of this work was to investigate radiomic analysis of contrast and non-contrast enhanced planning CT images of oesophageal cancer (OC) patients in terms of stability, dimensionality and contrast agent dependency. The prognostic significance of CT-based radiomic features was also evaluated. Different 2D and 3D radiomic features were extracted from contrast and non-contrast enhanced CT images of 213 patients from the multi-centre SCOPE1 randomised controlled trial (RCT) in OC. Feature stability was evaluated by randomly dividing patients into three groups and identifying textures with similar distributions among groups with a Kruskal-Wallis analysis. A paired two-sided Wilcoxon signed rank test was used to assess for significant differences in the remaining corresponding 2D and 3D stable features. A prognostic model was constructed using clinical characteristics and remaining filtered features. The discriminative ability of significant variables was tested using Kaplan-Meier analysis. A total of 238 2D and 3D radiomic features were computed from oesophageal CT images. More than 75 features were stable if extracted from homogeneous cohort (contrast or non-contrast enhanced CT images) and inhomogeneous cohort (contrast and non-contrast enhanced CT images). Among the remaining corresponding stable features computed from both cohorts, only 4 features did not show a statistically significant difference if obtained in 2D or in 3D (p-value < 0.05). A Cox regression model constructed using 5 clinical variables (age, sex, tumour, node and metastasis (TNM) stage, WHO performance status and contrast administration) and 4 radiomic variables (inverse varianceGLCM, large distance emphasisGLDZM, zone distance non uniformity normGLDZM, zone distance varianceGLDZM), identified one radiomic feature (zone distance varianceGLDZM) that was significantly associated with overall survival (p-value = 0.032, HR = 1.25, 95% CI = 1.02-1.52). A significant difference in overall survival between groups was found when considering a threshold of zone distance varianceGLDZM equals to 1.70 (X2 = 7.692, df = 1, p-value = 0.006). Zone distance varianceGLDZM was identified as the only stable CT radiomic feature statistically correlated with overall survival, independent of dimensionality and contrast administration. This feature was able to identify high-risk patients and if validated, could be the subject of a future clinical trial aiming to improve clinical decision making and personalise OC treatment.

Imaging-Based Individualized Response Prediction Of Carbon Ion Radiotherapy For Prostate Cancer Patients

Purpose

To explore the value of the pre-treatment MRI radiomic features in individualized prediction of the therapeutic response of carbon ion radiotherapy (CIRT) for prostate cancer patients.

Patients and methods

Twenty-three patients with localized prostate cancer treated by CIRT were enrolled for analysis. Prostate tumors were manually delineated on T2-weighted (T2w) images and apparent diffusion coefficient (ADC) maps acquired before CIRT. Abundant radiomic features were extracted from the delineations, which were randomly deformed to account for delineation uncertainty. The robust features were selected and then compared between patient groups of different CIRT responses. Support vector machine (SVM) was subsequently applied to demonstrate the role of the radiomic features to predict individualized CIRT response in the way of artificial intelligence.

Results

Radiomic features from ADC had significantly higher intra-correlation coefficient (ICC) (0.71±0.28) than T2w features (0.60±0.31) (p<0.01), indicating higher robustness of ADC features against delineation uncertainty. More features were excellently robust in ADC (58.2% of all the radiomic feature candidates, compared to 41.3% in T2w). By combining the excellently robust radiomic features of T2w and ADC, SVM achieved high performance to predict individualized therapeutic response of CIRT, ie, area-under-curve (AUC) = 0.88.

Conclusion

Radiomic features extracted from T2w and ADC images displayed great robustness to quantify the tumor characteristics of prostate cancer and high accuracy to predict the individualized therapeutic response of CIRT. After further validation, the selected radiomic features may become potential imaging biomarkers in the management of prostate cancer through CIRT.

Combined CT radiomics of primary tumor and metastatic lymph nodes improves prediction of loco-regional control in head and neck cancer

Loco-regional control (LRC) is a major clinical endpoint after definitive radiochemotherapy (RCT) of head and neck cancer (HNC). Radiomics has been shown a promising biomarker in cancer research, however closer related to primary tumor control than composite endpoints. Radiomics studies often focus on the analysis of primary tumor (PT). We hypothesize that the combination of PT and lymph nodes (LN) radiomics better predicts LRC in HNC treated with RCT. Radiomics analysis was performed in CT images of 128 patients using Z-Rad implementation (training n = 77, validation n = 51). 285 features were extracted from PT and involved LN. Features were preselected with the maximum relevance minimum redundancy method and the multivariate Cox model was trained using least absolute shrinkage and selection operator. The mixed model was based on the combination of PT and LN radiomics, whereas the PT model included only the PT features. The mixed model showed significantly higher performance than the PT model (p < 0.01), c-index of 0.67 and 0.63, respectively; and better risk group stratification. The clinical nodal status was not a significant predictor in the combination with PT radiomics. This study shows that the LRC can be better predicted by expansion of radiomics analysis with LN features.

