Measuring Computed Tomography Scanner Variability of Radiomics Features. Invest Radiol. 2015;50:757-765. [PMID: 26115366 DOI: 10.1097/rli.0000000000000180]

Impact of hepatocellular carcinoma heterogeneity on computed tomography as a prognostic indicator

We assessed the relationship between the heterogeneity of HCC on preoperative non-contrast-enhanced CT and patient prognosis. The heterogeneity of CT images from 122 patients was assessed and texture feature parameters such as mean, standard deviation (SD), entropy, mean of the positive pixels (MPP), skewness, and kurtosis were obtained using filtration. The relationship between CT texture features and 5-year overall survival (OS) or disease-free survival (DFS) was assessed. Multivariate regression analysis was performed to evaluate the independence of texture feature from clinical or pathological parameters. The Kaplan-Meier curves for OS or DFS was significantly different between patient groups dichotomized by cut-off values for all CT texture parameters with filtration at at least one filter level. Multivariate regression analysis showed the independence of most CT texture parameters on clinical and pathological parameters for OS with filtration at at least one filter level and without filtration except kurtosis. SD, entropy, and MPP with coarse filter, and skewness without filtration showed a significant correlation for DFS. CT texture features of non-contrast-enhanced CT images showed a relationship with HCC prognosis. Multivariate regression analysis showed the possibility of CT texture feature increase the prognostic prediction of HCC by clinical and pathological information.

Radiomics: the bridge between medical imaging and personalized medicine

Radiomics, the high-throughput mining of quantitative image features from standard-of-care medical imaging that enables data to be extracted and applied within clinical-decision support systems to improve diagnostic, prognostic, and predictive accuracy, is gaining importance in cancer research. Radiomic analysis exploits sophisticated image analysis tools and the rapid development and validation of medical imaging data that uses image-based signatures for precision diagnosis and treatment, providing a powerful tool in modern medicine. Herein, we describe the process of radiomics, its pitfalls, challenges, opportunities, and its capacity to improve clinical decision making, emphasizing the utility for patients with cancer. Currently, the field of radiomics lacks standardized evaluation of both the scientific integrity and the clinical relevance of the numerous published radiomics investigations resulting from the rapid growth of this area. Rigorous evaluation criteria and reporting guidelines need to be established in order for radiomics to mature as a discipline. Herein, we provide guidance for investigations to meet this urgent need in the field of radiomics.

Harmonizing the pixel size in retrospective computed tomography radiomics studies

Consistent pixel sizes are of fundamental importance for assessing texture features that relate intensity and spatial information in radiomics studies. To correct for the effects of variable pixel sizes, we combined image resampling with Butterworth filtering in the frequency domain and tested the correction on computed tomography (CT) scans of lung cancer patients reconstructed 5 times with pixel sizes varying from 0.59 to 0.98 mm. One hundred fifty radiomics features were calculated for each preprocessing and field-of-view combination. Intra-patient agreement and inter-patient agreement were compared using the overall concordance correlation coefficient (OCCC). To further evaluate the corrections, hierarchical clustering was used to identify patient scans before and after correction. To assess the general applicability of the corrections, they were applied to 17 CT scans of a radiomics phantom. The reduction in the inter-scanner variability relative to non–small cell lung cancer patient scans was quantified. The variation in pixel sizes caused the intra-patient variability to be large (OCCC <95%) relative to the inter-patient variability in 79% of the features. However, with the resampling and filtering corrections, the intra-patient variability was relatively large in only 10% of the features. With the filtering correction, 8 of 8 patients were correctly clustered, in contrast to only 2 of 8 without the correction. In the phantom study, resampling and filtering the images of a rubber particle cartridge substantially reduced variability in 61% of the radiomics features and substantially increased variability in only 6% of the features. Surprisingly, resampling without filtering tended to increase the variability. In conclusion, applying a correction based on resampling and Butterworth low-pass filtering in the frequency domain effectively reduced variability in CT radiomics features caused by variations in pixel size. This correction may also reduce the variability introduced by other CT scan acquisition parameters.

Development and clinical application of radiomics in lung cancer

Since the discovery of X-rays at the end of the 19^th century, medical imageology has progressed for 100 years, and medical imaging has become an important auxiliary tool for clinical diagnosis. With the launch of the human genome project (HGP) and the development of various high-throughput detection techniques, disease exploration in the post-genome era has extended beyond investigations of structural changes to in-depth analyses of molecular abnormalities in tissues, organs and cells, on the basis of gene expression and epigenetics. These techniques have given rise to genomics, proteomics, metabolomics and other systems biology subspecialties, including radiogenomics. Radiogenomics is an important revolution in the traditional visually identifiable imaging technology and constitutes a new branch, radiomics. Radiomics is aimed at extracting quantitative imaging features automatically and developing models to predict lesion phenotypes in a non-invasive manner. Here, we summarize the advent and development of radiomics, the basic process and challenges in clinical practice, with a focus on applications in pulmonary nodule evaluations, including diagnostics, pathological and molecular classifications, treatment response assessments and prognostic predictions, especially in radiotherapy.

