Test-retest reproducibility analysis of lung CT image features

Stability of Radiomic Features against Variations in Lesion Segmentations Computed on Apparent Diffusion Coefficient Maps of Breast Lesions

Diffusion-weighted imaging (DWI) combined with radiomics can aid in the differentiation of breast lesions. Segmentation characteristics, however, might influence radiomic features. To evaluate feature stability, we implemented a standardized pipeline featuring shifts and shape variations of the underlying segmentations. A total of 103 patients were retrospectively included in this IRB-approved study after multiparametric diagnostic breast 3T MRI with a spin-echo diffusion-weighted sequence with echoplanar readout (b-values: 50, 750 and 1500 s/mm2). Lesion segmentations underwent shifts and shape variations, with >100 radiomic features extracted from apparent diffusion coefficient (ADC) maps for each variation. These features were then compared and ranked based on their stability, measured by the Overall Concordance Correlation Coefficient (OCCC) and Dynamic Range (DR). Results showed variation in feature robustness to segmentation changes. The most stable features, excluding shape-related features, were FO (Mean, Median, RootMeanSquared), GLDM (DependenceNonUniformity), GLRLM (RunLengthNonUniformity), and GLSZM (SizeZoneNonUniformity), which all had OCCC and DR > 0.95 for both shifting and resizing the segmentation. Perimeter, MajorAxisLength, MaximumDiameter, PixelSurface, MeshSurface, and MinorAxisLength were the most stable features in the Shape category with OCCC and DR > 0.95 for resizing. Considering the variability in radiomic feature stability against segmentation variations is relevant when interpreting radiomic analysis of breast DWI data.

A CT texture-based nomogram for predicting futile reperfusion in patients with intraparenchymal hyperdensity after endovascular thrombectomy for acute anterior circulation large vessel occlusion

Background

Post-thrombectomy intraparenchymal hyperdensity (PTIH) in patients with acute anterior circulation large vessel occlusion is a common CT sign associated with a higher incidence of futile reperfusion (FR). We aimed to develop a nomogram to predict FR specifically in patients with PTIH.

Methods

We retrospectively collected information on patients with acute ischemic stroke who underwent endovascular thrombectomy (EVT) at two stroke centers. A total of 398 patients with PTIH were included to develop and validate the nomogram, including 214 patients in the development cohort, 92 patients in the internal validation cohort and 92 patients in the external validation cohort. The nomogram was developed according to the independent predictors obtained from multivariate logistic regression analysis, including clinical factors and CT texture features extracted from hyperdense areas on CT images within half an hour after EVT. The performance of the nomogram was evaluated with integrated discrimination improvement (IDI), category-free net reclassification improvement (NRI), the area under the receiver operating characteristic curve (AUC-ROC), calibration plots, and decision curve analyses for discrimination, calibration ability, and clinical net benefits, respectively.

Results

Our nomogram was constructed based on three clinical factors (age, NIHSS score and ASPECT score) and two CT texture features (entropy and kurtosis), with AUC-ROC of 0.900, 0.897, and 0.870 in the development, internal validation, and external validation cohorts, respectively. NRI and IDI further validated the superior predictive ability of the nomogram compared to the clinical model. The calibration plot revealed good consistency between the predicted and the actual outcome. The decision curve indicated good positive net benefit and clinical validity of the nomogram.

Conclusion

The nomogram enables clinicians to accurately predict FR specifically in patients with PTIH within half an hour after EVT and helps to formulate more appropriate treatment plans in the early post-EVT period.

Protocol selection formalism for minimizing detectable differences in morphological radiomics features of lung lesions in repeated CT acquisitions

Background

The accuracy of morphological radiomic features (MRFs) can be affected by various acquisition settings and imaging conditions. To ensure that clinically irrelevant changes do not reduce sensitivity to capture the radiomics changes between successive acquisitions, it is essential to determine the optimal imaging systems and protocols to use.

Purpose

The main goal of our study was to optimize CT protocols and minimize the minimum detectable difference (MDD) in successive acquisitions of MRFs.

Method

MDDs were derived based on the previous research involving 15 realizations of nodule models at two different sizes. Our study involved simulations of two consecutive acquisitions using 297 different imaging conditions, representing variations in scanners' reconstruction kernels, dose levels, and slice thicknesses. Parametric polynomial models were developed to establish correlations between imaging system characteristics, lesion size, and MDDs. Additionally, polynomial models were used to model the correlation of the imaging system parameters. Optimization problems were formulated for each MRF to minimize the approximated function. Feature importance was determined for each MRF through permutation feature analysis. The proposed method was compared to the recommended guidelines by the quantitative imaging biomarkers alliance (QIBA).

