Reproducibility of radiomics for deciphering tumor phenotype with imaging. Sci Rep. 2016;6:23428. [PMID: 27009765 PMCID: PMC4806325 DOI: 10.1038/srep23428]

Differential Diagnosis of Nonhypervascular Pancreatic Neuroendocrine Neoplasms From Pancreatic Ductal Adenocarcinomas, Based on Computed Tomography Radiological Features and Texture Analysis

RATIONALE AND OBJECTIVES

To determine computed tomography (CT) radiological features and texture features that are rewarding in differentiating nonhypervascular pancreatic neuroendocrine neoplasms (PNENs) from pancreatic ductal adenocarcinomas (PDACs).

MATERIALS AND METHODS

We compared patients to pathologically proven nonhypervascular PNENs and age-matched controls with pathologically proven PDACs in a 1:2 ratio. Preoperative CT images in the arterial phase (AP) and portal vein phase (PVP) were obtained. Two radiologists independently reviewed the morphological characteristics of each tumor. Three-dimensional regions of interest (ROIs), drawn using ITK-SNAP software, were input into AK software (Artificial Intelligent Kit, GE) to extract texture features from AP and PVP images. Differences between PNENs and PDACs were analyzed with the chi-squared test, least absolute shrinkage and selection operator, kappa statistics, and uni- and multivariate logistic regression analyses.

RESULTS

In total, 40 nonhypervascular PNENs and 80 PDACs were evaluated. Maximum diameter on axial section, margin, calcification, vascularity in the tumor, and tumor heterogeneity were significantly different between PDACs and nonhypervascular PNENs. Multivariate analysis showed well-defined tumor margin (odds ratio: 21.0) and presence of calcification (odds ratio: 4.4) were significant predictors of nonhypervascular PNENs. The area under the receiver operating characteristic curve of the radiological feature model, AP texture model, and PVP texture model were 0.780, 0.855, and 0.929, respectively, based on logistic regression.

CONCLUSION

A well-defined margin and calcification in the tumor were helpful in discriminating nonhypervascular PNENs from PDACs. Texture analysis of contrast-enhanced CT images could be beneficial in differentially diagnosing nonhypervascular PNENs and PDACs.

Stability analysis of CT radiomic features with respect to segmentation variation in oropharyngeal cancer

INTRODUCTION

Accurate segmentation of tumors and quantification of tumor features are important for cancer detection, diagnosis, monitoring, and planning therapeutic intervention. Due to inherent noise components in multi-parametric imaging and inter-observer and intra-observer variations, it is common that various segmentation methods may produce large segmentation errors in tumor volumes and their associated radiomic features. The purpose of this study is to carry out the stability analysis for radiomic features with respect to segmentation variation in oropharyngeal cancer (OPC).

METHODS

In this study, 436 contrast-enhanced computed tomography (CT) axial images were collected from patients with OPC. In order to derive various segmentations of tumor volumes, two additional segmentations were obtained via resizing the original segmented regions of interest (ROIs) based on their geometric information on the boundary. For three ROI image groups, we calculated 109 radiomic features. Then, a logistic regression model was built to investigate the correlation between the radiomic features extracted from GTVp and the response to chemotherapy and radiation in terms of overall survival (OS). Finally, in order to evaluate the stability of each feature with respect to segmentation results, based on the prediction probabilities, we assessed the inter-rater reliability and reproducibility by calculating the intra-class correlation coefficients (ICC) and concordance correlation coefficients (CCC).

RESULTS

Most radiomic features in this study varied a lot when the ROIs were not well segmented. For both the representation agreement and predictive agreement, the ICC and CCC were below 0.5 for all the features. We still found some robust features with relatively high ICC and CCC compared to most features. For example, 25percentile (ICC = 0.38, CCC = 0.37 in representation agreement and ICC = CCC = 0.27 in predictive agreement) is a quantile based feature, which is robust to the extremely high or low values; and Hu_1_std (ICC = 0.31, CCC = 0.31 in representation agreement) is a feature calculated based on the first Hu moment, which is invariant to the transformation of ROIs.

CONCLUSION

In OPC studies, the tumor segmentation variation affects the radiomic features from CT images in terms of both representation and prediction. Some features that are robust to the extreme values or invariant to the transformation of ROIs may be treated as radiomic markers to assist with OPC treatment monitoring and prognostic prediction.

Radiographic Image Radiomics Feature Reproducibility: A Preliminary Study on the Impact of Field Size

RATIONALE AND OBJECTIVES

Radiomics is an approach to quantifying diseases. Recently, several studies have indicated that radiomics features are vulnerable against imaging parameters. The aim of this study is to assess how radiomics features change with radiographic field sizes, positions in the field size, and mAs.

MATERIALS AND METHODS

A large and small wood phantom and a cotton phantom were prepared and imaged in different field sizes, mAs, and placement in the radiographic field size. A region of interest was drawn on the image features, and twenty two features were extracted. Radiomics feature reproducibility was obtained based on coefficient of variation, Bland-Altman analysis, and intraclass correlation coefficient. Features with coefficient of variation ≤ 5%, intraclass correlation coefficient ≤ 90%, and 1% ≤ U/LRL ≤30% were introduced as robust features. U/LRL is upper/lower reproducibility limits in Bland-Altman.

RESULTS

For all field sizes and all phantoms, features including Difference Variance, Inverse Different Moment, Fraction, Long Run Emphasis, Run Length Non Uniformity, and Short Run Emphasis were found as highly reproducible features. For change in the position of field size, Fraction was the most reproducible in all field sizes and all phantoms. On the mAs change, we found that feature, Short Run Emphasis field 15 × 15 for small wood phantom, and Correlation in all field sizes for Cotton are the most reproducible features.

CONCLUSION

We demonstrated that radiomics features are strongly vulnerable against radiographic field size, positions in the radiation field, mAs, and phantom materials, and reproducibility analyses should be performed before each radiomics study. Moreover, these changing parameters should be considered, and their effects should be minimized in future radiomics studies.

A quantitative imaging biomarker for predicting disease-free-survival-associated histologic subgroups in lung adenocarcinoma

OBJECTIVES

Classification of histologic subgroups has significant prognostic value for lung adenocarcinoma patients who undergo surgical resection. However, clinical histopathology assessment is generally performed on only a small portion of the overall tumor from biopsy or surgery. Our objective is to identify a noninvasive quantitative imaging biomarker (QIB) for the classification of histologic subgroups in lung adenocarcinoma patients.

METHODS

We retrospectively collected and reviewed 1313 CT scans of patients with resected lung adenocarcinomas from two geographically distant institutions who were seen between January 2014 and October 2017. Three study cohorts, the training, internal validation, and external validation cohorts, were created, within which lung adenocarcinomas were divided into two disease-free-survival (DFS)-associated histologic subgroups, the mid/poor and good DFS groups. A comprehensive machine learning- and deep learning-based analytical system was adopted to identify reproducible QIBs and help to understand QIBs' significance.

RESULTS

Intensity-Skewness, a QIB quantifying tumor density distribution, was identified as the optimal biomarker for predicting histologic subgroups. Intensity-Skewness achieved high AUCs (95% CI) of 0.849(0.813,0.881), 0.820(0.781,0.856) and 0.863(0.827,0.895) on the training, internal validation, and external validation cohorts, respectively. A criterion of Intensity-Skewness ≤ 1.5, which indicated high tumor density, showed high specificity of 96% (sensitivity 46%) and 99% (sensitivity 53%) on predicting the mid/poor DFS group in the training and external validation cohorts, respectively.

CONCLUSIONS

A QIB derived from routinely acquired CT was able to predict lung adenocarcinoma histologic subgroups, providing a noninvasive method that could potentially benefit personalized treatment decision-making for lung cancer patients.

KEY POINTS

• A noninvasive imaging biomarker, Intensity-Skewness, which described the distortion of pixel-intensity distribution within lesions on CT images, was identified as a biomarker to predict disease-free-survival-associated histologic subgroups in lung adenocarcinoma. • An Intensity-Skewness of ≤ 1.5 has high specificity in predicting the mid/poor disease-free survival histologic patient group in both the training cohort and the external validation cohort. • The Intensity-Skewness is a feature that can be automatically computed with high reproducibility and robustness.