Radiomic machine-learning classifiers for prognostic biomarkers of advanced nasopharyngeal carcinoma

We aimed to identify optimal machine-learning methods for radiomics-based prediction of local failure and distant failure in advanced nasopharyngeal carcinoma (NPC). We enrolled 110 patients with advanced NPC. A total of 970 radiomic features were extracted from MRI images for each patient. Six feature selection methods and nine classification methods were evaluated in terms of their performance. We applied the 10-fold cross-validation as the criterion for feature selection and classification. We repeated each combination for 50 times to obtain the mean area under the curve (AUC) and test error. We observed that the combination methods Random Forest (RF) + RF (AUC, 0.8464 ± 0.0069; test error, 0.3135 ± 0.0088) had the highest prognostic performance, followed by RF + Adaptive Boosting (AdaBoost) (AUC, 0.8204 ± 0.0095; test error, 0.3384 ± 0.0097), and Sure Independence Screening (SIS) + Linear Support Vector Machines (LSVM) (AUC, 0.7883 ± 0.0096; test error, 0.3985 ± 0.0100). Our radiomics study identified optimal machine-learning methods for the radiomics-based prediction of local failure and distant failure in advanced NPC, which could enhance the applications of radiomics in precision oncology and clinical practice.

Preoperative prediction of muscular invasiveness of bladder cancer with radiomic features on conventional MRI and its high-order derivative maps

PURPOSE

To determine radiomic features which are capable of reflecting muscular invasiveness of bladder cancer (BC) and propose a non-invasive strategy for the differentiation of muscular invasiveness preoperatively.

METHODS

Sixty-eight patients with clinicopathologically confirmed BC were included in this retrospective study. A total of 118 cancerous volumes of interest (VOI) were segmented from patients' T2 weighted MR images (T2WI), including 34 non-muscle invasive bladder carcinomas (NMIBCs, stage <T2) and 84 muscle invasive ones (MIBCs, stage ≥T2). The radiomic features quantifying tumor signal intensity and textures were extracted from each VOI and its high-order derivative maps to characterize heterogeneity of tumor tissues. Statistical analysis was used to build radiomic signatures with significant inter-group differences of NMIBCs and MIBCs. The synthetic minority oversampling technique (SMOTE) and a support vector machine (SVM)-based feature selection and classification strategy were proposed to first rebalance the imbalanced sample size and then further select the most predictive and compact signature subset to verify its differentiation capability.

RESULTS

From each tumor VOI, a total of 63 radiomic features were derived and 30 of them showed significant inter-group differences (P ≤ 0.01). By using the SVM-based feature selection algorithm with rebalanced samples, an optimal subset including 13 radiomic signatures was determined. The area under receiver operating characteristic curve and Youden index were improved to 0.8610 and 0.7192, respectively.

CONCLUSION

3D radiomic signatures derived from T2WI and its high-order derivative maps could reflect muscular invasiveness of bladder cancer, and the proposed strategy can be used to facilitate the preoperative prediction of muscular invasiveness in patients with bladder cancer.

Beyond imaging: The promise of radiomics

The domain of investigation of radiomics consists of large-scale radiological image analysis and association with biological or clinical endpoints. The purpose of the present study is to provide a recent update on the status of this rapidly emerging field by performing a systematic review of the literature on radiomics, with a primary focus on oncologic applications. The systematic literature search, performed in Pubmed using the keywords: "radiomics OR radiomic" provided 97 research papers. Based on the results of this search, we describe the methods used for building a model of prognostic value from quantitative analysis of patient images. Then, we provide an up-to-date overview of the results achieved in this field, and discuss the current challenges and future developments of radiomics for oncology.

Infrastructure and distributed learning methodology for privacy-preserving multi-centric rapid learning health care: euroCAT

•

Developed and implemented IT infrastructure in 5 radiation clinics across 3 countries.

•

Proof-of-principle for ‘big data’ infrastructure and distributed learning studies.

•

General framework to execute learning algorithms on distributed data.

Machine learning applications for personalized medicine are highly dependent on access to sufficient data. For personalized radiation oncology, datasets representing the variation in the entire cancer patient population need to be acquired and used to learn prediction models. Ethical and legal boundaries to ensure data privacy hamper collaboration between research institutes. We hypothesize that data sharing is possible without identifiable patient data leaving the radiation clinics and that building machine learning applications on distributed datasets is feasible.

We developed and implemented an IT infrastructure in five radiation clinics across three countries (Belgium, Germany, and The Netherlands). We present here a proof-of-principle for future ‘big data’ infrastructures and distributed learning studies. Lung cancer patient data was collected in all five locations and stored in local databases. Exemplary support vector machine (SVM) models were learned using the Alternating Direction Method of Multipliers (ADMM) from the distributed databases to predict post-radiotherapy dyspnea grade ⩾2. The discriminative performance was assessed by the area under the curve (AUC) in a five-fold cross-validation (learning on four sites and validating on the fifth). The performance of the distributed learning algorithm was compared to centralized learning where datasets of all institutes are jointly analyzed.

The euroCAT infrastructure has been successfully implemented in five radiation clinics across three countries. SVM models can be learned on data distributed over all five clinics. Furthermore, the infrastructure provides a general framework to execute learning algorithms on distributed data. The ongoing expansion of the euroCAT network will facilitate machine learning in radiation oncology. The resulting access to larger datasets with sufficient variation will pave the way for generalizable prediction models and personalized medicine.