Results

The feature importance analysis showed that lesion size is the most influential parameter to estimate the MDDs in most of the MRFs. Our study revealed that thinner slices and higher doses had a measurable impact on reducing the MDDs. Higher spatial resolution and lower noise magnitude were identified as the most suitable or noninferior acquisition settings. Compared to QIBA, the proposed protocol selection guideline demonstrated a reduced coefficient of variation, with values decreasing from 1.49 to 1.11 for large lesions and from 1.68 to 1.12 for small lesions.

Conclusion

The protocol optimization framework provides means to assess and optimize protocols to minimize the MDD to increase the sensitivity of the measurements in lung cancer screening.

CT Texture-Based Nomogram in Ischemic Stroke to Differentiate Intracerebral Hemorrhage from Contrast Extravasation after Thrombectomy

INTRODUCTION

Post-thrombectomy intraparenchymal hyperdensity (PTIH) in patients with acute ischemic stroke is a common CT sign, making it difficult for physicians to distinguish intracerebral hemorrhage in the early post-thrombectomy period. The aim of this study was to develop an effective model to differentiate intracerebral hemorrhage from contrast extravasation in patients with PTIH.

METHODS

We retrospectively collected information on patients who underwent endovascular thrombectomy at two stroke centers between August 2017 and January 2023. A total of 222 patients were included in the study, including 118 patients in the development cohort, 52 patients in the internal validation cohort, and 52 patients in the external validation cohort. The nomogram was constructed using R software based on independent predictors derived from the multivariate logistic regression analysis, including clinical factors and CT texture features extracted from hyperdense areas on CT images. The performance and accuracy of the derived nomogram were assessed by the area under the receiver operating characteristic curve (AUC-ROC) and calibration curves. Additionally, decision curve analysis was conducted to appraise the clinical utility of the nomogram.

RESULTS

Our nomogram was derived from two clinical factors (ASPECT score and onset to reperfusion time) and two CT texture features (variance and uniformity), with AUC-ROC of 0.943, 0.930, and 0.937 in the development, internal validation, and external validation cohorts, respectively. Furthermore, the calibration plot exhibited a strong agreement between the predicted outcome and the actual outcome. In addition, the decision curve analysis revealed the clinical utility of the nomogram in accurately predicting hemorrhage in patients with PTIH.

CONCLUSION

The developed nomogram, based on clinical factors and CT texture features, proves to be effective in distinguishing intracerebral hemorrhage from contrast extravasation in patients with PTIH.

Radiomics and liver: Where we are and where we are headed?

Hepatic diffuse conditions and focal liver lesions represent two of the most common scenarios to face in everyday radiological clinical practice. Thanks to the advances in technology, radiology has gained a central role in the management of patients with liver disease, especially due to its high sensitivity and specificity. Since the introduction of computed tomography (CT) and magnetic resonance imaging (MRI), radiology has been considered the non-invasive reference modality to assess and characterize liver pathologies. In recent years, clinical practice has moved forward to a quantitative approach to better evaluate and manage each patient with a more fitted approach. In this setting, radiomics has gained an important role in helping radiologists and clinicians characterize hepatic pathological entities, in managing patients, and in determining prognosis. Radiomics can extract a large amount of data from radiological images, which can be associated with different liver scenarios. Thanks to its wide applications in ultrasonography (US), CT, and MRI, different studies were focused on specific aspects related to liver diseases. Even if broadly applied, radiomics has some advantages and different pitfalls. This review aims to summarize the most important and robust studies published in the field of liver radiomics, underlying their main limitations and issues, and what they can add to the current and future clinical practice and literature.

Radiomics signature based on robust features derived from diffusion data for differentiation between benign and malignant solitary pulmonary lesions

BACKGROUND

Classifying and characterizing pulmonary lesions are critical for clinical decision-making process to identify optimal therapeutic strategies. The purpose of this study was to develop and validate a radiomics nomogram for distinguishing between benign and malignant pulmonary lesions based on robust features derived from diffusion images.

MATERIAL AND METHODS

The study was conducted in two phases. In the first phase, we prospectively collected 30 patients with pulmonary nodule/mass who underwent twice EPI-DWI scans. The robustness of features between the two scans was evaluated using the concordance correlation coefficient (CCC) and dynamic range (DR). In the second phase, 139 patients who underwent pulmonary DWI were randomly divided into training and test sets in a 7:3 ratio. Maximum relevance minimum redundancy, least absolute shrinkage and selection operator, and logistic regression were used for feature selection and construction of radiomics signatures. Nomograms were established incorporating clinical features, radiomics signatures, and ADC(0, 800). The diagnostic efficiency of different models was evaluated using the area under the curve (AUC) and decision curve analysis.