Multiparametric radiomics methods for breast cancer tissue characterization using radiological imaging

BACKGROUND AND PURPOSE

Multiparametric radiological imaging is vital for detection, characterization, and diagnosis of many different diseases. Radiomics provide quantitative metrics from radiological imaging that may infer potential biological meaning of the underlying tissue. However, current methods are limited to regions of interest extracted from a single imaging parameter or modality, which limits the amount of information available within the data. This limitation can directly affect the integration and applicable scope of radiomics into different clinical settings, since single image radiomics are not capable of capturing the true underlying tissue characteristics in the multiparametric radiological imaging space. To that end, we developed a multiparametric imaging radiomic (mpRad) framework for extraction of first and second order radiomic features from multiparametric radiological datasets.

METHODS

We developed five different radiomic techniques that extract different aspects of the inter-voxel and inter-parametric relationships within the high-dimensional multiparametric magnetic resonance imaging breast datasets. Our patient cohort consisted of 138 breast patients, where, 97 patients had malignant lesions and 41 patients had benign lesions. Sensitivity, specificity, receiver operating characteristic (ROC) and areas under the curve (AUC) analysis were performed to assess diagnostic performance of the mpRad parameters. Statistical significance was set at p < 0.05.

RESULTS

The mpRad features successfully classified malignant from benign breast lesions with excellent sensitivity and specificity of 82.5% and 80.5%, respectively, with Area Under the receiver operating characteristic Curve (AUC) of 0.87 (0.81-0.93). mpRad provided a 9-28% increase in AUC metrics over single radiomic parameters.

CONCLUSIONS

We have introduced the mpRad framework that extends radiomic analysis from single images to multiparametric datasets for better characterization of the underlying tissue biology.

Automated Identification of Optimal Portal Venous Phase Timing with Convolutional Neural Networks

OBJECTIVES

To develop a deep learning-based algorithm to automatically identify optimal portal venous phase timing (PVP-timing) so that image analysis techniques can be accurately performed on post contrast studies.

METHODS

681 CT-scans (training: 479 CT-scans; validation: 202 CT-scans) from a multicenter clinical trial in patients with liver metastases from colorectal cancer were retrospectively analyzed for algorithm development and validation. An additional external validation was performed on a cohort of 228 CT-scans from gastroenteropancreatic neuroendocrine cancer patients. Image acquisition was performed according to each centers' standard CT protocol for single portal venous phase, portal venous acquisition. The reference gold standard for the classification of PVP-timing as either optimal or nonoptimal was based on experienced radiologists' consensus opinion. The algorithm performed automated localization (on axial slices) of the portal vein and aorta upon which a novel dual input Convolutional Neural Network calculated a probability of the optimal PVP-timing.

RESULTS

The algorithm automatically computed a PVP-timing score in 3 seconds and reached area under the curve of 0.837 (95% CI: 0.765, 0.890) in validation set and 0.844 (95% CI: 0.786, 0.889) in external validation set.

CONCLUSION

A fully automated, deep-learning derived PVP-timing algorithm was developed to classify scans' contrast-enhancement timing and identify scans with optimal PVP-timing. The rapid identification of such scans will aid in the analysis of quantitative (radiomics) features used to characterize tumors and changes in enhancement with treatment in a multitude of settings including quantitative response criteria such as Choi and MASS which rely on reproducible measurement of enhancement.

The Emerging Role of Radiomics in COPD and Lung Cancer

Medical imaging plays a key role in evaluating and monitoring lung diseases such as chronic obstructive pulmonary disease (COPD) and lung cancer. The application of artificial intelligence in medical imaging has transformed medical images into mineable data, by extracting and correlating quantitative imaging features with patients' outcomes and tumor phenotype - a process termed radiomics. While this process has already been widely researched in lung oncology, the evaluation of COPD in this fashion remains in its infancy. Here we outline the main applications of radiomics in lung cancer and briefly review the workflow from image acquisition to the evaluation of model performance. Finally, we discuss the current assessments of COPD and the potential application of radiomics in COPD.

Image-Based Cardiac Diagnosis With Machine Learning: A Review

Cardiac imaging plays an important role in the diagnosis of cardiovascular disease (CVD). Until now, its role has been limited to visual and quantitative assessment of cardiac structure and function. However, with the advent of big data and machine learning, new opportunities are emerging to build artificial intelligence tools that will directly assist the clinician in the diagnosis of CVDs. This paper presents a thorough review of recent works in this field and provide the reader with a detailed presentation of the machine learning methods that can be further exploited to enable more automated, precise and early diagnosis of most CVDs.

Deep Learning to Distinguish Benign from Malignant Renal Lesions Based on Routine MR Imaging

PURPOSE

With increasing incidence of renal mass, it is important to make a pretreatment differentiation between benign renal mass and malignant tumor. We aimed to develop a deep learning model that distinguishes benign renal tumors from renal cell carcinoma (RCC) by applying a residual convolutional neural network (ResNet) on routine MR imaging.

EXPERIMENTAL DESIGN

Preoperative MR images (T2-weighted and T1-postcontrast sequences) of 1,162 renal lesions definitely diagnosed on pathology or imaging in a multicenter cohort were divided into training, validation, and test sets (70:20:10 split). An ensemble model based on ResNet was built combining clinical variables and T1C and T2WI MR images using a bagging classifier to predict renal tumor pathology. Final model performance was compared with expert interpretation and the most optimized radiomics model.

RESULTS

Among the 1,162 renal lesions, 655 were malignant and 507 were benign. Compared with a baseline zero rule algorithm, the ensemble deep learning model had a statistically significant higher test accuracy (0.70 vs. 0.56, P = 0.004). Compared with all experts averaged, the ensemble deep learning model had higher test accuracy (0.70 vs. 0.60, P = 0.053), sensitivity (0.92 vs. 0.80, P = 0.017), and specificity (0.41 vs. 0.35, P = 0.450). Compared with the radiomics model, the ensemble deep learning model had higher test accuracy (0.70 vs. 0.62, P = 0.081), sensitivity (0.92 vs. 0.79, P = 0.012), and specificity (0.41 vs. 0.39, P = 0.770).

CONCLUSIONS

Deep learning can noninvasively distinguish benign renal tumors from RCC using conventional MR imaging in a multi-institutional dataset with good accuracy, sensitivity, and specificity comparable with experts and radiomics.

Radiomics approaches in gastric cancer: a frontier in clinical decision making

Objective:

To review the application of radiomics in gastric cancer and its challenges as well as future prospects.

Data sources:

A research for relevant studies were performed in PubMed with the terms of “radiomics,” “texture analysis,” and “gastric cancer.” The search was updated until February 28^th, 2019.

Study selection:

All original articles regarding the investigation of texture analysis or radiomics in gastric cancer were retrieved. Only papers written in English were included.

Results:

A total of 17 original articles were selected in final. It is shown that radiomics has yielded moderate to excellent performance in a spectrum of respects including differential diagnosis, assessment of histological differential degree, evaluation of tumor stage, prediction of response to therapy, and prognosis in gastric cancer. Yet, a number of challenges are facing both radiomics itself and its application in gastric cancer.

Conclusions:

Radiomics holds great potential in facilitating decision-making in gastric cancer. With the standardization of work-flow and advancement of machine learning methods, radiomics is expected to make great breakthroughs in precision medicine of gastric cancer.

Robustness and Reproducibility of Radiomics in Magnetic Resonance Imaging: A Phantom Study

OBJECTIVES

The aim of this study was to investigate the robustness and reproducibility of radiomic features in different magnetic resonance imaging sequences.

MATERIALS AND METHODS

A phantom was scanned on a clinical 3 T system using fluid-attenuated inversion recovery (FLAIR), T1-weighted (T1w), and T2-weighted (T2w) sequences with low and high matrix size. For retest data, scans were repeated after repositioning of the phantom. Test and retest datasets were segmented using a semiautomated approach. Intraobserver and interobserver comparison was performed. Radiomic features were extracted after standardized preprocessing of images. Test-retest robustness was assessed using concordance correlation coefficients, dynamic range, and Bland-Altman analyses. Reproducibility was assessed by intraclass correlation coefficients.

RESULTS

The number of robust features (concordance correlation coefficient and dynamic range ≥ 0.90) was higher for features calculated from FLAIR than from T1w and T2w images. High-resolution FLAIR images provided the highest percentage of robust features (n = 37/45, 81%). No considerable difference in the number of robust features was observed between low- and high-resolution T1w and T2w images (T1w low: n = 26/45, 56%; T1w high: n = 25/45, 54%; T2 low: n = 21/45, 46%; T2 high: n = 24/45, 52%). A total of 15 (33%) of 45 features showed excellent robustness across all sequences and demonstrated excellent intraobserver and interobserver reproducibility (intraclass correlation coefficient ≥ 0.75).