Precision of quantitative computed tomography texture analysis using image filtering: A phantom study for scanner variability

Quantitative computed tomography (CT) texture analyses for images with and without filtration are gaining attention to capture the heterogeneity of tumors. The aim of this study was to investigate how quantitative texture parameters using image filtering vary among different computed tomography (CT) scanners using a phantom developed for radiomics studies.A phantom, consisting of 10 different cartridges with various textures, was scanned under 6 different scanning protocols using four CT scanners from four different vendors. CT texture analyses were performed for both unfiltered images and filtered images (using a Laplacian of Gaussian spatial band-pass filter) featuring fine, medium, and coarse textures. Forty-five regions of interest were placed for each cartridge (x) in a specific scan image set (y), and the average of the texture values (T(x,y)) was calculated. The interquartile range (IQR) of T(x,y) among the 6 scans was calculated for a specific cartridge (IQR(x)), while the IQR of T(x,y) among the 10 cartridges was calculated for a specific scan (IQR(y)), and the median IQR(y) was then calculated for the 6 scans (as the control IQR, IQRc). The median of their quotient (IQR(x)/IQRc) among the 10 cartridges was defined as the variability index (VI).The VI was relatively small for the mean in unfiltered images (0.011) and for standard deviation (0.020-0.044) and entropy (0.040-0.044) in filtered images. Skewness and kurtosis in filtered images featuring medium and coarse textures were relatively variable across different CT scanners, with VIs of 0.638-0.692 and 0.430-0.437, respectively.Various quantitative CT texture parameters are robust and variable among different scanners, and the behavior of these parameters should be taken into consideration.

Quantitative computed tomography texture analysis for estimating histological subtypes of thymic epithelial tumors

OBJECTIVES

To investigate whether high-risk thymic epithelial tumor (TET) (HTET) can be differentiated from low-risk TET (LTET) using computed tomography (CT) quantitative texture analysis.

MATERIALS AND METHODS

The data of 39 patients (mean age, 58.6±14.1 years) (39 unenhanced CT (UECT) and 33 contrast-enhanced CT (CECT)) who underwent thymectomy for TET were retrospectively analyzed. A region of interest was placed to include the entire TET within the slice at its maximum diameter. Texture analysis was performed for images with or without a Laplacian of Gaussian filter (with various spatial scaling factors [SSFs]). Two radiologists evaluated the visual heterogeneity of TET using a 3-point scale.

RESULTS

The mean in the unfiltered image (mean0u) and entropy in the filtered image (SSF: 6mm) (entropy6u) for UECT, and the mean in the unfiltered image (mean0c) for CECT were significant parameters for differentiating between HTET and LTET as determined by logistic regression analysis. The area under the receiver operating characteristics curve (AUC) for differentiating HTET from LTET using mean0u, entropy6u, and mean0c was 0.75, 0.76, and 0.89, respectively. And the combination of mean0u and entropy6u allowed AUC of 0.87. Entropy6u provided a higher diagnostic performance compared with visual heterogeneity analysis (p≤0.018).

CONCLUSION

Using CT quantitative texture analysis, HTET can be differentiated from LTET with a high diagnostic performance.

Delta-radiomics features for the prediction of patient outcomes in non-small cell lung cancer

Radiomics is the use of quantitative imaging features extracted from medical images to characterize tumor pathology or heterogeneity. Features measured at pretreatment have successfully predicted patient outcomes in numerous cancer sites. This project was designed to determine whether radiomics features measured from non–small cell lung cancer (NSCLC) change during therapy and whether those features (delta-radiomics features) can improve prognostic models. Features were calculated from pretreatment and weekly intra-treatment computed tomography images for 107 patients with stage III NSCLC. Pretreatment images were used to determine feature-specific image preprocessing. Linear mixed-effects models were used to identify features that changed significantly with dose-fraction. Multivariate models were built for overall survival, distant metastases, and local recurrence using only clinical factors, clinical factors and pretreatment radiomics features, and clinical factors, pretreatment radiomics features, and delta-radiomics features. All of the radiomics features changed significantly during radiation therapy. For overall survival and distant metastases, pretreatment compactness improved the c-index. For local recurrence, pretreatment imaging features were not prognostic, while texture-strength measured at the end of treatment significantly stratified high- and low-risk patients. These results suggest radiomics features change due to radiation therapy and their values at the end of treatment may be indicators of tumor response.

Intrinsic dependencies of CT radiomic features on voxel size and number of gray levels

PURPOSE

Many radiomics features were originally developed for non-medical imaging applications and therefore original assumptions may need to be reexamined. In this study, we investigated the impact of slice thickness and pixel spacing (or pixel size) on radiomics features extracted from Computed Tomography (CT) phantom images acquired with different scanners as well as different acquisition and reconstruction parameters. The dependence of CT texture features on gray-level discretization was also evaluated.

METHODS AND MATERIALS

A texture phantom composed of 10 different cartridges of different materials was scanned on eight different CT scanners from three different manufacturers. The images were reconstructed for various slice thicknesses. For each slice thickness, the reconstruction Field Of View (FOV) was varied to render pixel sizes ranging from 0.39 to 0.98 mm. A fixed spherical region of interest (ROI) was contoured on the images of the shredded rubber cartridge and the 3D printed, 20% fill, acrylonitrile butadiene styrene plastic cartridge (ABS20) for all phantom imaging sets. Radiomic features were extracted from the ROIs using an in-house program. Features categories were: shape (10), intensity (16), GLCM (24), GLZSM (11), GLRLM (11), and NGTDM (5), fractal dimensions (8) and first-order wavelets (128), for a total of 213 features. Voxel-size resampling was performed to investigate the usefulness of extracting features using a suitably chosen voxel size. Acquired phantom image sets were resampled to a voxel size of 1 × 1 × 2 mm³ using linear interpolation. Image features were therefore extracted from resampled and original datasets and the absolute value of the percent coefficient of variation (%COV) for each feature was calculated. Based on the %COV values, features were classified in 3 groups: (1) features with large variations before and after resampling (%COV >50); (2) features with diminished variation (%COV <30) after resampling; and (3) features that had originally moderate variation (%COV <50%) and were negligibly affected by resampling. Group 2 features were further studied by modifying feature definitions to include voxel size. Original and voxel-size normalized features were used for interscanner comparisons. A subsequent analysis investigated feature dependency on gray-level discretization by extracting 51 texture features from ROIs from each of the 10 different phantom cartridges using 16, 32, 64, 128, and 256 gray levels.