RESULTS

Among the features extracted from DWI and ADC images, 42.7% and 37.4% were stable (both CCC and DR ≥ 0.85). The AUCs for distinguishing pulmonary lesions in the test set for clinical model, ADC, ADC radiomics signatures, and DWI radiomics signatures were 0.694, 0.802, 0.885, and 0.767, respectively. The nomogram exhibited the best differentiation performance (AUC = 0.923). The decision curve showed that the nomogram consistently outperformed ADC value and clinical model in lesion differentiation.

CONCLUSION

Our study demonstrates the robustness of radiomics features derived from lung DWI. The ADC radiomics nomogram shows superior clinical net benefits compared to conventional clinical models or ADC values alone in distinguishing solitary pulmonary lesions, offering a promising tool for noninvasive, precision diagnosis in lung cancer.

State of the art: radiomics and radiomics-related artificial intelligence on the road to clinical translation

Radiomics and artificial intelligence carry the promise of increased precision in oncologic imaging assessments due to the ability of harnessing thousands of occult digital imaging features embedded in conventional medical imaging data. While powerful, these technologies suffer from a number of sources of variability that currently impede clinical translation. In order to overcome this impediment, there is a need to control for these sources of variability through harmonization of imaging data acquisition across institutions, construction of standardized imaging protocols that maximize the acquisition of these features, harmonization of post-processing techniques, and big data resources to properly power studies for hypothesis testing. For this to be accomplished, it will be critical to have multidisciplinary and multi-institutional collaboration.

Volumetric analysis: Rethinking brain metastases response assessment

Background

The Response Assessment in Neuro-Oncology for Brain Metastases (RANO-BM) criteria are the gold standard for assessing brain metastases (BMs) treatment response. However, they are limited by their reliance on 1D, despite the routine use of high-resolution T1-weighted MRI scans for BMs, which allows for 3D measurements. Our study aimed to investigate whether volumetric measurements could improve the response assessment in patients with BMs.

Methods

We retrospectively evaluated a dataset comprising 783 BMs and analyzed the response of 185 of them from 132 patients who underwent stereotactic radiotherapy between 2007 and 2021 at 5 hospitals. We used T1-weighted MRIs to compute the volume of the lesions. For the volumetric criteria, progressive disease was defined as at least a 30% increase in volume, and partial response was characterized by a 20% volume reduction.

Results

Our study showed that the proposed volumetric criteria outperformed the RANO-BM criteria in several aspects: (1) Evaluating every lesion, while RANO-BM failed to evaluate 9.2% of them. (2) Classifying response effectively in 140 lesions, compared to only 72 lesions classified by RANO-BM. (3) Identifying BM recurrences a median of 3.3 months earlier than RANO-BM criteria.

Conclusions

Our study demonstrates the superiority of volumetric criteria in improving the response assessment of BMs compared to the RANO-BM criteria. Our proposed criteria allow for evaluation of every lesion, regardless of its size or shape, better classification, and enable earlier identification of progressive disease. Volumetric criteria provide a standardized, reliable, and objective tool for assessing treatment response.

Impact of Image Enhancement on the Radiomics Stability of Diffusion-Weighted MRI Images of Cervical Cancer

The diffusion-weighted imaging (DWI) technique is known for its capability to differentiate the diffusion of water molecules between cancerous and non-cancerous cervix tissues, which enhances the accuracy of detection. Despite the potential of DWI-MRI, its accuracy is limited by technical factors influencing in vivo data acquisition, thus impacting the quantification of radiomics features. This study aimed to measure the radiomics stability of manual and semi-automated segmentation on contrast limited adaptive histogram equalization (CLAHE)-enhanced DWI-MRI cervical images. Eighty diffusion-weighted MRI images were obtained from patients diagnosed with cervical cancer, and an active contour model was used to analyze the data. Radiomics analysis was conducted to extract the first statistical order, shape, and textural features with intraclass correlation coefficient (ICC) measurement. The results of the CLAHE segmentation approach showed a marked improvement when compared to the manual and semi-automated segmentation methods, with an ICC value of 0.990 ± 0.005 (p<0.05), compared to 0.864 ± 0.033 (p<0.05) and 0.554 ± 0.185 (p>0.05), respectively. The CLAHE segmentation displayed a higher level of robustness than the manual groups in terms of the features present in both categories. Thus, CLAHE segmentation is owing to its potential to generate radiomics features that are more durable and consistent.

Predicting tumor deposits in rectal cancer: a combined deep learning model using T2-MR imaging and clinical features

BACKGROUND

Tumor deposits (TDs) are associated with poor prognosis in rectal cancer (RC). This study aims to develop and validate a deep learning (DL) model incorporating T2-MR image and clinical factors for the preoperative prediction of TDs in RC patients.