Early variation of 18-fluorine-labelled fluorodeoxyglucose PET-derived parameters after chemoradiotherapy as predictors of survival in locally advanced pancreatic carcinoma patients

OBJECTIVE

To investigate if early variation of PET-derived parameters after concomitant chemoradiotherapy (CRT) predicts overall survival (OS), local relapse free survival (LRFS), distant relapse free survival (DRFS) and progression free survival (PFS) in locally advanced pancreatic cancer (LAPC) patients.

METHODS

Fifty-two LAPC patients (median age: 61 years; range: 35-85) with available FDG PET/CT before and after RT (2-6 months, median: 2) were enrolled from May 2005 to June 2015. The predictive value of the percentage variation of mean/maximum standard uptake value (ΔSUVmean/max), metabolic tumour volume (ΔMTV) and total lesion glycolysis (ΔTLG), estimated considering different uptake thresholds (40-50-60%), was investigated between pre- and post-RT PET. The percentage difference between gastrointestinal cancer-associated antigen (ΔGICA) levels measured at the time of PET was also considered. Log-rank test and Cox regression analysis were performed to assess the prognostic value of considered PET-derived parameters on survival outcomes.

RESULTS

The median follow-up was 13 months (range: 4-130). At univariate analysis, ΔTLG50 showed borderline significance in predicting OS (P = 0.05) and was the most significant parameter correlated to LRFS and PFS (P = 0.001). Median LRFS was 4 and 33 months if ΔTLG50 was below or above 35% respectively (P = 0.0003); similarly, median PFS was 3 vs 6 months (P = 0.0009). No significant correlation was found between PET-derived parameters and DRFS, while the ΔGICA was the only borderline significant prognostic value for this endpoint (P = 0.05).

CONCLUSION

PET-derived parameters predict survival in LAPC patients; in particular, ΔTLG50 is the strongest predictor. The combination of these biochemical and imaging biomarkers is promising in identifying patients at higher risk of earlier relapse.

Bone Marrow and Tumor Radiomics at 18F-FDG PET/CT: Impact on Outcome Prediction in Non-Small Cell Lung Cancer

Background Primary tumor maximum standardized uptake value is a prognostic marker for non-small cell lung cancer. In the setting of malignancy, bone marrow activity from fluorine 18-fluorodeoxyglucose (FDG) PET may be informative for clinical risk stratification. Purpose To determine whether integrating FDG PET radiomic features of the primary tumor, tumor penumbra, and bone marrow identifies lung cancer disease-free survival more accurately than clinical features alone. Materials and Methods Patients were retrospectively analyzed from two distinct cohorts collected between 2008 and 2016. Each tumor, its surrounding penumbra, and bone marrow from the L3-L5 vertebral bodies was contoured on pretreatment FDG PET/CT images. There were 156 bone marrow and 512 tumor and penumbra radiomic features computed from the PET series. Randomized sparse Cox regression by least absolute shrinkage and selection operator identified features that predicted disease-free survival in the training cohort. Cox proportional hazards models were built and locked in the training cohort, then evaluated in an independent cohort for temporal validation. Results There were 227 patients analyzed; 136 for training (mean age, 69 years ± 9 [standard deviation]; 101 men) and 91 for temporal validation (mean age, 72 years ± 10; 91 men). The top clinical model included stage; adding tumor region features alone improved outcome prediction (log likelihood, -158 vs -152; P = .007). Adding bone marrow features continued to improve performance (log likelihood, -158 vs -145; P = .001). The top model integrated stage, two bone marrow texture features, one tumor with penumbra texture feature, and two penumbra texture features (concordance, 0.78; 95% confidence interval: 0.70, 0.85; P < .001). This fully integrated model was a predictor of poor outcome in the independent cohort (concordance, 0.72; 95% confidence interval: 0.64, 0.80; P < .001) and a binary score stratified patients into high and low risk of poor outcome (P < .001). Conclusion A model that includes pretreatment fluorine 18-fluorodeoxyglucose PET texture features from the primary tumor, tumor penumbra, and bone marrow predicts disease-free survival of patients with non-small cell lung cancer more accurately than clinical features alone. © RSNA, 2019 Online supplemental material is available for this article.