CONCLUSIONS

FLAIR delivers the most robust substrate for radiomic analyses. Only 15 of 45 features showed excellent robustness and reproducibility across all sequences. Care must be taken in the interpretation of clinical studies using nonrobust features.

Radiomic features of the pancreas on CT imaging accurately differentiate functional abdominal pain, recurrent acute pancreatitis, and chronic pancreatitis

PURPOSE

Patients with recurrent abdominal pain and pancreatic enzyme elevations may be diagnosed clinically with recurrent acute pancreatitis (RAP) even with normal imaging or no imaging at all. Since neither abdominal pain nor enzyme elevations are specific for acute pancreatitis (AP), and patients with RAP often have a normal appearing pancreas on CT after resolution of an AP episode, RAP diagnosis can be challenging. This study aims to determine if quantitative radiomic features of the pancreas on CT can differentiate patients with functional abdominal pain, RAP, and chronic pancreatitis (CP).

METHOD

Contrast enhanced CT abdominal images of adult patients evaluated in a pancreatitis clinic from 2010 to 2018 with the diagnosis of RAP, functional abdominal pain, or CP were retrospectively reviewed. The pancreas was outlined by drawing region of interest (ROI) on images. 54 radiomic features were extracted from each ROI and were compared between the patient groups. A one-vs-one Isomap and Support Vector Machine (IsoSVM) classifier was also trained and tested to classify patients into one of the three diagnostic groups based on their radiomic features.

RESULTS

Among the study's 56 patients, 20 (35.7 %) had RAP, 19 (33.9 %) had functional abdominal pain, and 17 (30.4 %) had CP. On univariate analysis, 11 radiomic features (10 GLCM features and one NGTDM feature) were significantly different between the patient groups. The IsoSVM classifier for prediction of patient diagnosis had an overall accuracy of 82.1 %.

CONCLUSIONS

Certain radiomic features on CT imaging can differentiate patients with functional abdominal pain, RAP, and CP.

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.

Assessment of Stability and Discrimination Capacity of Radiomic Features on Apparent Diffusion Coefficient Images

The objectives of the study are to develop a new way to assess stability and discrimination capacity of radiomic features without the need of test-retest or multiple delineations and to use information obtained to perform a preliminary feature selection. Apparent diffusion coefficient (ADC) maps were computed from diffusion-weighted magnetic resonance images (DW-MRI) of two groups of patients: 18 with soft tissue sarcomas (STS) and 18 with oropharyngeal cancers (OPC). Sixty-nine radiomic features were computed, using three different histogram discretizations (16, 32, and 64 bins). Geometrical transformations (translations) of increasing entity were applied to the regions of interest (ROIs), and the intra-class correlation coefficient (ICC) was used to compare the features computed on the original and modified ROIs. The distribution of ICC values for minimal and maximal entity translations (ICC₁₀ and ICC₁₀₀, respectively) was used to adjust thresholds of ICC (ICC_min and ICC_max) used to discriminate between good, unstable (ICC₁₀ < ICC_min), and non-discriminative features (ICC₁₀₀ > ICC_max). Fifty-four and 59 radiomic features passed the stability-based selection for all the three histogram discretizations for the OPC and STS datasets, respectively. The excluded features were similar across the different histogram discretizations (Jaccard’s index 0.77 ± 0.13 and 0.9 ± 0.1 for OPC and STS, respectively) but different between datasets (Jaccard’s index 0.19 ± 0.02). The results suggest that the observed radiomic features are mainly stable and discriminative, but the stability depends on the region of the body under observation. The method provides a way to assess stability without the need of test-retest or multiple delineations.

Imaging for Response Assessment in Radiation Oncology: Current and Emerging Techniques

Imaging in radiation oncology is essential for the evaluation of treatment response in tumors and organs at risk. This influences further treatment decisions and could possibly be used to adapt therapy. This review article focuses on the currently used imaging modalities for response assessment in radiation oncology and gives an overview of new and promising techniques within this field.

Diffusion radiomics as a diagnostic model for atypical manifestation of primary central nervous system lymphoma: development and multicenter external validation

Background

Radiomics is a rapidly growing field in neuro-oncology, but studies have been limited to conventional MRI, and external validation is critically lacking. We evaluated technical feasibility, diagnostic performance, and generalizability of a diffusion radiomics model for identifying atypical primary central nervous system lymphoma (PCNSL) mimicking glioblastoma.

Methods

A total of 1618 radiomics features were extracted from diffusion and conventional MRI from 112 patients (training set, 70 glioblastomas and 42 PCNSLs). Feature selection and classification were optimized using a machine-learning algorithm. The diagnostic performance was tested in 42 patients of internal and external validation sets. The performance was compared with that of human readers (2 neuroimaging experts), cerebral blood volume (90% histogram cutoff, CBV90), and apparent diffusion coefficient (10% histogram, ADC10) using the area under the receiver operating characteristic curve (AUC).

Results

The diffusion radiomics was optimized with the combination of recursive feature elimination and a random forest classifier (AUC 0.983, stability 2.52%). In internal validation, the diffusion model (AUC 0.984) showed similar performance with conventional (AUC 0.968) or combined diffusion and conventional radiomics (AUC 0.984) and better than human readers (AUC 0.825-0.908), CBV90 (AUC 0.905), or ADC10 (AUC 0.787) in atypical PCNSL diagnosis. In external validation, the diffusion radiomics showed robustness (AUC 0.944) and performed better than conventional radiomics (AUC 0.819) and similar to combined radiomics (AUC 0.946) or human readers (AUC 0.896-0.930).

Conclusion

The diffusion radiomics model had good generalizability and yielded a better diagnostic performance than conventional radiomics or single advanced MRI in identifying atypical PCNSL mimicking glioblastoma.

Vol-PACT: A Foundation for the NIH Public-Private Partnership That Supports Sharing of Clinical Trial Data for the Development of Improved Imaging Biomarkers in Oncology

PURPOSE

To develop a public-private partnership to study the feasibility of a new approach in collecting and analyzing clinically annotated imaging data from landmark phase III trials in advanced solid tumors.

PATIENTS AND METHODS

The collection of clinical trials fulfilled the following inclusion criteria: completed randomized trials of > 300 patients, highly measurable solid tumors (non-small-cell lung cancer, colorectal cancer, renal cell cancer, and melanoma), and required sponsor and institutional review board sign-offs. The new approach in analyzing computed tomography scans was to transfer to an academic image analysis laboratory, draw contours semi-automatically by using in-house-developed algorithms integrated into the open source imaging platform Weasis, and perform serial volumetric measurement.

RESULTS

The median duration of contracting with five sponsors was 12 months. Ten trials in 7,085 patients that covered 12 treatment regimens across 20 trial arms were collected. To date, four trials in 3,954 patients were analyzed. Source imaging data were transferred to the academic core from 97% of trial patients (n = 3,837). Tumor imaging measurements were extracted from 82% of transferred computed tomography scans (n = 3,162). Causes of extraction failure were nonmeasurable disease (n = 392), single imaging time point (n = 224), and secondary captured images (n = 59). Overall, clinically annotated imaging data were extracted in 79% of patients (n = 3,055), and the primary trial end point analysis in each trial remained representative of each original trial end point.

CONCLUSION

The sharing and analysis of source imaging data from large randomized trials is feasible and offer a rich and reusable, but largely untapped, resource for future research on novel trial-level response and progression imaging metrics.

A machine learning algorithm predicts molecular subtypes in pancreatic ductal adenocarcinoma with differential response to gemcitabine-based versus FOLFIRINOX chemotherapy

PURPOSE

Development of a supervised machine-learning model capable of predicting clinically relevant molecular subtypes of pancreatic ductal adenocarcinoma (PDAC) from diffusion-weighted-imaging-derived radiomic features.

METHODS

The retrospective observational study assessed 55 surgical PDAC patients. Molecular subtypes were defined by immunohistochemical staining of KRT81. Tumors were manually segmented and 1606 radiomic features were extracted with PyRadiomics. A gradient-boosted-tree algorithm was trained on 70% of the patients (N = 28) and tested on 30% (N = 17) to predict KRT81+ vs. KRT81- tumor subtypes. A gradient-boosted survival regression model was fit to the disease-free and overall survival data. Chemotherapy response and survival were assessed stratified by subtype and radiomic signature. Radiomic feature importance was ranked.