METHODS AND METHODS

A total of 327 RC patients with pathologically confirmed TDs status from January 2016 to December 2019 were retrospectively recruited, and the T2-MR images and clinical variables were collected. Patients were randomly split into a development dataset (n = 246) and an independent testing dataset (n = 81). A single-channel DL model, a multi-channel DL model, a hybrid DL model, and a clinical model were constructed. The performance of these predictive models was assessed by using receiver operating characteristics (ROC) analysis and decision curve analysis (DCA).

RESULTS

The areas under the curves (AUCs) of the clinical, single-DL, multi-DL, and hybrid-DL models were 0.734 (95% CI, 0.674-0.788), 0.710 (95% CI, 0.649-0.766), 0.767 (95% CI, 0.710-0.819), and 0.857 (95% CI, 0.807-0.898) in the development dataset. The AUC of the hybrid-DL model was significantly higher than the single-DL and multi-DL models (both p < 0.001) in the development dataset, and the single-DL model (p = 0.028) in the testing dataset. Decision curve analysis demonstrated the hybrid-DL model had higher net benefit than other models across the majority range of threshold probabilities.

CONCLUSIONS

The proposed hybrid-DL model achieved good predictive efficacy and could be used to predict tumor deposits in rectal cancer.

CRITICAL RELEVANCE STATEMENT

The proposed hybrid-DL model achieved good predictive efficacy and could be used to predict tumor deposits in rectal cancer.

KEY POINTS

• Preoperative non-invasive identification of TDs is of great clinical significance. • The combined hybrid-DL model achieved good predictive efficacy and could be used to predict tumor deposits in rectal cancer. • A preoperative nomogram provides gastroenterologist with an accurate and effective tool.

Reproducibility and location-stability of radiomic features derived from cone-beam computed tomography: a phantom study

OBJECTIVES

This study aims to determine the reproducibility and location-stability of cone-beam computed tomography (CBCT) radiomic features.

METHODS

Centrifugal tubes with six concentrations of K2HPO4 solutions (50, 100, 200, 400, 600, and 800 mg ml-1) were imaged within a customized phantom. For each concentration, images were captured twice as test and retest sets. Totally, 69 radiomic features were extracted by LIFEx. The reproducibility was assessed between the test and retest sets. We used the concordance correlation coefficient (CCC) to screen qualified features and then compared the differences in the numbers of them under 24 series (four locations groups * six concentrations). The location-stability was assessed using the Kruskal-Wallis test under different concentration sets; likewise, the numbers of qualified features under six test sets were analyzed.

RESULTS

There were 20 and 23 qualified features in the reproducibility and location-stability experiments, respectively. In the reproducibility experiment, the performance of the peripheral groups and high-concentration sets was significantly better than the center groups and low-concentration sets. The effect of concentration on the location-stability of features was not monotonic, and the number of qualified features in the low-concentration sets was greater than that in the high-concentration sets. No features were qualified in both experiments.

CONCLUSIONS

The density and location of the target object can affect the number of reproducible radiomic features, and its density can also affect the number of location-stable radiomic features. The problem of feature reliability should be treated cautiously in radiomic research on CBCT.

Emerging Role of Nuclear Medicine in Prostate Cancer: Current State and Future Perspectives

Prostate cancer is the most frequent epithelial neoplasia after skin cancer in men starting from 50 years and prostate-specific antigen (PSA) dosage can be used as an early screening tool. Prostate cancer imaging includes several radiological modalities, ranging from ultrasonography, computed tomography (CT), and magnetic resonance to nuclear medicine hybrid techniques such as single-photon emission computed tomography (SPECT)/CT and positron emission tomography (PET)/CT. Innovation in radiopharmaceutical compounds has introduced specific tracers with diagnostic and therapeutic indications, opening the horizons to targeted and very effective clinical care for patients with prostate cancer. The aim of the present review is to illustrate the current knowledge and future perspectives of nuclear medicine, including stand-alone diagnostic techniques and theragnostic approaches, in the clinical management of patients with prostate cancer from initial staging to advanced disease.

Development of a multi-feature-combined model: proof-of-concept with application to local failure prediction of post-SBRT or surgery early-stage NSCLC patients

Objective

To develop a Multi-Feature-Combined (MFC) model for proof-of-concept in predicting local failure (LR) in NSCLC patients after surgery or SBRT using pre-treatment CT images. This MFC model combines handcrafted radiomic features, deep radiomic features, and patient demographic information in an integrated machine learning workflow.