CT radiomics and PET radiomics: ready for clinical implementation?

INTRODUCTION

Today, rapid technical and clinical developments result in an increasing number of treatment options for oncological diseases. Thus, decision support systems are needed to offer the right treatment to the right patient. Imaging biomarkers hold great promise in patient-individual treatment guidance. Routinely performed for diagnosis and staging, imaging datasets are expected to hold more information than used in the clinical practice. Radiomics describes the extraction of a large number of meaningful quantitative features from medical images, such as computed tomography (CT) and positron emission tomography (PET). Due to the non-invasive nature and ability to capture 3D image-based heterogeneity, radiomic features are potential surrogate markers of the cancer phenotype. Several radiomic studies are published per day, owing to encouraging results of many radiomics-based patient outcome models. Despite this comparably large number of studies, radiomics is mainly studied in proof of principle concept. Hence, a translation of radiomics from a hot topic research field into an essential clinical decision-making tool is lacking, but of high clinical interest.

EVIDENCE ACQUISITION

Herein, we present a literature review addressing the clinical evidence of CT and PET radiomics. An extensive literature review was conducted in PubMed, including papers on robustness and clinical applications.

EVIDENCE SYNTHESIS

We summarize image-modality related influences on the robustness of radiomic features and provide an overview of clinical evidence reported in the literature. Today, more evidence has been provided for CT imaging, however, PET imaging offers the promise of direct imaging of biological processes and functions. We provide a summary of future research directions, which needs to be addressed in order to successfully introduce radiomics into clinical medicine. In comparison to CT, more focus should be directed towards harmonization of PET acquisition and reconstruction protocols, which is important for transferable modelling.

CONCLUSIONS

Both CT and PET radiomics are promising pre-treatment and intra-treatment biomarkers for outcome prediction. Most studies are performed in retrospective setting, however their validation in prospective data collections is ongoing.

Is tumour sphericity an important prognostic factor in patients with lung cancer?

BACKGROUND AND PURPOSE

Quantitative tumour shape features extracted from radiotherapy planning scans have shown potential as prognostic markers. In this study, we investigated if sphericity of the gross tumour volume (GTV) on planning computed tomography (CT) is an independent predictor of overall survival (OS) in lung cancer patients treated with standard radiotherapy. In the analysis, we considered whether tumour sphericity is correlated with clinical prognostic factors or influenced by the inclusion of lymph nodes in the GTV.

MATERIALS AND METHODS

Sphericity of single GTV delineation was extracted for 457 lung cancer patients. Relationships between sphericity, and common patient and tumour characteristics were investigated via correlation analysis and multivariate Cox regression to assess prognostic value of GTV sphericity. A subset analysis was performed for 290 nodal stage N0 patients to determine prognostic value of primary tumour sphericity.

RESULTS

Sphericity is correlated with clinical variables: tumour volume, mean lung dose, N stage, and T stage. Sphericity is strongly associated with OS (p < 0.001, hazard ratio (HR) (95% confidence interval (CI)) = 0.13 (0.04-0.41)) in univariate analysis. However, this association did not remain significant in multivariate analysis (p = 0.826, HR (95% CI) = 0.83 (0.16-4.31), and inclusion of sphericity to a clinical model did not improve model performance. In addition, no significant relationship between sphericity and OS was detected in univariate (p = 0.072) or multivariate (p = 0.920) analysis of N0 subset.

CONCLUSION

Sphericity correlates clearly with clinical prognostic factors, which are often unaccounted for in radiomic studies. Sphericity is also influenced by the presence of nodal involvement within the GTV contour.

Radiomic feature reproducibility in contrast-enhanced CT of the pancreas is affected by variabilities in scan parameters and manual segmentation

OBJECTIVES

This study aims to measure the reproducibility of radiomic features in pancreatic parenchyma and ductal adenocarcinomas (PDAC) in patients who underwent consecutive contrast-enhanced computed tomography (CECT) scans.