RESULTS

The mean±STDEV sensitivity, specificity and ROC-AUC were 0.90±0.07, 0.92±0.11, and 0.93±0.07, respectively. The mean±STDEV concordance indices between the disease-free and overall survival predicted by the model based on the radiomic parameters and actual patient survival were 0.76±0.05 and 0.71±0.06, respectively. Patients with a KRT81+ subtype experienced significantly diminished median overall survival compared to KRT81- patients (7.0 vs. 22.6 months, HR 4.03, log-rank-test P = <0.001) and a significantly improved response to gemcitabine-based chemotherapy over FOLFIRINOX (10.14 vs. 3.8 months median overall survival, HR 2.33, P = 0.037) compared to KRT81- patients, who responded significantly better to FOLFIRINOX over gemcitabine-based treatment (30.8 vs. 13.4 months median overall survival, HR 2.41, P = 0.027). Entropy was ranked as the most important radiomic feature.

CONCLUSIONS

The machine-learning based analysis of radiomic features enables the prediction of subtypes of PDAC, which are highly relevant for disease-free and overall patient survival and response to chemotherapy.

Adaptive statistical iterative reconstruction (ASIR) affects CT radiomics quantification in primary colorectal cancer

OBJECTIVES

To investigate whether adaptive statistical iterative reconstruction (ASIR), a hybrid iterative CT image reconstruction algorithm, affects radiomics feature quantification in primary colorectal cancer compared to filtered back projection. Additionally, to establish whether radiomics from single-slice analysis undergo greater change than those from multi-slice analysis.

METHODS

Following review board approval, contrast-enhanced CT studies from 32 prospective primary colorectal cancer patients were reconstructed with 20% ASIR level increments, from 0 to 100%. Radiomics analysis was applied to single-slice and multi-slice regions of interest outlining the tumour: 70 features, including statistical (first-, second- and high-order) and fractal radiomics, were generated per dataset. The effect of ASIR was calculated by means of multilevel linear regression.

RESULTS

Twenty-eight CT datasets were suitable for analysis. Incremental ASIR levels determined a significant change (p < 0.001) in most statistical radiomics features, best described by a simple linear relationship. First-order statistical features, including mean, standard deviation, skewness, kurtosis, energy and entropy, underwent a relatively small change in both single-slice and multi-slice analysis (median standardised effect size B = 0.08). Second-order statistical features, including grey-level co-occurrence and difference matrices, underwent a greater change in single-slice analysis (median B = 0.36) than in multi-slice analysis (median B = 0.13). Fractal features underwent a significant change only in single-slice analysis (median B = 0.49).

CONCLUSIONS

Incremental levels of ASIR affect significantly CT radiomics quantification in primary colorectal cancer. Second-order statistical and fractal features derived from single-slice analysis undergo greater change than those from multi-slice analysis.

KEY POINTS

• Incremental levels of ASIR determine a significant change in most statistical (first-, second- and high-order) CT radiomics features measured in primary colorectal cancer, best described by a linear relationship. • First-order statistical features undergo a small change, both from single-slice and multi-slice radiomics analyses. • Most second-order statistical features undergo a greater change in single-slice analysis than in multi-slice analysis. Fractal features are only affected in single-slice analysis.

Application of Radiomics in Predicting the Malignancy of Pulmonary Nodules in Different Sizes

OBJECTIVE. The purpose of this study was to investigate the utility of radiomics for predicting the malignancy of pulmonary nodules (PNs) of different sizes using unenhanced, thin-section CT images. MATERIALS AND METHODS. Patients with a single PN (n = 373) who underwent a preoperative chest CT were recruited retrospectively at Beijing Friendship Hospital from March 2015 to March 2018. Of the 373 PNs studied, 192 were benign and 181 were malignant. The lesions were classified into three groups (T1a, T1b, or T1c according to the 8th edition of the TNM staging system for lung cancer) on the basis of lesion diameters: T1a (diameter, 0-1 cm), T1b (1 cm < diameter ≤ 2 cm) and T1c (2 cm < diameter ≤ 3 cm). A total of 1160 radiomic features were extracted from PN segmentation on unenhanced CT images. We developed three radiomic models to predict PN malignancy in each group on the basis of the extracted radiomic features. Fivefold cross-validation was used to estimate AUC, accuracy, sensitivity, and specificity for indicating the performance of prediction models. RESULTS. The AUC, accuracy, sensitivity, and specificity for predicting PN malignancy in each group were 0.84, 0.77, 0.89, and 0.74 with the T1a model; 0.78, 0.73, 0.74, and 0.71 with the T1b model, and 0.79, 0.76, 0.77, and 0.73 with the T1c model, respectively. The most contributive radiomic features for predicting PN malignancy for groups T1a, T1b, and T1c were LoG_X_Uniformity, Intensity_Minimum, and Shape_SI9, respectively. CONCLUSION. Radiomic features based on unenhanced CT images can be used to predict the malignancy of pulmonary nodules. The radiomic T1a model showed superior prediction performance to the T1b and T1c models, and the best performance in terms of AUC and sensitivity was found for predicting the malignancy of T1a PN.

Radiomics features of the primary tumor fail to improve prediction of overall survival in large cohorts of CT- and PET-imaged head and neck cancer patients

Radiomics studies require many patients in order to power them, thus patients are often combined from different institutions and using different imaging protocols. Various studies have shown that imaging protocols affect radiomics feature values. We examined whether using data from cohorts with controlled imaging protocols improved patient outcome models. We retrospectively reviewed 726 CT and 686 PET images from head and neck cancer patients, who were divided into training or independent testing cohorts. For each patient, radiomics features with different preprocessing were calculated and two clinical variables-HPV status and tumor volume-were also included. A Cox proportional hazards model was built on the training data by using bootstrapped Lasso regression to predict overall survival. The effect of controlled imaging protocols on model performance was evaluated by subsetting the original training and independent testing cohorts to include only patients whose images were obtained using the same imaging protocol and vendor. Tumor volume, HPV status, and two radiomics covariates were selected for the CT model, resulting in an AUC of 0.72. However, volume alone produced a higher AUC, whereas adding radiomics features reduced the AUC. HPV status and one radiomics feature were selected as covariates for the PET model, resulting in an AUC of 0.59, but neither covariate was significantly associated with survival. Limiting the training and independent testing to patients with the same imaging protocol reduced the AUC for CT patients to 0.55, and no covariates were selected for PET patients. Radiomics features were not consistently associated with survival in CT or PET images of head and neck patients, even within patients with the same imaging protocol.

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.

Radiomics for Classification of Lung Cancer Histological Subtypes Based on Nonenhanced Computed Tomography

OBJECTIVES

To evaluate the performance of using radiomics method to classify lung cancer histological subtypes based on nonenhanced computed tomography images.

MATERIALS AND METHODS

278 patients with pathologically confirmed lung cancer were collected, including 181 nonsmall cell lung cancer (NSCLC) and 97 small cell lung cancers (SCLC) patients. Among the NSCLC patients, 88 patients were adenocarcinomas (AD) and 93 patients were squamous cell carcinomas (SCC). In total, 1695 quantitative radiomic features (QRF) were calculated from the primary lung cancer tumor in each patient. To build radiomic classification model based on the extracted QRFs, several machine-learning algorithms were applied sequentially. First, unsupervised hierarchical clustering was used to exclude highly correlated QRFs; second, the minimum Redundancy Maximum Relevance feature selection algorithm was employed to select informative and nonredundant QRFs; finally, the Incremental Forward Search and Support Vector Machine classification algorithms were used to combine the selected QRFs and build the model. In our work, to study the phenotypic differences among lung cancer histological subtypes, four classification models were built. They were models of SCLC vs NSCLC, SCLC vs AD, SCLC vs SCC, and AD vs SCC. The performance of the classification models was evaluated by the area under the receiver operating characteristic curve (AUC) estimated by three-fold cross-validation.

RESULTS

The AUC (95% confidence interval) for the model of SCLC vs NSCLC was 0.741(0.678, 0.795). For the models of SCLC vs AD and SCLC vs SCC, the AUCs were 0.822(0.755, 0.875) and 0.665(0.583, 0.738), respectively. The AUC for the model of AD vs SCC was 0.655(0.570, 0.731). Several QRFs ("Law_15," "LoG_Uniformity," "GLCM_Contrast," and "Compactness Factor") that characterize tumor heterogeneity and shape were selected as the significant features to build the models.

CONCLUSION

Our results show that phenotypic differences exist among different lung cancer histological subtypes on nonenhanced computed tomography image.