Methods

The MFC model comprised three key steps. (1) Extraction of 92 handcrafted radiomic features from the GTV segmented on pre-treatment CT images. (2) Extraction of 512 deep radiomic features from pre-trained U-Net encoder. (3) The extracted handcrafted radiomic features, deep radiomic features, along with 4 patient demographic information (i.e., gender, age, tumor volume, and Charlson comorbidity index), were concatenated as a multi-dimensional input to the classifiers for LR prediction. Two NSCLC patient cohorts from our institution were investigated: (1) the surgery cohort includes 83 patients with segmentectomy or wedge resection (7 LR), and (2) the SBRT cohort includes 84 patients with lung SBRT (9 LR). The MFC model was developed and evaluated independently for both cohorts, and was subsequently compared against the prediction models based on only handcrafted radiomic features (R models), patient demographic information (PI models), and deep learning modeling (DL models). ROC with AUC was adopted to evaluate model performance with leave-one-out cross-validation (LOOCV) and 100-fold Monte Carlo random validation (MCRV). The t-test was performed to identify the statistically significant differences.

Results

In LOOCV, the AUC range (surgery/SBRT) of the MFC model was 0.858-0.895/0.868-0.913, which was higher than the three other models: 0.356-0.480/0.322-0.650 for PI models, 0.559-0.618/0.639-0.682 for R models, and 0.809/0.843 for DL models. In 100-fold MCRV, the MFC model again showed the highest AUC results (surgery/SBRT): 0.742-0.825/0.888-0.920, which were significantly higher than PI models: 0.464-0.564/0.538-0.628, R models: 0.557-0.652/0.551-0.732, and DL models: 0.702/0.791.

Conclusion

We successfully developed an MFC model that combines feature information from multiple sources for proof-of-concept prediction of LR in patients with surgical and SBRT early-stage NSCLC. Initial results suggested that incorporating pre-treatment patient information from multiple sources improves the ability to predict the risk of local failure.

Optimization and validation of voxel size-related radiomics variability by combatting batch effect harmonization in pulmonary nodules: a phantom and clinical study

Background

Broad generalization of radiomics-assisted models may be impeded by concerns about variability. This study aimed to evaluate the merit of combatting batch effect (ComBat) harmonization in reducing the variability of voxel size-related radiomics in both phantom and clinical study in comparison with image resampling correction method.

Methods

A pulmonary phantom with 22 different types of nodules was scanned by computed tomography (CT) with different voxel sizes. The variability of voxel size-related radiomics features was evaluated using concordance correlation coefficient (CCC), dynamic range (DR), and intraclass correlation coefficient (ICC). ComBat and image resampling compensation methods were used to reduce variability of voxel size-related radiomics. The percentage of robust radiomics features was compared before and after optimization. Pathologically differential diagnosis of invasive adenocarcinoma (IAC) from adenocarcinoma in situ (AIS) and minimally invasive adenocarcinoma (MIA) (AIS-MIA group) was used for clinical validation in 134 patients.

Results

Before optimization, the number of excellent features in the phantom and clinical data was 26.12% and 32.31%, respectively. The excellent features were increased after image resampling and ComBat correction. For clinical optimization, the effect of the ComBat compensation method was significantly better than that of image resampling, with excellent features reaching 90.96% and poor features only amounting to 4.96%. In addition, the hierarchical clustering analysis showed that the first-order and shape features had better robustness than did texture features. In clinical validation, the area under the curve (AUC) of the testing set was 0.865 after ComBat correction.

Conclusions

The ComBat harmonization can optimize voxel size-related CT radiomics variability in pulmonary nodules more efficiently than image resampling harmonization.

Exploring the Utility of Cardiovascular Magnetic Resonance Radiomic Feature Extraction for Evaluation of Cardiac Sarcoidosis

BACKGROUND

The aim of this study is to explore the utility of cardiac magnetic resonance (CMR) imaging of radiomic features to distinguish active and inactive cardiac sarcoidosis (CS).

METHODS

Subjects were classified into active cardiac sarcoidosis (CS_active) and inactive cardiac sarcoidosis (CS_inactive) based on PET-CMR imaging. CS_active was classified as featuring patchy [¹⁸F]fluorodeoxyglucose ([¹⁸F]FDG) uptake on PET and presence of late gadolinium enhancement (LGE) on CMR, while CS_inactive was classified as featuring no [¹⁸F]FDG uptake in the presence of LGE on CMR. Among those screened, thirty CS_active and thirty-one CS_inactive patients met these criteria. A total of 94 radiomic features were subsequently extracted using PyRadiomics. The values of individual features were compared between CS_active and CS_inactive using the Mann-Whitney U test. Subsequently, machine learning (ML) approaches were tested. ML was applied to two sub-sets of radiomic features (signatures A and B) that were selected by logistic regression and PCA, respectively.