RESULTS

Out of the 213 features extracted, 150 were reproducible across voxel sizes, 42 improved significantly (%COV <30, Group 2) after resampling, and 21 had large variations before and after resampling (Group 1). Ten features improved significantly after definition modification effectively removed their voxel-size dependency. Interscanner comparison indicated that feature variability among scanners nearly vanished for 8 of these 10 features. Furthermore, 17 out of 51 texture features were found to be dependent on the number of gray levels. These features were redefined to include the number of gray levels which greatly reduced this dependency.

CONCLUSION

Voxel-size resampling is an appropriate pre-processing step for image datasets acquired with variable voxel sizes to obtain more reproducible CT features. We found that some of the radiomics features were voxel size and gray-level discretization-dependent. The introduction of normalizing factors in their definitions greatly reduced or removed these dependencies.

Variability in CT lung-nodule quantification: Effects of dose reduction and reconstruction methods on density and texture based features

Purpose:

To investigate the effects of dose level and reconstruction method on density and texture based features computed from CT lung nodules.

Methods:

This study had two major components. In the first component, a uniform water phantom was scanned at three dose levels and images were reconstructed using four conventional filtered backprojection (FBP) and four iterative reconstruction (IR) methods for a total of 24 different combinations of acquisition and reconstruction conditions. In the second component, raw projection (sinogram) data were obtained for 33 lung nodules from patients scanned as a part of their clinical practice, where low dose acquisitions were simulated by adding noise to sinograms acquired at clinical dose levels (a total of four dose levels) and reconstructed using one FBP kernel and two IR kernels for a total of 12 conditions. For the water phantom, spherical regions of interest (ROIs) were created at multiple locations within the water phantom on one reference image obtained at a reference condition. For the lung nodule cases, the ROI of each nodule was contoured semiautomatically (with manual editing) from images obtained at a reference condition. All ROIs were applied to their corresponding images reconstructed at different conditions. For 17 of the nodule cases, repeat contours were performed to assess repeatability. Histogram (eight features) and gray level co-occurrence matrix (GLCM) based texture features (34 features) were computed for all ROIs. For the lung nodule cases, the reference condition was selected to be 100% of clinical dose with FBP reconstruction using the B45f kernel; feature values calculated from other conditions were compared to this reference condition. A measure was introduced, which the authors refer to as Q, to assess the stability of features across different conditions, which is defined as the ratio of reproducibility (across conditions) to repeatability (across repeat contours) of each feature.

Results:

The water phantom results demonstrated substantial variability among feature values calculated across conditions, with the exception of histogram mean. Features calculated from lung nodules demonstrated similar results with histogram mean as the most robust feature (Q ≤ 1), having a mean and standard deviation Q of 0.37 and 0.22, respectively. Surprisingly, histogram standard deviation and variance features were also quite robust. Some GLCM features were also quite robust across conditions, namely, diff. variance, sum variance, sum average, variance, and mean. Except for histogram mean, all features have a Q of larger than one in at least one of the 3% dose level conditions.

Conclusions:

As expected, the histogram mean is the most robust feature in their study. The effects of acquisition and reconstruction conditions on GLCM features vary widely, though trending toward features involving summation of product between intensities and probabilities being more robust, barring a few exceptions. Overall, care should be taken into account for variation in density and texture features if a variety of dose and reconstruction conditions are used for the quantification of lung nodules in CT, otherwise changes in quantification results may be more reflective of changes due to acquisition and reconstruction conditions than in the nodule itself.

Texture analysis in radiology: Does the emperor have no clothes?

Texture analysis is more and more frequently used in radiology research. Is this a new technology, and if not, what has changed? Is texture analysis the great diagnostic and prognostic tool we have been searching for in radiology? This commentary answers these questions and places texture analysis into its proper perspective.

Associations of Radiomic Data Extracted from Static and Respiratory-Gated CT Scans with Disease Recurrence in Lung Cancer Patients Treated with SBRT