Validated imaging biomarkers as decision-making tools in clinical trials and routine practice: current status and recommendations from the EIBALL* subcommittee of the European Society of Radiology (ESR)

Observer-driven pattern recognition is the standard for interpretation of medical images. To achieve global parity in interpretation, semi-quantitative scoring systems have been developed based on observer assessments; these are widely used in scoring coronary artery disease, the arthritides and neurological conditions and for indicating the likelihood of malignancy. However, in an era of machine learning and artificial intelligence, it is increasingly desirable that we extract quantitative biomarkers from medical images that inform on disease detection, characterisation, monitoring and assessment of response to treatment. Quantitation has the potential to provide objective decision-support tools in the management pathway of patients. Despite this, the quantitative potential of imaging remains under-exploited because of variability of the measurement, lack of harmonised systems for data acquisition and analysis, and crucially, a paucity of evidence on how such quantitation potentially affects clinical decision-making and patient outcome. This article reviews the current evidence for the use of semi-quantitative and quantitative biomarkers in clinical settings at various stages of the disease pathway including diagnosis, staging and prognosis, as well as predicting and detecting treatment response. It critically appraises current practice and sets out recommendations for using imaging objectively to drive patient management decisions.

Image-based Classification of Tumor Type and Growth Rate using Machine Learning: a preclinical study

Medical images such as magnetic resonance (MR) imaging provide valuable information for cancer detection, diagnosis, and prognosis. In addition to the anatomical information these images provide, machine learning can identify texture features from these images to further personalize treatment. This study aims to evaluate the use of texture features derived from T₁-weighted post contrast scans to classify different types of brain tumors and predict tumor growth rate in a preclinical mouse model. To optimize prediction models this study uses varying gray-level co-occurrence matrix (GLCM) sizes, tumor region selection and different machine learning models. Using a random forest classification model with a GLCM of size 512 resulted in 92%, 91%, and 92% specificity, and 89%, 85%, and 73% sensitivity for GL261 (mouse glioma), U87 (human glioma) and Daoy (human medulloblastoma), respectively. A tenfold cross-validation of the classifier resulted in 84% accuracy when using the entire tumor volume for feature extraction and 74% accuracy for the central tumor region. A two-layer feedforward neural network using the same features is able to predict tumor growth with 16% mean squared error. Broadly applicable, these predictive models can use standard medical images to classify tumor type and predict tumor growth, with model performance, varying as a function of GLCM size, tumor region, and tumor type.

Convolution kernel and iterative reconstruction affect the diagnostic performance of radiomics and deep learning in lung adenocarcinoma pathological subtypes

Background

The aim of this study was to investigate the influence of convolution kernel and iterative reconstruction on the diagnostic performance of radiomics and deep learning (DL) in lung adenocarcinomas.

Methods

A total of 183 patients with 215 lung adenocarcinomas were included in this study. All CT imaging data was reconstructed with three reconstruction algorithms (ASiR at 0%, 30%, 60% strength), each with two convolution kernels (bone and standard). A total of 171 nodules were selected as the training‐validation set, whereas 44 nodules were selected as the testing set. Logistic regression and a DL framework‐DenseNets were selected to tackle the task. Three logical experiments were implemented to fully explore the influence of the studied parameters on the diagnostic performance. The receiver operating characteristic curve (ROC) was used to evaluate the performance of constructed models.

Results

In Experiments A and B, no statistically significant results were found in the radiomic method, whereas two and six pairs were statistically significant (P < 0.05) in the DL method. In Experiment_C, significant differences in one and four models were found in the radiomics and DL methods, respectively. Moreover, models constructed with standard convolution kernel data outperformed that constructed with bone convolution kernel data in all studied ASiR levels in the DL method. In the DL method, B0 and S60 performed best in bone and standard convolution kernel, respectively.

Conclusion

The results demonstrated that DL was more susceptible to CT parameter variability than radiomics. Standard convolution kernel images seem to be more appropriate for imaging analysis. Further investigation with a larger sample size is needed.

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.

Systematic analysis of bias and variability of texture measurements in computed tomography

Texture is a key radiomics measurement for quantification of disease and disease progression. The sensitivity of the measurements to image acquisition, however, is uncertain. We assessed bias and variability of computed tomography (CT) texture feature measurements across many clinical image acquisition settings and reconstruction algorithms. Diverse, anatomically informed textures (texture A, B, and C) were simulated across 1188 clinically relevant CT imaging conditions representing four in-plane pixel sizes (0.4, 0.5, 0.7, and 0.9 mm), three slice thicknesses (0.625, 1.25, and 2.5 mm), three dose levels ( CTDI vol 1.90, 3.75, and 7.50 mGy), and 33 reconstruction kernels. Imaging conditions corresponded to noise and resolution properties representative of five commercial scanners (GE LightSpeed VCT, GE Discovery 750 HD, GE Revolution, Siemens Definition Flash, and Siemens Force) in filtered backprojection and iterative reconstruction. About 21 texture features were calculated and compared between the ground-truth phantom (i.e., preimaging) and its corresponding images. Each feature was measured with four unique volumes of interest (VOIs) sizes (244, 579, 1000, and 1953    mm 3 . To characterize the bias, the percentage relative difference [PRD(%)] in each feature was calculated between the imaged scenario and the ground truth for all VOI sizes. Feature variability was assessed in terms of (1)  σ PRD ( % ) indicating the variability between the ground truth and simulated image scenario based on the PRD(%), (2)  COV f indicating the simulation-based variability, and (3)  COV T indicating the natural variability present in the ground-truth phantom. The PRD ranged widely from - 97 % to 1220%, with an underlying variability ( σ ) of up to 241%. Features such as gray-level nonuniformity, texture entropy, sum average, and homogeneity exhibited low susceptibility to reconstruction kernel effects ( PRD < 3 % ) with relatively small σ PRD ( % ) ( ≤ 5 % ) across imaging conditions. The dynamic range of results indicates that image acquisition and reconstruction conditions of in-plane pixel sizes, slice thicknesses, dose levels, and reconstruction kernels can lead to significant bias and variability in feature measurements.

Impact of Variability in Portal Venous Phase Acquisition Timing in Tumor Density Measurement and Treatment Response Assessment: Metastatic Colorectal Cancer as a Paradigm

PURPOSE

New response patterns to anticancer drugs have led tumor size-based response criteria to shift to also include density measurements. Choi criteria, for instance, categorize antiangiogenic therapy response as a decrease in tumor density > 15% at the portal venous phase (PVP). We studied the effect that PVP timing has on measurement of the density of liver metastases (LM) from colorectal cancer (CRC).

METHODS

Pretreatment PVP computed tomography images from 291 patients with LM-CRC from the CRYSTAL trial (Cetuximab Combined With Irinotecan in First-Line Therapy for Metastatic Colorectal Cancer; ClinicalTrials.gov identifier: NCT00154102) were included. Four radiologists independently scored the scans' timing according to a three-point scoring system: early, optimal, late PVP. Using this, we developed, by machine learning, a proprietary computer-aided quality-control algorithm to grade PVP timing. The reference standard was a computer-refined consensus. For each patient, we contoured target liver lesions and calculated their mean density.

RESULTS

Contrast-product administration data were not recorded in the digital imaging and communications in medicine headers for injection volume (94%), type (93%), and route (76%). The PVP timing was early, optimal, and late in 52, 194, and 45 patients, respectively. The mean (95% CI) accuracy of the radiologists for detection of optimal PVP timing was 81.7% (78.3 to 85.2) and was outperformed by the 88.6% (84.8 to 92.4) computer accuracy. The mean ± standard deviation of LM-CRC density was 68 ± 15 Hounsfield units (HU) overall and 59.5 ± 14.9 HU, 71.4 ± 14.1 HU, 62.4 ± 12.5 HU at early, optimal, and late PVP timing, respectively. LM-CRC density was thus decreased at nonoptimal PVP timing by 14.8%: 16.7% at early PVP ( P < .001) and 12.6% at late PVP ( P < .001).

CONCLUSION

Nonoptimal PVP timing should be identified because it significantly decreased tumor density by 14.8%. Our computer-aided quality-control system outperformed the accuracy, reproducibility, and speed of radiologists' visual scoring. PVP-timing scoring could improve the extraction of tumor quantitative imaging biomarkers and the monitoring of anticancer therapy efficacy at the patient and clinical trial levels.