RESULTS

Univariate analysis of individual features showed no significant differences. Of all features, gray level co-occurrence matrix (GLCM) joint entropy had a good area under the curve (AUC) and accuracy with the smallest confidence interval, suggesting it may be a good target for further investigation. Some ML classifiers achieved reasonable discrimination between CS_active and CS_inactive patients. With signature A, support vector machine and k-neighbors showed good performance with AUC (0.77 and 0.73) and accuracy (0.67 and 0.72), respectively. With signature B, decision tree demonstrated AUC and accuracy around 0.7; Conclusion: CMR radiomic analysis in CS provides promising results to distinguish patients with active and inactive disease.

Artificial intelligence-driven radiomics study in cancer: the role of feature engineering and modeling

Modern medicine is reliant on various medical imaging technologies for non-invasively observing patients' anatomy. However, the interpretation of medical images can be highly subjective and dependent on the expertise of clinicians. Moreover, some potentially useful quantitative information in medical images, especially that which is not visible to the naked eye, is often ignored during clinical practice. In contrast, radiomics performs high-throughput feature extraction from medical images, which enables quantitative analysis of medical images and prediction of various clinical endpoints. Studies have reported that radiomics exhibits promising performance in diagnosis and predicting treatment responses and prognosis, demonstrating its potential to be a non-invasive auxiliary tool for personalized medicine. However, radiomics remains in a developmental phase as numerous technical challenges have yet to be solved, especially in feature engineering and statistical modeling. In this review, we introduce the current utility of radiomics by summarizing research on its application in the diagnosis, prognosis, and prediction of treatment responses in patients with cancer. We focus on machine learning approaches, for feature extraction and selection during feature engineering and for imbalanced datasets and multi-modality fusion during statistical modeling. Furthermore, we introduce the stability, reproducibility, and interpretability of features, and the generalizability and interpretability of models. Finally, we offer possible solutions to current challenges in radiomics research.

Robustness of radiomics features of virtual unenhanced and virtual monoenergetic images in dual-energy CT among different imaging platforms and potential role of CT number variability

OBJECTIVES

To evaluate robustness of dual-energy CT (DECT) radiomics features of virtual unenhanced (VUE) image and virtual monoenergetic image (VMI) among different imaging platforms.

METHODS

A phantom with sixteen clinical-relevant densities was scanned on ten DECT platforms with comparable scan parameters. Ninety-four radiomic features were extracted via Pyradiomics from VUE images and VMIs at energy level of 70 keV (VMI_70keV). Test-retest repeatability was assessed by Bland-Altman analysis. Inter-platform reproducibility of VUE images and VMI_70keV was evaluated by coefficient of variation (CV) and quartile coefficient of dispersion (QCD) among platforms, and by intraclass correlation coefficient (ICC) and concordance correlation coefficient (CCC) between platform pairs. The correlation between variability of CT number radiomics reproducibility was estimated.

RESULTS

92.02% and 92.87% of features were repeatable between scan-rescans for VUE images and VMI_70keV, respectively. Among platforms, 11.30% and 28.39% features of VUE images, and 15.16% and 28.99% features of VMI_70keV were with CV < 10% and QCD < 10%. The average percentages of radiomics features with ICC > 0.90 and CCC > 0.90 between platform pairs were 10.00% and 9.86% in VUE images and 11.23% and 11.23% in VMI_70keV. The CT number inter-platform reproducibility using CV and QCD showed negative correlations with percentage of the first-order radiomics features with CV < 10% and QCD < 10%, in both VUE images and VMI_70keV (r² 0.3870-0.6178, all p < 0.001).

CONCLUSIONS

The majority of DECT radiomics features were non-reproducible. The differences in CT number were considered as an indicator of inter-platform DECT radiomics variation. Critical relevance statement: The majority of radiomics features extracted from the VUE images and the VMI70keV were non-reproducible among platforms, while synchronizing energy levels of VMI to reduce the CT number value variability may be a potential way to mitigate radiomics instability.

Repeatability of radiomics studies in colorectal cancer: a systematic review

BACKGROUND

Recently, radiomics has been widely used in colorectal cancer, but many variable factors affect the repeatability of radiomics research. This review aims to analyze the repeatability of radiomics studies in colorectal cancer and to evaluate the current status of radiomics in the field of colorectal cancer.

METHODS

The included studies in this review by searching from the PubMed and Embase databases. Then each study in our review was evaluated using the Radiomics Quality Score (RQS). We analyzed the factors that may affect the repeatability in the radiomics workflow and discussed the repeatability of the included studies.