METHODS

In this IRB-approved and HIPAA-compliant retrospective study, 37 pairs of scans from 37 unique patients who underwent CECTs within a 2-week interval were included in the analysis of the reproducibility of features derived from pancreatic parenchyma, and a subset of 18 pairs of scans were further analyzed for the reproducibility of features derived from PDAC. In each patient, pancreatic parenchyma and pancreatic tumor (when present) were manually segmented by two radiologists independently. A total of 266 radiomic features were extracted from the pancreatic parenchyma and tumor region and also the volume and diameter of the tumor. The concordance correlation coefficient (CCC) was calculated to assess feature reproducibility for each patient in three scenarios: (1) different radiologists, same CECT; (2) same radiologist, different CECTs; and (3) different radiologists, different CECTs.

RESULTS

Among pancreatic parenchyma-derived features, using a threshold of CCC > 0.90, 58/266 (21.8%) and 48/266 (18.1%) features met the threshold for scenario 1, 14/266 (5.3%) and 15/266 (5.6%) for scenario 2, and 14/266 (5.3%) and 10/266 (3.8%) for scenario 3. Among pancreatic tumor-derived features, 11/268 (4.1%) and 17/268 (6.3%) features met the threshold for scenario 1, 1/268 (0.4%) and 5/268 (1.9%) features met the threshold for scenario 2, and no features for scenario 3 met the threshold, respectively.

CONCLUSIONS

Variations between CECT scans affected radiomic feature reproducibility to a greater extent than variation in segmentation. A smaller number of pancreatic tumor-derived radiomic features were reproducible compared with pancreatic parenchyma-derived radiomic features under the same conditions.

KEY POINTS

• For pancreatic-derived radiomic features from contrast-enhanced CT (CECT), fewer than 25% are reproducible (with a threshold of CCC < 0.9) in a clinical heterogeneous dataset. • Variations between CECT scans affected the number of reproducible radiomic features to a greater extent than variations in radiologist segmentation. • A smaller number of pancreatic tumor-derived radiomic features were reproducible compared with pancreatic parenchyma-derived radiomic features under the same conditions.

Predicting acute radiation induced xerostomia in head and neck Cancer using MR and CT Radiomics of parotid and submandibular glands

PURPOSE

To analyze baseline CT/MR-based image features of salivary glands to predict radiation-induced xerostomia 3-months after head-and-neck cancer (HNC) radiotherapy.

METHODS

A retrospective analysis was performed on 266 HNC patients who were treated using radiotherapy at our institution between 2009 and 2018. CT and T1 post-contrast MR images along with NCI-CTCAE xerostomia grade (3-month follow-up) were prospectively collected at our institution. CT and MR images were registered on which parotid/submandibular glands were contoured. Image features were extracted for ipsilateral/contralateral parotid and submandibular glands relative to the location of the primary tumor. Dose-volume-histogram (DVH) parameters were also acquired. Features were pre-selected based on Spearman correlation before modelling by examining the correlation with xerostomia (p < 0.05). A shrinkage regression analysis of the pre-selected features was performed using LASSO. The internal validity of the variable selection was estimated by repeating the entire variable selection procedure using a leave-one-out-cross-validation. The most frequently selected variables were considered in the final model. A generalized linear regression with repeated ten-fold cross-validation was developed to predict radiation-induced xerostomia at 3-months after radiotherapy. This model was tested in an independent dataset (n = 50) of patients who were treated at the same institution in 2017-2018. We compared the prediction performances under eight conditions (DVH-only, CT-only, MR-only, CT + MR, DVH + CT, DVH + CT + MR, Clinical+CT + MR, and Clinical+DVH + CT + MR) using the area under the receiver operating characteristic curve (ROC-AUC).

RESULTS

Among extracted features, 7 CT, 5 MR, and 2 DVH features were selected. The internal cohort (n = 216) ROC-AUC values for DVH, CT, MR, and Clinical+DVH + CT + MR features were 0.73 ± 0.01, 0.69 ± 0.01, 0.70 ± 0.01, and 0.79 ± 0.01, respectively. The validation cohort (n = 50) ROC-AUC values for DVH, CT, MR, and Clinical+DVH + CT + MR features were 0.63, 0.57, 0.66, and 0.68, respectively. The DVH-ROC was not significantly different than the CT-ROC (p = 0.8) or MR-ROC (p = 0.4). However, the CT + MR-ROC was significantly different than the CT-ROC (p = 0.03), but not the Clinical+DVH + CT + MR model (p = 0.5).