Repeatability of Multiparametric Prostate MRI Radiomics Features

In this study we assessed the repeatability of radiomics features on small prostate tumors using test-retest Multiparametric Magnetic Resonance Imaging (mpMRI). The premise of radiomics is that quantitative image-based features can serve as biomarkers for detecting and characterizing disease. For such biomarkers to be useful, repeatability is a basic requirement, meaning its value must remain stable between two scans, if the conditions remain stable. We investigated repeatability of radiomics features under various preprocessing and extraction configurations including various image normalization schemes, different image pre-filtering, and different bin widths for image discretization. Although we found many radiomics features and preprocessing combinations with high repeatability (Intraclass Correlation Coefficient > 0.85), our results indicate that overall the repeatability is highly sensitive to the processing parameters. Neither image normalization, using a variety of approaches, nor the use of pre-filtering options resulted in consistent improvements in repeatability. We urge caution when interpreting radiomics features and advise paying close attention to the processing configuration details of reported results. Furthermore, we advocate reporting all processing details in radiomics studies and strongly recommend the use of open source implementations.

Development and validation of radiomic signatures of head and neck squamous cell carcinoma molecular features and subtypes

BACKGROUND

Radiomics-based non-invasive biomarkers are promising to facilitate the translation of therapeutically related molecular subtypes for treatment allocation of patients with head and neck squamous cell carcinoma (HNSCC).

METHODS

We included 113 HNSCC patients from The Cancer Genome Atlas (TCGA-HNSCC) project. Molecular phenotypes analyzed were RNA-defined HPV status, five DNA methylation subtypes, four gene expression subtypes and five somatic gene mutations. A total of 540 quantitative image features were extracted from pre-treatment CT scans. Features were selected and used in a regularized logistic regression model to build binary classifiers for each molecular subtype. Models were evaluated using the average area under the Receiver Operator Characteristic curve (AUC) of a stratified 10-fold cross-validation procedure repeated 10 times. Next, an HPV model was trained with the TCGA-HNSCC, and tested on a Stanford cohort (N = 53).

FINDINGS

Our results show that quantitative image features are capable of distinguishing several molecular phenotypes. We obtained significant predictive performance for RNA-defined HPV+ (AUC = 0.73), DNA methylation subtypes MethylMix HPV+ (AUC = 0.79), non-CIMP-atypical (AUC = 0.77) and Stem-like-Smoking (AUC = 0.71), and mutation of NSD1 (AUC = 0.73). We externally validated the HPV prediction model (AUC = 0.76) on the Stanford cohort. When compared to clinical models, radiomic models were superior to subtypes such as NOTCH1 mutation and DNA methylation subtype non-CIMP-atypical while were inferior for DNA methylation subtype CIMP-atypical and NSD1 mutation.

INTERPRETATION

Our study demonstrates that radiomics can potentially serve as a non-invasive tool to identify treatment-relevant subtypes of HNSCC, opening up the possibility for patient stratification, treatment allocation and inclusion in clinical trials. FUND: Dr. Gevaert reports grants from National Institute of Dental & Craniofacial Research (NIDCR) U01 DE025188, grants from National Institute of Biomedical Imaging and Bioengineering of the National Institutes of Health (NIBIB), R01 EB020527, grants from National Cancer Institute (NCI), U01 CA217851, during the conduct of the study; Dr. Huang and Dr. Zhu report grants from China Scholarship Council (Grant NO:201606320087), grants from China Medical Board Collaborating Program (Grant NO:15-216), the Cyrus Tang Foundation, and the Zhejiang University Education Foundation during the conduct of the study; Dr. Cintra reports grants from São Paulo State Foundation for Teaching and Research (FAPESP), during the conduct of the study.

Radiomics: an Introductory Guide to What It May Foretell

PURPOSE OF REVIEW

To briefly review the radiomics concept, its applications, and challenges in oncology in the era of precision medicine.

RECENT FINDINGS

Over the last 5 years, more than 500 studies have evaluated the role of radiomics to predict tumor diagnosis, genetic pattern, tumor response to therapy, and survival in multiple cancers. This new post-processing method is aimed at extracting multiple quantitative features from the image and converting them into mineable data. Radiomics models developed have shown promising results and may play a role in the near future in the daily patient management especially to assess tumor heterogeneity acting as a whole tumor virtual biopsy. For now, radiomics is limited by its lack of standardization; future challenges will be to provide robust and reproducible metrics extracted from large multicenter databases.

Design and fabrication of heterogeneous lung nodule phantoms for assessing the accuracy and variability of measured texture radiomics features in CT

We aimed to design and fabricate synthetic lung nodules with patient-informed internal heterogeneity to assess the variability and accuracy of measured texture features in CT. To that end, 190 lung nodules from a publicly available database of chest CT images (Lung Image Database Consortium) were selected based on size ( > 3    mm ) and malignancy. The texture features of the nodules were used to train a statistical texture synthesis model based on clustered lumpy background. The model parameters were ascertained based on a genetic optimization of a Mahalanobis distance objective function. The resulting texture model defined internal heterogeneity within 24 anthropomorphic lesion models which were subsequently fabricated into physical phantoms using a multimaterial three-dimensional (3-D) printer. The 3-D-printed lesions were imbedded in an anthropomorphic chest phantom and imaged with a clinical scanner using different acquisition parameters including slice thickness, dose level, and reconstruction kernel. The imaged lesions were analyzed in terms of texture features to ascertain the impact of CT imaging on lesion texture quantification. The texture modeling method produced lesion models with low and stable Mahalanobis distance between real and synthetic textures. The virtual lesions were successfully printed into 3-D phantoms. The accuracy and variability of the measured features extracted from the CT images of the phantoms showed notable influence from the imaging acquisition parameters with contrast, energy, and texture entropy exhibiting most sensitivity in terms of accuracy, and contrast, dissimilarity, and texture entropy most variability. Thinner slice thicknesses yielded more accurate and edge reconstruction kernels more stable results. We conclude that printed textured models of lesions can be developed using a method that can target and minimize the mathematical distance between real and synthetic lesions. The synthetic lesions can be used as the basis to investigate how CT imaging conditions might affect radiomics features derived from CT images.

Reliability of CT-based texture features: Phantom study

Objective

To determine the intra‐, inter‐ and test‐retest variability of CT‐based texture analysis (CTTA) metrics.

Materials and methods

In this study, we conducted a series of CT imaging experiments using a texture phantom to evaluate the performance of a CTTA panel on routine abdominal imaging protocols. The phantom comprises of three different regions with various textures found in tumors. The phantom was scanned on two CT scanners viz. the Philips Brilliance 64 CT and Toshiba Aquilion Prime 160 CT scanners. The intra‐scanner variability of the CTTA metrics was evaluated across imaging parameters such as slice thickness, field of view, post‐reconstruction filtering, tube voltage, and tube current. For each scanner and scanning parameter combination, we evaluated the performance of eight different types of texture quantification techniques on a predetermined region of interest (ROI) within the phantom image using 235 different texture metrics. We conducted the repeatability (test‐retest) and robustness (intra‐scanner) test on both the scanners and the reproducibility test was conducted by comparing the inter‐scanner differences in the repeatability and robustness to identify reliable CTTA metrics. Reliable metrics are those metrics that are repeatable, reproducible and robust.

Results

As expected, the robustness, repeatability and reproducibility of CTTA metrics are variably sensitive to various scanner and scanning parameters. Entropy of Fast Fourier Transform‐based texture metrics was overall most reliable across the two scanners and scanning conditions. Post‐processing techniques that reduce image noise while preserving the underlying edges associated with true anatomy or pathology bring about significant differences in radiomic reliability compared to when they were not used.

Conclusion

Following large‐scale validation, identification of reliable CTTA metrics can aid in conducting large‐scale multicenter CTTA analysis using sample sets acquired using different imaging protocols, scanners etc.

Deep Learning-based Image Conversion of CT Reconstruction Kernels Improves Radiomics Reproducibility for Pulmonary Nodules or Masses

Background Intratumor heterogeneity in lung cancer may influence outcomes. CT radiomics seeks to assess tumor features to provide detailed imaging features. However, CT radiomic features vary according to the reconstruction kernel used for image generation. Purpose To investigate the effect of different reconstruction kernels on radiomic features and assess whether image conversion using a convolutional neural network (CNN) could improve reproducibility of radiomic features between different kernels. Materials and Methods In this retrospective analysis, patients underwent non-contrast material-enhanced and contrast material-enhanced axial chest CT with soft kernel (B30f) and sharp kernel (B50f) reconstruction using a single CT scanner from April to June 2017. To convert different kernels without sinogram, the CNN model was developed using residual learning and an end-to-end way. Kernel-converted images were generated, from B30f to B50f and from B50f to B30f. Pulmonary nodules or masses were semiautomatically segmented and 702 radiomic features (tumor intensity, texture, and wavelet features) were extracted. Measurement variability in radiomic features was evaluated using the concordance correlation coefficient (CCC). Results A total of 104 patients were studied, including 54 women and 50 men, with pulmonary nodules or masses (mean age, 63.2 years ± 10.5). The CCC between two readers using the same kernel was 0.92, and 592 of 702 (84.3%) of the radiomic features were reproducible (CCC ≥ 0.85); using different kernels, the CCC was 0.38 and only 107 of 702 (15.2%) of the radiomic features were reliable. Texture features and wavelet features were predominantly affected by reconstruction kernel (CCC, from 0.88 to 0.61 for texture features and from 0.92 to 0.35 for wavelet features). After applying image conversion, CCC improved to 0.84 and 403 of 702 (57.4%) radiomic features were reproducible (CCC, 0.85 for texture features and 0.84 for wavelet features). Conclusion Chest CT image conversion using a convolutional neural network effectively reduced the effect of two different reconstruction kernels and may improve the reproducibility of radiomic features in pulmonary nodules or masses. © RSNA, 2019 Online supplemental material is available for this article. See also the editorial by Park in this issue.