RESULTS

A total of 188 studies was included in this review, of which only two (2/188, 1.06%) studies controlled the influence of individual factors. In addition, the median score of RQS was 11 (out of 36), range-1 to 27.

CONCLUSIONS

The RQS score was moderately low, and most studies did not consider the repeatability of radiomics features, especially in terms of Intra-individual, scanners, and scanning parameters. To improve the generalization of the radiomics model, it is necessary to further control the variable factors of repeatability.

Feasibility and intra-and interobserver reproducibility of quantitative susceptibility mapping with radiomic features for intracranial dissecting intramural hematomas and atherosclerotic calcifications

Quantitative susceptibility mapping (QSM) for 61 patients with dissecting intramural hematomas (n = 36) or atherosclerotic calcifications (n = 25) in intracranial vertebral arteries were collected to assess intra- and interobserver reproducibility in a 3.0-T MR system between January 2015 and December 2017. Two independent observers each segmented regions of interest for lesions twice. The reproducibility was evaluated using intra-class correlation coefficients (ICC) and within-subject coefficients of variation (wCV) for means and concordance correlation coefficients (CCC) and ICC for radiomic features (CCC and ICC > 0.85) were used. Mean QSM values were 0.277 ± 0.092 ppm for dissecting intramural hematomas and - 0.208 ± 0.078 ppm for atherosclerotic calcifications. ICCs and wCVs were 0.885-0.969 and 6.5-13.7% in atherosclerotic calcifications and 0.712-0.865 and 12.4-18.7% in dissecting intramural hematomas, respectively. A total of 9 and 19 reproducible radiomic features were observed in dissecting intramural hematomas and atherosclerotic calcifications, respectively. QSM measurements in dissecting intramural hematomas and atherosclerotic calcifications were feasible and reproducible between intra- and interobserver comparisons, and some reproducible radiomic features were demonstrated.

Prediction of pathological response after neoadjuvant chemotherapy using baseline FDG PET heterogeneity features in breast cancer

Complete pathological response to neoadjuvant systemic treatment (NAST) in some subtypes of breast cancer (BC) has been used as a surrogate of long-term outcome. The possibility of predicting BC pathological response to NAST based on the baseline 18F-Fluorodeoxyglucose positron emission tomography (FDG PET), without the need of an interim study, is a focus of recent discussion. This review summarises the characteristics and results of the available studies regarding the potential impact of heterogeneity features of the primary tumour burden on baseline FDG PET in predicting pathological response to NAST in BC patients. Literature search was conducted on PubMed database and relevant data from each selected study were collected. A total of 13 studies were eligible for inclusion, all of them published over the last 5 years. Eight out of 13 analysed studies indicated an association between FDG PET-based tumour uptake heterogeneity features and prediction of response to NAST. When features associated with predicting response to NAST were derived, these varied between studies. Therefore, definitive reproducible findings across series were difficult to establish. This lack of consensus may reflect the heterogeneity and low number of included series. The clinical relevance of this topic justifies further investigation about the predictive role of baseline FDG PET.

Influence of Image Processing on Radiomic Features From Magnetic Resonance Imaging

OBJECTIVE

Before implementing radiomics in routine clinical practice, comprehensive knowledge about the repeatability and reproducibility of radiomic features is required. The aim of this study was to systematically investigate the influence of image processing parameters on radiomic features from magnetic resonance imaging (MRI) in terms of feature values as well as test-retest repeatability.

MATERIALS AND METHODS

Utilizing a phantom consisting of 4 onions, 4 limes, 4 kiwifruits, and 4 apples, we acquired a test-retest dataset featuring 3 of the most commonly used MRI sequences on a 3 T scanner, namely, a T1-weighted, a T2-weighted, and a fluid-attenuated inversion recovery sequence, each at high and low resolution. After semiautomatic image segmentation, image processing with systematic variation of image processing parameters was performed, including spatial resampling, intensity discretization, and intensity rescaling. For each respective image processing setting, a total of 45 radiomic features were extracted, corresponding to the following 7 matrices/feature classes: conventional indices, histogram matrix, shape matrix, gray-level zone length matrix, gray-level run length matrix, neighboring gray-level dependence matrix, and gray-level cooccurrence matrix. Systematic differences of individual features between different resampling steps were assessed using 1-way analysis of variance with Tukey-type post hoc comparisons to adjust for multiple testing. Test-retest repeatability of radiomic features was measured using the concordance correlation coefficient, dynamic range, and intraclass correlation coefficient.