Radiomics aims to quantitatively capture the complex tumor phenotype contained in medical images to associate them with clinical outcomes. This study investigates the impact of different types of computed tomography (CT) images on the prognostic performance of radiomic features for disease recurrence in early stage non-small cell lung cancer (NSCLC) patients treated with stereotactic body radiation therapy (SBRT). 112 early stage NSCLC patients treated with SBRT that had static free breathing (FB) and average intensity projection (AIP) images were analyzed. Nineteen radiomic features were selected from each image type (FB or AIP) for analysis based on stability and variance. The selected FB and AIP radiomic feature sets had 6 common radiomic features between both image types and 13 unique features. The prognostic performances of the features for distant metastasis (DM) and locoregional recurrence (LRR) were evaluated using the concordance index (CI) and compared with two conventional features (tumor volume and maximum diameter). P-values were corrected for multiple testing using the false discovery rate procedure. None of the FB radiomic features were associated with DM, however, seven AIP radiomic features, that described tumor shape and heterogeneity, were (CI range: 0.638-0.676). Conventional features from FB images were not associated with DM, however, AIP conventional features were (CI range: 0.643-0.658). Radiomic and conventional multivariate models were compared between FB and AIP images using cross validation. The differences between the models were assessed using a permutation test. AIP radiomic multivariate models (median CI = 0.667) outperformed all other models (median CI range: 0.601-0.630) in predicting DM. None of the imaging features were prognostic of LRR. Therefore, image type impacts the performance of radiomic models in their association with disease recurrence. AIP images contained more information than FB images that were associated with disease recurrence in early stage NSCLC patients treated with SBRT, which suggests that AIP images may potentially be more optimal for the development of an imaging biomarker.

Assessing Agreement between Radiomic Features Computed for Multiple CT Imaging Settings

OBJECTIVES

Radiomics utilizes quantitative image features (QIFs) to characterize tumor phenotype. In practice, radiological images are obtained from different vendors' equipment using various imaging acquisition settings. Our objective was to assess the inter-setting agreement of QIFs computed from CT images by varying two parameters, slice thickness and reconstruction algorithm.

MATERIALS AND METHODS

CT images from an IRB-approved/HIPAA-compliant study assessing thirty-two lung cancer patients were included for the analysis. Each scan's raw data were reconstructed into six imaging series using combinations of two reconstruction algorithms (Lung[L] and Standard[S]) and three slice thicknesses (1.25mm, 2.5mm and 5mm), i.e., 1.25L, 1.25S, 2.5L, 2.5S, 5L and 5S. For each imaging-setting, 89 well-defined QIFs were computed for each of the 32 tumors (one tumor per patient). The six settings led to 15 inter-setting comparisons (combinatorial pairs). To reduce QIF redundancy, hierarchical clustering was done. Concordance correlation coefficients (CCCs) were used to assess inter-setting agreement of the non-redundant feature groups. The CCC of each group was assessed by averaging CCCs of QIFs in the group.

RESULTS

Twenty-three non-redundant feature groups were created. Across all feature groups, the best inter-setting agreements (CCCs>0.8) were 1.25S vs 2.5S, 1.25L vs 2.5L, and 2.5S vs 5S; the worst (CCCs<0.51) belonged to 1.25L vs 5S and 2.5L vs 5S. Eight of the feature groups related to size, shape, and coarse texture had an average CCC>0.8 across all imaging settings.

CONCLUSIONS

Varying degrees of inter-setting disagreements of QIFs exist when features are computed from CT images reconstructed using different algorithms and slice thicknesses. Our findings highlight the importance of harmonizing imaging acquisition for obtaining consistent QIFs to study tumor imaging phonotype.

Quantitative radiomics studies for tissue characterization: a review of technology and methodological procedures

Quantitative analysis of tumour characteristics based on medical imaging is an emerging field of research. In recent years, quantitative imaging features derived from CT, positron emission tomography and MR scans were shown to be of added value in the prediction of outcome parameters in oncology, in what is called the radiomics field. However, results might be difficult to compare owing to a lack of standardized methodologies to conduct quantitative image analyses. In this review, we aim to present an overview of the current challenges, technical routines and protocols that are involved in quantitative imaging studies. The first issue that should be overcome is the dependency of several features on the scan acquisition and image reconstruction parameters. Adopting consistent methods in the subsequent target segmentation step is evenly crucial. To further establish robust quantitative image analyses, standardization or at least calibration of imaging features based on different feature extraction settings is required, especially for texture- and filter-based features. Several open-source and commercial software packages to perform feature extraction are currently available, all with slightly different functionalities, which makes benchmarking quite challenging. The number of imaging features calculated is typically larger than the number of patients studied, which emphasizes the importance of proper feature selection and prediction model-building routines to prevent overfitting. Even though many of these challenges still need to be addressed before quantitative imaging can be brought into daily clinical practice, radiomics is expected to be a critical component for the integration of image-derived information to personalize treatment in the future.

Non-small cell lung cancer: quantitative phenotypic analysis of CT images as a potential marker of prognosis

This was a retrospective study to investigate the predictive and prognostic ability of quantitative computed tomography phenotypic features in patients with non-small cell lung cancer (NSCLC). 661 patients with pathological confirmed as NSCLC were enrolled between 2007 and 2014. 592 phenotypic descriptors was automatically extracted on the pre-therapy CT images. Firstly, support vector machine (SVM) was used to evaluate the predictive value of each feature for pathology and TNM clinical stage. Secondly, Cox proportional hazards model was used to evaluate the prognostic value of these imaging signatures selected by SVM which subjected to a primary cohort of 138 patients, and an external independent validation of 61 patients. The results indicated that predictive accuracy for histopathology, N staging, and overall clinical stage was 75.16%, 79.40% and 80.33%, respectively. Besides, Cox models indicated the signatures selected by SVM: “correlation of co-occurrence after wavelet transform” was significantly associated with overall survival in the two datasets (hazard ratio [HR]: 1.65, 95% confidence interval [CI]: 1.41–2.75, p = 0.010; and HR: 2.74, 95%CI: 1.10–6.85, p = 0.027, respectively). Our study indicates that the phenotypic features might provide some insight in metastatic potential or aggressiveness for NSCLC, which potentially offer clinical value in directing personalized therapeutic regimen selection for NSCLC.