CONCLUSION

Our results suggest that baseline CT and MR image features may reflect baseline salivary gland function and potential risk for radiation injury. The integration of baseline image features into prediction models has the potential to improve xerostomia risk stratification with the ultimate goal of truly personalized HNC radiotherapy.

Reproducibility and Generalizability in Radiomics Modeling: Possible Strategies in Radiologic and Statistical Perspectives

Radiomics, which involves the use of high-dimensional quantitative imaging features for predictive purposes, is a powerful tool for developing and testing medical hypotheses. Radiologic and statistical challenges in radiomics include those related to the reproducibility of imaging data, control of overfitting due to high dimensionality, and the generalizability of modeling. The aims of this review article are to clarify the distinctions between radiomics features and other omics and imaging data, to describe the challenges and potential strategies in reproducibility and feature selection, and to reveal the epidemiological background of modeling, thereby facilitating and promoting more reproducible and generalizable radiomics research.

Predicting survival and local control after radiochemotherapy in locally advanced head and neck cancer by means of computed tomography based radiomics

PURPOSE

To appraise the ability of a radiomics signature to predict clinical outcome after definitive radiochemotherapy (RCT) of stage III-IV head and neck cancer.

METHODS

A cohort of 110 patients was included in a retrospective analysis. Radiomics texture features were extracted from the gross tumor volumes contoured on planning computed tomography (CT) images. The cohort of patients was randomly divided into a training (70 patients) and a validation (40 patients) cohorts. Textural features were correlated to survival and control data to build predictive models. All the significant predictors of the univariate analysis were included in a multivariate model. The quality of the models was appraised by means of the concordance index (CI).

RESULTS

A signature with 3 features was identified as predictive of overall survival (OS) with CI = 0.88 and 0.90 for the training and validation cohorts, respectively. A signature with 2 features was identified for progression-free survival (PFS; CI = 0.72 and 0.80); 2 features also characterized the signature for local control (LC; CI = 0.72 and 0.82). In all cases, the stratification in high- and low-risk groups for the training and validation cohorts led to significant differences in the actuarial curves. In the validation cohort the mean OS times (in months) were 78.9 ± 2.1 vs 67.4 ± 6.0 in the low- and high-risk groups, respectively, the PFS was 73.1 ± 3.7 and 50.7 ± 7.2, while the LC was 78.7 ± 2.1 and 63.9 ± 6.5.

CONCLUSION

CT-based radiomic signatures that correlate with survival and control after RCT were identified and allow low- and high-risk groups of patients to be identified.

CT Texture Analysis-Correlations With Histopathology Parameters in Head and Neck Squamous Cell Carcinomas

Introduction: Texture analysis is an emergent imaging technique to quantify heterogeneity in radiological images. It is still unclear whether this technique is capable to reflect tumor microstructure. The present study sought to correlate histopathology parameters with texture features derived from contrast-enhanced CT images in head and neck squamous cell carcinomas (HNSCC).

Materials and Methods: Twenty-eight patients with histopathological proven HNSCC were retrospectively analyzed. In every case EGFR, VEGF, Hif1-alpha, Ki67, p53 expression derived from immunhistochemical specimen were semiautomatically calculated. Furthermore, mean cell count was estimated. Texture analysis was performed on contrast-enhanced CT images as a whole lesion measurement. Spearman's correlation analysis was performed, adjusted with Benjamini-Hochberg correction for multiple tests.

Results: Several texture features correlated with histopathological parameters. After correction only CT texture joint entropy and CT entropy correlation with Hif1-alpha expression remained statistically significant (ρ = −0.60 and ρ = −0.50, respectively).

Conclusions: CT texture joint entropy and CT entropy were associated with Hif1-alpha expression in HNSCC and might be able to reflect hypoxic areas in this entity.