Quantitative Imaging features Improve Discrimination of Malignancy in Pulmonary nodules

Pulmonary nodules are frequently detected radiological abnormalities in lung cancer screening. Nodules of the highest- and lowest-risk for cancer are often easily diagnosed by a trained radiologist there is still a high rate of indeterminate pulmonary nodules (IPN) of unknown risk. Here, we test the hypothesis that computer extracted quantitative features (“radiomics”) can provide improved risk-assessment in the diagnostic setting. Nodules were segmented in 3D and 219 quantitative features are extracted from these volumes. Using these features novel malignancy risk predictors are formed with various stratifications based on size, shape and texture feature categories. We used images and data from the National Lung Screening Trial (NLST), curated a subset of 479 participants (244 for training and 235 for testing) that included incident lung cancers and nodule-positive controls. After removing redundant and non-reproducible features, optimal linear classifiers with area under the receiver operator characteristics (AUROC) curves were used with an exhaustive search approach to find a discriminant set of image features, which were validated in an independent test dataset. We identified several strong predictive models, using size and shape features the highest AUROC was 0.80. Using non-size based features the highest AUROC was 0.85. Combining features from all the categories, the highest AUROC were 0.83.

Radiomics nomogram for differentiating between benign and malignant soft-tissue masses of the extremities

BACKGROUND

Preoperative differentiation between malignant and benign tumors is important for treatment decisions.

PURPOSE/HYPOTHESIS

To investigate/validate a radiomics nomogram for preoperative differentiation between malignant and benign masses.

STUDY TYPE

Retrospective.

POPULATION

Imaging data of 91 patients.

FIELD STRENGTH/SEQUENCE

T₁ -weighted images (570 msec repetition time [TR]; 17.9 msec echo time [TE], 200-400 mm field of view [FOV], 208-512 × 208-512 matrix), fat-suppressed fast-spin-echo (FSE) T₂ -weighted images (T₂ WIs) (4331 msec TR; 87.9 msec TE, 200-400 mm FOV, 312 × 312 matrix), slice thickness 4 mm, and slice spacing 1 mm.

ASSESSMENT

Fat-suppressed FSE T₂ WIs were selected for extraction of features. Radiomics features were extracted from fat-suppressed T₂ WIs. A radiomics signature was generated from the training dataset using least absolute shrinkage and selection operator algorithms. Independent risk factors were identified by multivariate logistic regression analysis and a radiomics nomogram was constructed. Nomogram capability was evaluated in the training dataset and validated in the validation dataset. Performance of the nomogram, radiomics signature, and clinical model were compared.

STATISTICAL TESTS

1) Independent t-test or Mann-Whitney U-test: for continuous variables. Fisher's exact test or χ² test: comparing categorical variables between two groups. Univariate analysis: evaluating associations between clinical/morphological characteristics and malignancy. 2) Least absolute shrinkage and selection operator (LASSO)-logistic regression model: selection of malignancy features. 3) Significant clinical/morphological characteristics and radiomics signature were input variables for multiple logistic regression analysis. Area under the curve (AUC): evaluation of ability of the nomogram to identify malignancy. Hosmer-Lemeshow test and decision curve: evaluation and validation of nomogram results.

RESULTS

The radiomics nomogram was able to differentiate malignancy from benignity in the training and validation datasets with an AUC of 0.94. The nomogram outperformed both the radiomics signature and clinical model alone.

DATA CONCLUSION

This radiomics nomogram is a noninvasive, low-cost preoperative prediction method combining the radiomics signature and clinical model.

LEVEL OF EVIDENCE

3 Technical Efficacy: Stage 2 J. Magn. Reson. Imaging 2020;51:155-163.

NCTN Assessment on Current Applications of Radiomics in Oncology

Radiomics is a fast-growing research area based on converting standard-of-care imaging into quantitative minable data and building subsequent predictive models to personalize treatment. Radiomics has been proposed as a study objective in clinical trial concepts and a potential biomarker for stratifying patients across interventional treatment arms. In recognizing the growing importance of radiomics in oncology, a group of medical physicists and clinicians from NRG Oncology reviewed the current status of the field and identified critical issues, providing a general assessment and early recommendations for incorporation in oncology studies.

Radiogenomics: bridging imaging and genomics

From diagnostics to prognosis to response prediction, new applications for radiomics are rapidly being developed. One of the fastest evolving branches involves linking imaging phenotypes to the tumor genetic profile, a field commonly referred to as "radiogenomics." In this review, a general outline of radiogenomic literature concerning prominent mutations across different tumor sites will be provided. The field of radiogenomics originates from image processing techniques developed decades ago; however, many technical and clinical challenges still need to be addressed. Nevertheless, increasingly accurate and robust radiogenomic models are being presented and the future appears to be bright.

Comparison of prediction models with radiological semantic features and radiomics in lung cancer diagnosis of the pulmonary nodules: a case-control study

PURPOSE

To compare the ability of radiological semantic and quantitative texture features in lung cancer diagnosis of pulmonary nodules.

MATERIALS AND METHODS

A total of N = 121 subjects with confirmed non-small-cell lung cancer were matched with 117 controls based on age and gender. Radiological semantic and quantitative texture features were extracted from CT images with or without contrast enhancement. Three different models were compared using LASSO logistic regression: "CS" using clinical and semantic variables, "T" using texture features, and "CST" using clinical, semantic, and texture variables. For each model, we performed 100 trials of fivefold cross-validation and the average receiver operating curve was accessed. The AUC of the cross-validation study (AUC_CV) was calculated together with its 95% confidence interval.

RESULTS

The AUC_CV (and 95% confidence interval) for models T, CS, and CST was 0.85 (0.71-0.96), 0.88 (0.77-0.96), and 0.88 (0.77-0.97), respectively. After separating the data into two groups with or without contrast enhancement, the AUC (without cross-validation) of the model T was 0.86 both for images with and without contrast enhancement, suggesting that contrast enhancement did not impact the utility of texture analysis.

CONCLUSIONS

The models with semantic and texture features provided cross-validated AUCs of 0.85-0.88 for classification of benign versus cancerous nodules, showing potential in aiding the management of patients.

KEY POINTS

• Pretest probability of cancer can aid and direct the physician in the diagnosis and management of pulmonary nodules in a cost-effective way. • Semantic features (qualitative features reported by radiologists to characterize lung lesions) and radiomic (e.g., texture) features can be extracted from CT images. • Input of these variables into a model can generate a pretest likelihood of cancer to aid clinical decision and management of pulmonary nodules.

Radiomic feature stability across 4D respiratory phases and its impact on lung tumor prognosis prediction

Radiomic analysis has recently demonstrated versatile uses in improving diagnostic and prognostic prediction accuracy for lung cancer. However, since lung tumors are subject to substantial motion due to respiration, the stability of radiomic features over the respiratory cycle of the patient needs to be investigated to better evaluate the robustness of the inter-patient feature variability for clinical applications, and its impact in such applications needs to be assessed. A full panel of 841 radiomic features, including tumor intensity, shape, texture, and wavelet features, were extracted from individual phases of a four-dimensional (4D) computed tomography on 20 early-stage non-small-cell lung cancer (NSCLC) patients. The stability of each radiomic feature was assessed across different phase images of the same patient using the coefficient of variation (COV). The relationship between individual COVs and tumor motion magnitude was inspected. Population COVs, the mean COVs of all 20 patients, were used to evaluate feature motion stability and categorize the radiomic features into 4 different groups. The two extremes, the Very Small group (COV≤5%) and the Large group (COV>20%), each accounted for about a quarter of the features. Shape features were the most stable, with COV≤10% for all features. A clinical study was subsequently conducted using 140 early-stage NSCLC patients. Radiomic features were employed to predict the overall survival with a 500-round bootstrapping. Identical multiple regression model development process was applied, and the model performance was compared between models with and without a feature pre-selection step based on 4D COV to pre-exclude unstable features. Among the systematically tested cutoff values, feature pre-selection with 4D COV≤5% achieved the optimal model performance. The resulting 3-feature radiomic model significantly outperformed its counterpart with no 4D COV pre-selection, with P = 2.16x10^-27 in the one-tailed t-test comparing the prediction performances of the two models.