Imaging Phenotyping Using Radiomics to Predict Micropapillary Pattern within Lung Adenocarcinoma

INTRODUCTION

Lung adenocarcinomas (ADCs) with a micropapillary pattern have been reported to have a poor prognosis. However, few studies have reported on the imaging-based identification of a micropapillary component, and all of them have been subjective studies dealing with qualitative computed tomography variables. We aimed to explore imaging phenotyping using a radiomics approach for predicting a micropapillary pattern within lung ADC.

METHODS

We enrolled 339 patients who underwent complete resection for lung ADC. Histologic subtypes and grades of the ADC were classified. The amount of micropapillary component was determined. Clinical features and conventional imaging variables such as tumor disappearance rate and maximum standardized uptake value on positron emission tomography were assessed. Quantitative computed tomography analysis was performed on the basis of histogram, size and shape, Gray level co-occurrence matrix-based features, and intensity variance and size zone variance-based features.

RESULTS

Higher tumor stage (OR = 3.270, 95% confidence interval [CI]: 1.483-7.212), intermediate grade (OR = 2.977, 95% CI: 1.066-8.316), lower value of the minimum of the whole pixel value (OR = 0.725, 95% CI: 0.527-0.98800), and lower value of the variance of the positive pixel value (OR = 0.961, 95% CI: 0.927-0.997) were identified as being predictive of a micropapillary component within lung ADC. On the other hand, maximum standardized uptake value and tumor disappearance rate were not significantly different in groups with a micropapillary pattern constituting at least 5% or less than 5% of the entire tumor.

CONCLUSION

A radiomics approach can be used to interrogate an entire tumor in a noninvasive manner. Combining imaging parameters with clinical features can provide added diagnostic value to identify the presence of a micropapillary component and thus, can influence proper treatment planning.

Stability of radiomic features in CT perfusion maps

This study aimed to identify a set of stable radiomic parameters in CT perfusion (CTP) maps with respect to CTP calculation factors and image discretization, as an input for future prognostic models for local tumor response to chemo-radiotherapy. Pre-treatment CTP images of eleven patients with oropharyngeal carcinoma and eleven patients with non-small cell lung cancer (NSCLC) were analyzed. 315 radiomic parameters were studied per perfusion map (blood volume, blood flow and mean transit time). Radiomics robustness was investigated regarding the potentially standardizable (image discretization method, Hounsfield unit (HU) threshold, voxel size and temporal resolution) and non-standardizable (artery contouring and noise threshold) perfusion calculation factors using the intraclass correlation (ICC). To gain added value for our model radiomic parameters correlated with tumor volume, a well-known predictive factor for local tumor response to chemo-radiotherapy, were excluded from the analysis. The remaining stable radiomic parameters were grouped according to inter-parameter Spearman correlations and for each group the parameter with the highest ICC was included in the final set. The acceptance level was 0.9 and 0.7 for the ICC and correlation, respectively. The image discretization method using fixed number of bins or fixed intervals gave a similar number of stable radiomic parameters (around 40%). The potentially standardizable factors introduced more variability into radiomic parameters than the non-standardizable ones with 56-98% and 43-58% instability rates, respectively. The highest variability was observed for voxel size (instability rate  >97% for both patient cohorts). Without standardization of CTP calculation factors none of the studied radiomic parameters were stable. After standardization with respect to non-standardizable factors ten radiomic parameters were stable for both patient cohorts after correction for inter-parameter correlations. Voxel size, image discretization, HU threshold and temporal resolution have to be standardized to build a reliable predictive model based on CTP radiomics analysis.

Impact of Reconstruction Algorithms on CT Radiomic Features of Pulmonary Tumors: Analysis of Intra- and Inter-Reader Variability and Inter-Reconstruction Algorithm Variability

PURPOSE

To identify the impact of reconstruction algorithms on CT radiomic features of pulmonary tumors and to reveal and compare the intra- and inter-reader and inter-reconstruction algorithm variability of each feature.

METHODS

Forty-two patients (M:F = 19:23; mean age, 60.43±10.56 years) with 42 pulmonary tumors (22.56±8.51mm) underwent contrast-enhanced CT scans, which were reconstructed with filtered back projection and commercial iterative reconstruction algorithm (level 3 and 5). Two readers independently segmented the whole tumor volume. Fifteen radiomic features were extracted and compared among reconstruction algorithms. Intra- and inter-reader variability and inter-reconstruction algorithm variability were calculated using coefficients of variation (CVs) and then compared.

RESULTS

Among the 15 features, 5 first-order tumor intensity features and 4 gray level co-occurrence matrix (GLCM)-based features showed significant differences (p<0.05) among reconstruction algorithms. As for the variability, effective diameter, sphericity, entropy, and GLCM entropy were the most robust features (CV≤5%). Inter-reader variability was larger than intra-reader or inter-reconstruction algorithm variability in 9 features. However, for entropy, homogeneity, and 4 GLCM-based features, inter-reconstruction algorithm variability was significantly greater than inter-reader variability (p<0.013).

CONCLUSIONS

Most of the radiomic features were significantly affected by the reconstruction algorithms. Inter-reconstruction algorithm variability was greater than inter-reader variability for entropy, homogeneity, and GLCM-based features.