Reliability of Single-Slice-Based 2D CT Texture Analysis of Renal Masses: Influence of Intra- and Interobserver Manual Segmentation Variability on Radiomic Feature Reproducibility

OBJECTIVE. The objective of our study was to investigate the potential influence of intra- and interobserver manual segmentation variability on the reliability of single-slice-based 2D CT texture analysis of renal masses. MATERIALS AND METHODS. For this retrospective study, 30 patients with clear cell renal cell carcinoma were included from a public database. For intra- and interobserver analyses, three radiologists with varying degrees of experience segmented the tumors from unenhanced CT and corticomedullary phase contrast-enhanced CT (CECT) in different sessions. Each radiologist was blind to the image slices selected by other radiologists and him- or herself in the previous session. A total of 744 texture features were extracted from original, filtered, and transformed images. The intraclass correlation coefficient was used for reliability analysis. RESULTS. In the intraobserver analysis, the rates of features with good to excellent reliability were 84.4-92.2% for unenhanced CT and 85.5-93.1% for CECT. Considering the mean rates of unenhanced CT and CECT, having high experience resulted in better reliability rates in terms of the intraobserver analysis. In the interobserver analysis, the rates were 76.7% for unenhanced CT and 84.9% for CECT. The gray-level cooccurrence matrix and first-order feature groups yielded higher good to excellent reliability rates on both unenhanced CT and CECT. Filtered and transformed images resulted in more features with good to excellent reliability than the original images did on both unenhanced CT and CECT. CONCLUSION. Single-slice-based 2D CT texture analysis of renal masses is sensitive to intra- and interobserver manual segmentation variability. Therefore, it may lead to nonreproducible results in radiomic analysis unless a reliability analysis is considered in the workflow.

Assessing robustness of radiomic features by image perturbation

Image features need to be robust against differences in positioning, acquisition and segmentation to ensure reproducibility. Radiomic models that only include robust features can be used to analyse new images, whereas models with non-robust features may fail to predict the outcome of interest accurately. Test-retest imaging is recommended to assess robustness, but may not be available for the phenotype of interest. We therefore investigated 18 combinations of image perturbations to determine feature robustness, based on noise addition (N), translation (T), rotation (R), volume growth/shrinkage (V) and supervoxel-based contour randomisation (C). Test-retest and perturbation robustness were compared for combined total of 4032 morphological, statistical and texture features that were computed from the gross tumour volume in two cohorts with computed tomography imaging: I) 31 non-small-cell lung cancer (NSCLC) patients; II): 19 head-and-neck squamous cell carcinoma (HNSCC) patients. Robustness was determined using the 95% confidence interval (CI) of the intraclass correlation coefficient (1, 1). Features with CI ≥ 0:90 were considered robust. The NTCV, TCV, RNCV and RCV perturbation chain produced similar results and identified the fewest false positive robust features (NSCLC: 0.2–0.9%; HNSCC: 1.7–1.9%). Thus, these perturbation chains may be used as an alternative to test-retest imaging to assess feature robustness.

Quantifying the robustness of [ 18 F]FDG-PET/CT radiomic features with respect to tumor delineation in head and neck and pancreatic cancer patients

PURPOSE

To investigate the robustness of PET radiomic features (RF) against tumour delineation uncertainty in two clinically relevant situations.

METHODS

Twenty-five head-and-neck (HN) and 25 pancreatic cancer patients previously treated with ¹⁸F-Fluorodeoxyglucose (FDG) positron emission tomography/computed tomography (PET/CT)-based planning optimization were considered. Seven FDG-based contours were delineated for tumour (T) and positive lymph nodes (N, for HN patients only) following manual (2 observers), semi-automatic (based on SUV maximum gradient: PET_Edge) and automatic (40%, 50%, 60%, 70% SUV_max thresholds) methods. Seventy-three RF (14 of first order and 59 of higher order) were extracted using the CGITA software (v.1.4). The impact of delineation on volume agreement and RF was assessed by DICE and Intra-class Correlation Coefficients (ICC).

RESULTS

A large disagreement between manual and SUV_max method was found for thresholds  ≥50%. Inter-observer variability showed median DICE values between 0.81 (HN-T) and 0.73 (pancreas). Volumes defined by PET_Edge were better consistent with the manual ones compared to SUV40%. Regarding RF, 19%/19%/47% of the features showed ICC < 0.80 between observers for HN-N/HN-T/pancreas, mostly in the Voxel-alignment matrix and in the intensity-size zone matrix families. RFs with ICC < 0.80 against manual delineation (taking the worst value) increased to 44%/36%/61% for PET_Edge and to 69%/53%/75% for SUV40%.

CONCLUSIONS

About 80%/50% of 72 RF were consistent between observers for HN/pancreas patients. PET_edge was sufficiently robust against manual delineation while SUV40% showed a worse performance. This result suggests the possibility to replace manual with semi-automatic delineation of HN and pancreas tumours in studies including PET radiomic analyses.