Matching and Homogenizing Convolution Kernels for Quantitative Studies in Computed Tomography

The sharpness of the kernels used for image reconstruction in computed tomography affects the values of the quantitative image features. We sought to identify the kernels that produce similar feature values to enable a more effective comparison of images produced using scanners from different manufactures. We also investigated a new image filter designed to change the kernel-related component of the frequency spectrum of a postreconstruction image from that of the initial kernel to that of a preferred kernel. A radiomics texture phantom was imaged using scanners from GE, Philips, Siemens, and Toshiba. Images were reconstructed multiple times, varying the kernel from smooth to sharp. The phantom comprised 10 cartridges of various textures. A semiautomated method was used to produce 8 × 2 × 2 cm regions of interest for each cartridge and for all scans. For each region of interest, 38 radiomics features from the categories intensity direct (n = 12), gray-level co-occurrence matrix (n = 21), and neighborhood gray-tone difference matrix (n = 5) were extracted. We then calculated the fractional differences of the features from those of the baseline kernel (GE Standard). To gauge the importance of the differences, we scaled them by the coefficient of variation of the same feature from a cohort of patients with non-small cell lung cancer. The noise power spectra for each kernel were estimated from the phantom's solid acrylic cartridge, and kernel-homogenization filters were developed from these estimates. The Philips C, Siemens B30f, and Toshiba FC24 kernels produced feature values most similar to GE Standard. The kernel homogenization filters reduced the median differences from baseline to less than 1 coefficient of variation in the patient population for all of the GE, Philips, and Siemens kernels except for GE Edge and Toshiba kernels. For prospective computed tomographic radiomics studies, the scanning protocol should specify kernels that have been shown to produce similar feature values. For retrospective studies, kernel homogenization filters can be designed and applied to reduce the kernel-related differences in the feature values.

MRI heterogeneity analysis for prediction of recurrence and disease free survival in anal cancer

BACKGROUND

The aim of this study was to evaluate the role of image heterogeneity analysis of standard care magnetic resonance imaging (MRI) in patients with anal squamous cell carcinoma (ASCC) to predict chemoradiotherapy (CRT) outcome. The ability to predict disease recurrence following CRT has the potential to inform personalized radiotherapy approaches currently being explored in novel clinical trials.

METHODS

An IRB waiver was obtained for retrospective analysis of standard care MRIs from ASCC patients presenting between 2010 and 2014. Whole tumor 3D volume-of-interest (VOI) was outlined on T2-weighted (T2w) and diffusion weighted imaging (DWI) of the pre- and post-treatment scans. Independent imaging features most predictive of disease recurrence were added to the baseline clinico-pathological model and the predictive value of respective extended models was calculated using net reclassification improvement (NRI) algorithm. Cross-validation analysis was carried out to determine percentage error reduction with inclusion of imaging features to the baseline model for both endpoints.

RESULTS

Forty patients who underwent 1.5 T pelvic MRI at baseline and following completion of CRT were included. A combination of two baseline MR heterogeneity features (baseline T2w energy and DWI coefficient of variation) was most predictive of disease recurrence resulting in significant NRI (p = 0 < 0.001). This was confirmed in cross-validation analysis with 34.8% percentage error reduction for the primary endpoint and 18.1% reduction for the secondary endpoint with addition of imaging variables to baseline model.

CONCLUSION

MRI heterogeneity analysis offers complementary information, in addition to clinical staging, in predicting outcome of CRT in anal SCC, warranting validation in larger datasets.

Systematic analysis of bias and variability of morphologic features for lung lesions in computed tomography

We propose to characterize the bias and variability of quantitative morphology features of lung lesions across a range of computed tomography (CT) imaging conditions. A total of 15 lung lesions were simulated (five in each of three spiculation classes: low, medium, and high). For each lesion, a series of simulated CT images representing different imaging conditions were synthesized by applying three-dimensional blur and adding correlated noise based on the measured noise and resolution properties of five commercial multislice CT systems, representing three dose levels ( CTDI vol of 1.90, 3.75, 7.50 mGy), three slice thicknesses (0.625, 1.25, 2.5 mm), and 33 clinical reconstruction kernels from five clinical scanners. The images were segmented using three segmentation algorithms and each algorithm was evaluated by computing a Sørensen-Dice coefficient between the ground truth and the segmentation. A series of 21 shape-based morphology features were extracted from both "ground truth" (i.e., preblur without noise) and "image rendered" lesions (i.e., postblur and with noise). For each morphology feature, the bias was quantified by comparing the percentage relative error in the morphology metric between the imaged lesions and the ground-truth lesions. The variability was characterized by calculating the average coefficient of variation averaged across repeats and imaging conditions. The active contour segmentation had the highest average Dice coefficient of 0.80 followed by 0.63 for threshold, and 0.39 for fuzzy c-means. The bias of the features was segmentation algorithm and feature-dependent, with sharper kernels being less biased and smoother kernels being more biased in general. The feature variability from simulated images ranged from 0.30% to 10% for repeats of the same condition and from 0.74% to 25.3% for different lesions in the same spiculation class. In conclusion, the bias of morphology features is dependent on the acquisition protocol in combination with the segmentation algorithm used and the variability is primarily dependent on the segmentation algorithm.

Radiomics for Classifying Histological Subtypes of Lung Cancer Based on Multiphasic Contrast-Enhanced Computed Tomography

OBJECTIVES

The aim of this study was to evaluate the performance of the radiomics method in classifying lung cancer histological subtypes based on multiphasic contrast-enhanced computed tomography (CT) images.

METHODS

A total of 229 patients with pathologically confirmed lung cancer were retrospectively recruited. All recruited patients underwent nonenhanced and dual-phase chest contrast-enhanced CT; 1160 quantitative radiomics features were calculated to build a radiomics classification model. The performance of the classification models was evaluated by the receiver operating characteristic curve.

RESULTS

The areas under the curve of radiomics models in classifying adenocarcinoma and squamous cell carcinoma, adenocarcinoma and small cell lung cancer, and squamous cell carcinoma and small cell lung cancer were 0.801, 0.857, and 0.657 (nonenhanced); 0.834, 0.855, and 0.619 (arterial phase); and 0.864, 0.864, and 0.664 (venous phase), respectively. Moreover, the application of contrast-enhanced CT may affect the selection of radiomics features.

CONCLUSIONS

Our study indicates that radiomics may be a promising tool for noninvasive predicting histological subtypes of lung cancer based on the multiphasic contrast-enhanced CT images.

Texture analysis of computed tomography data using morphologic and metabolic delineation of esophageal cancer-relation to tumor type and neoadjuvant therapy response

The prognostic values of image-based tumor texture analysis based on computed tomography (CT) and of limiting the segmented tumor volume to metabolically active regions using fludeoxyglucose-positron emission tomography (FDG-PET) were studied in 25 patients with esophageal adenocarcinoma and 11 patients with squamous cell carcinoma. The aims of this study are to describe their CT-image-based texture characteristics before and after neoadjuvant therapy and to evaluate whether limiting the examined tumor volume to metabolically active regions detected with FDG-PET image data would further improve their value. Textural parameters (homogeneity, energy, entropy, contrast, and correlation) based on gray-level co-occurrence matrices (GLCM) were calculated for 3D volumes of segmented esophageal tumors before and after neoadjuvant chemotherapy or radiochemotherapy. Histopathological data after surgical resection and textural parameters before and after neoadjuvant treatment were compared using the Mann-Whitney U test. Significant differences in the textural parameters were observed between adenocarcinoma and squamous cell carcinoma for homogeneity, energy, inertia, and correlation. The use of contrast media during scanning resulted in significant differences in homogeneity, energy, entropy, and inertia for adenocarcinoma but not squamous cell carcinoma. There was also a significant difference in all textural parameters between pathological T status for ypT0-ypT2 and ypT3-ypT4 adenocarcinomas, but not in squamous cell carcinoma patients. No additional value was found from using PET image data to aid segmentation of CT images.