Effects of contrast-enhancement, reconstruction slice thickness and convolution kernel on the diagnostic performance of radiomics signature in solitary pulmonary nodule

The Effects of contrast-enhancement, reconstruction slice thickness and convolution kernel on the diagnostic performance of radiomics signature in solitary pulmonary nodule (SPN) remains unclear. 240 patients with SPNs (malignant, n = 180; benign, n = 60) underwent non-contrast CT (NECT) and contrast-enhanced CT (CECT) which were reconstructed with different slice thickness and convolution kernel. 150 radiomics features were extracted separately from each set of CT and diagnostic performance of each feature were assessed. After feature selection and radiomics signature construction, diagnostic performance of radiomics signature for discriminating benign and malignant SPN was also assessed with respect to the discrimination and classification and compared with net reclassification improvement (NRI). Our results showed NECT-based radiomics signature demonstrated better discrimination and classification capability than CECT in both primary (AUC: 0.862 vs. 0.829, p = 0.032; NRI = 0.578) and validation cohort (AUC: 0.750 vs. 0.735, p = 0.014; NRI = 0.023). Thin-slice (1.25 mm) CT-based radiomics signature had better diagnostic performance than thick-slice CT (5 mm) in both primary (AUC: 0.862 vs. 0.785, p = 0.015; NRI = 0.867) and validation cohort (AUC: 0.750 vs. 0.725, p = 0.025; NRI = 0.467). Standard convolution kernel-based radiomics signature had better diagnostic performance than lung convolution kernel-based CT in both primary (AUC: 0.785 vs. 0.770, p = 0.015; NRI = 0.156) and validation cohort (AUC: 0.725 vs.0.686, p = 0.039; NRI = 0.467). Therefore, this study indicates that the contrast-enhancement, reconstruction slice thickness and convolution kernel can affect the diagnostic performance of radiomics signature in SPN, of which non-contrast, thin-slice and standard convolution kernel-based CT is more informative.

Can radiomics features be reproducibly measured from CBCT images for patients with non-small cell lung cancer?

PURPOSE

Increasing evidence suggests radiomics features extracted from computed tomography (CT) images may be useful in prognostic models for patients with nonsmall cell lung cancer (NSCLC). This study was designed to determine whether such features can be reproducibly obtained from cone-beam CT (CBCT) images taken using medical Linac onboard-imaging systems in order to track them through treatment.

METHODS

Test-retest CBCT images of ten patients previously enrolled in a clinical trial were retrospectively obtained and used to determine the concordance correlation coefficient (CCC) for 68 different texture features. The volume dependence of each feature was also measured using the Spearman rank correlation coefficient. Features with a high reproducibility (CCC > 0.9) that were not due to volume dependence in the patient test-retest set were further examined for their sensitivity to differences in imaging protocol, level of scatter, and amount of motion by using two phantoms. The first phantom was a texture phantom composed of rectangular cartridges to represent different textures. Features were measured from two cartridges, shredded rubber and dense cork, in this study. The texture phantom was scanned with 19 different CBCT imagers to establish the features' interscanner variability. The effect of scatter on these features was studied by surrounding the same texture phantom with scattering material (rice and solid water). The effect of respiratory motion on these features was studied using a dynamic-motion thoracic phantom and a specially designed tumor texture insert of the shredded rubber material. The differences between scans acquired with different Linacs and protocols, varying amounts of scatter, and with different levels of motion were compared to the mean intrapatient difference from the test-retest image set.

RESULTS

Of the original 68 features, 37 had a CCC >0.9 that was not due to volume dependence. When the Linac manufacturer and imaging protocol were kept consistent, 4-13 of these 37 features passed our criteria for reproducibility more than 50% of the time, depending on the manufacturer-protocol combination. Almost all of the features changed substantially when scatter material was added around the phantom. For the dense cork, 23 features passed in the thoracic scans and 11 features passed in the head scans when the differences between one and two layers of scatter were compared. Using the same test for the shredded rubber, five features passed the thoracic scans and eight features passed the head scans. Motion substantially impacted the reproducibility of the features. With 4 mm of motion, 12 features from the entire volume and 14 features from the center slice measurements were reproducible. With 6-8 mm of motion, three features (Laplacian of Gaussian filtered kurtosis, gray-level nonuniformity, and entropy), from the entire volume and seven features (coarseness, high gray-level run emphasis, gray-level nonuniformity, sum-average, information measure correlation, scaled mean, and entropy) from the center-slice measurements were considered reproducible.

CONCLUSIONS

Some radiomics features are robust to the noise and poor image quality of CBCT images when the imaging protocol is consistent, relative changes in the features are used, and patients are limited to those with less than 1 cm of motion.

Radiomics in head and neck cancer: from exploration to application

In the context of clinical oncology, a fundamental goal of radiomics is the extraction of large amounts of quantitative features whose subsequent analysis can be used for decision support towards personalized and actionable cancer care. Head and neck cancers present a unique set of diagnostic and therapeutic challenges by nature of its complex anatomy and heterogeneity. Radiomics holds the potential to address these barriers, but only if as a collective field we direct future effort towards investigating specific oncologic function and oncologic outcomes, with external validation and collaborative multi-institutional efforts to begin standardizing and refining radiomic signatures. Here we present an overview of radiomic texture analysis methods as well as the software infrastructure, review the developments of radiomics in head and neck cancer applications, discuss unmet challenges, and propose key recommendations for moving the field forward.

Tumour heterogeneity in glioblastoma assessed by MRI texture analysis: a potential marker of survival

OBJECTIVE:

The main objective of this retrospective work was the study of three-dimensional (3D) heterogeneity measures of post-contrast pre-operative MR images acquired with T₁ weighted sequences of patients with glioblastoma (GBM) as predictors of clinical outcome.

METHODS:

79 patients from 3 hospitals were included in the study. 16 3D textural heterogeneity measures were computed including run-length matrix (RLM) features (regional heterogeneity) and co-occurrence matrix (CM) features (local heterogeneity). The significance of the results was studied using Kaplan-Meier curves and Cox proportional hazards analysis. Correlation between the variables of the study was assessed using the Spearman's correlation coefficient.

RESULTS:

Kaplan-Meyer survival analysis showed that 4 of the 11 RLM features and 4 of the 5 CM features considered were robust predictors of survival. The median survival differences in the most significant cases were of over 6 months.

CONCLUSION:

Heterogeneity measures computed on the post-contrast pre-operative T₁ weighted MR images of patients with GBM are predictors of survival.

ADVANCES IN KNOWLEDGE:

Texture analysis to assess tumour heterogeneity has been widely studied. However, most works develop a two-dimensional analysis, focusing only on one MRI slice to state tumour heterogeneity. The study of fully 3D heterogeneity textural features as predictors of clinical outcome is more robust and is not dependent on the selected slice of the tumour.

Exploratory Study to Identify Radiomics Classifiers for Lung Cancer Histology

Background

Radiomics can quantify tumor phenotypic characteristics non-invasively by applying feature algorithms to medical imaging data. In this study of lung cancer patients, we investigated the association between radiomic features and the tumor histologic subtypes (adenocarcinoma and squamous cell carcinoma). Furthermore, in order to predict histologic subtypes, we employed machine-learning methods and independently evaluated their prediction performance.

Methods

Two independent radiomic cohorts with a combined size of 350 patients were included in our analysis. A total of 440 radiomic features were extracted from the segmented tumor volumes of pretreatment CT images. These radiomic features quantify tumor phenotypic characteristics on medical images using tumor shape and size, intensity statistics, and texture. Univariate analysis was performed to assess each feature’s association with the histological subtypes. In our multivariate analysis, we investigated 24 feature selection methods and 3 classification methods for histology prediction. Multivariate models were trained on the training cohort and their performance was evaluated on the independent validation cohort using the area under ROC curve (AUC). Histology was determined from surgical specimen.

Results

In our univariate analysis, we observed that fifty-three radiomic features were significantly associated with tumor histology. In multivariate analysis, feature selection methods ReliefF and its variants showed higher prediction accuracy as compared to other methods. We found that Naive Baye’s classifier outperforms other classifiers and achieved the highest AUC (0.72; p-value = 2.3 × 10⁻⁷) with five features: Stats_min, Wavelet_HLL_rlgl_lowGrayLevelRunEmphasis, Wavelet_HHL_stats_median, Wavelet_HLL_stats_skewness, and Wavelet_HLH_glcm_clusShade.

Conclusion

Histological subtypes can influence the choice of a treatment/therapy for lung cancer patients. We observed that radiomic features show significant association with the lung tumor histology. Moreover, radiomics-based multivariate classifiers were independently validated for the prediction of histological subtypes. Despite achieving lower than optimal prediction accuracy (AUC 0.72), our analysis highlights the impressive potential of non-invasive and cost-effective radiomics for precision medicine. Further research in this direction could lead us to optimal performance and therefore to clinical applicability, which could enhance the efficiency and efficacy of cancer care.

Radiomic Machine-Learning Classifiers for Prognostic Biomarkers of Head and Neck Cancer

INTRODUCTION

"Radiomics" extracts and mines a large number of medical imaging features in a non-invasive and cost-effective way. The underlying assumption of radiomics is that these imaging features quantify phenotypic characteristics of an entire tumor. In order to enhance applicability of radiomics in clinical oncology, highly accurate and reliable machine-learning approaches are required. In this radiomic study, 13 feature selection methods and 11 machine-learning classification methods were evaluated in terms of their performance and stability for predicting overall survival in head and neck cancer patients.

METHODS

Two independent head and neck cancer cohorts were investigated. Training cohort HN1 consisted of 101 head and neck cancer patients. Cohort HN2 (n = 95) was used for validation. A total of 440 radiomic features were extracted from the segmented tumor regions in CT images. Feature selection and classification methods were compared using an unbiased evaluation framework.

RESULTS

We observed that the three feature selection methods minimum redundancy maximum relevance (AUC = 0.69, Stability = 0.66), mutual information feature selection (AUC = 0.66, Stability = 0.69), and conditional infomax feature extraction (AUC = 0.68, Stability = 0.7) had high prognostic performance and stability. The three classifiers BY (AUC = 0.67, RSD = 11.28), RF (AUC = 0.61, RSD = 7.36), and NN (AUC = 0.62, RSD = 10.52) also showed high prognostic performance and stability. Analysis investigating performance variability indicated that the choice of classification method is the major factor driving the performance variation (29.02% of total variance).

CONCLUSION

Our study identified prognostic and reliable machine-learning methods for the prediction of overall survival of head and neck cancer patients. Identification of optimal machine-learning methods for radiomics-based prognostic analyses could broaden the scope of radiomics in precision oncology and cancer care.