RESULTS

Image processing influenced radiological feature values. Regardless of the acquired sequence and feature class, significant differences ( P < 0.05) in feature values were found when the size of the resampled voxels was too large, that is, bigger than 3 mm. Almost all higher-order features depended strongly on intensity discretization. The effects of intensity rescaling were negligible except for some features derived from T1-weighted sequences. For all sequences, the percentage of repeatable features (concordance correlation coefficient and dynamic range ≥ 0.9) varied considerably depending on the image processing settings. The optimal image processing setting to achieve the highest percentage of stable features varied per sequence. Irrespective of image processing, the fluid-attenuated inversion recovery sequence in high-resolution overall yielded the highest number of stable features in comparison with the other sequences (89% vs 64%-78% for the respective optimal image processing settings). Across all sequences, the most repeatable features were generally obtained for a spatial resampling close to the originally acquired voxel size and an intensity discretization to at least 32 bins.

CONCLUSION

Variation of image processing parameters has a significant impact on the values of radiomic features as well as their repeatability. Furthermore, the optimal image processing parameters differ for each MRI sequence. Therefore, it is recommended that these processing parameters be determined in corresponding test-retest scans before clinical application. Extensive repeatability, reproducibility, and validation studies as well as standardization are required before quantitative image analysis and radiomics can be reliably translated into routine clinical care.

Multiparametric Data-driven Imaging Markers: Guidelines for Development, Application and Reporting of Model Outputs in Radiomics

This paper is the fifth in a five-part series on statistical methodology for performance assessment of multi-parametric quantitative imaging biomarkers (mpQIBs) for radiomic analysis. Radiomics is the process of extracting visually imperceptible features from radiographic medical images using data-driven algorithms. We refer to the radiomic features as data-driven imaging markers (DIMs), which are quantitative measures discovered under a data-driven framework from images beyond visual recognition but evident as patterns of disease processes irrespective of whether or not ground truth exists for the true value of the DIM. This paper aims to set guidelines on how to build machine learning models using DIMs in radiomics and to apply and report them appropriately. We provide a list of recommendations, named RANDAM (an abbreviation of "Radiomic ANalysis and DAta Modeling"), for analysis, modeling, and reporting in a radiomic study to make machine learning analyses in radiomics more reproducible. RANDAM contains five main components to use in reporting radiomics studies: design, data preparation, data analysis and modeling, reporting, and material availability. Real case studies in lung cancer research are presented along with simulation studies to compare different feature selection methods and several validation strategies.

Data synthesis and adversarial networks: A review and meta-analysis in cancer imaging

Despite technological and medical advances, the detection, interpretation, and treatment of cancer based on imaging data continue to pose significant challenges. These include inter-observer variability, class imbalance, dataset shifts, inter- and intra-tumour heterogeneity, malignancy determination, and treatment effect uncertainty. Given the recent advancements in image synthesis, Generative Adversarial Networks (GANs), and adversarial training, we assess the potential of these technologies to address a number of key challenges of cancer imaging. We categorise these challenges into (a) data scarcity and imbalance, (b) data access and privacy, (c) data annotation and segmentation, (d) cancer detection and diagnosis, and (e) tumour profiling, treatment planning and monitoring. Based on our analysis of 164 publications that apply adversarial training techniques in the context of cancer imaging, we highlight multiple underexplored solutions with research potential. We further contribute the Synthesis Study Trustworthiness Test (SynTRUST), a meta-analysis framework for assessing the validation rigour of medical image synthesis studies. SynTRUST is based on 26 concrete measures of thoroughness, reproducibility, usefulness, scalability, and tenability. Based on SynTRUST, we analyse 16 of the most promising cancer imaging challenge solutions and observe a high validation rigour in general, but also several desirable improvements. With this work, we strive to bridge the gap between the needs of the clinical cancer imaging community and the current and prospective research on data synthesis and adversarial networks in the artificial intelligence community.

PET-based radiomics visualizes tumor-infiltrating CD8 T cell exhaustion to optimize radiotherapy/immunotherapy combination in mouse models of lung cancer

BACKGROUND

Cumulative preclinical and clinical evidences showed radiotherapy might augment systemic antitumoral responses to immunotherapy for metastatic non-small cell lung cancer, but the optimal timing of combination is still unclear. The overall infiltration and exhausted subpopulations of tumor-infiltrating CD8⁺ T cells might be a potential biomarker indicating the response to immune checkpoint inhibitors (ICI), the alteration of which is previously uncharacterized during peri-irradiation period, while dynamic monitoring is unavailable via repeated biopsies in clinical practice.

METHODS

Basing on tumor-bearing mice model, we investigated the dynamics of overall infiltration and exhausted subpopulations of CD8⁺ T cells after ablative irradiation. With the understanding of distinct metabolic characteristics accompanied with T cell exhaustion, we developed a PET radiomics approach to identify and visualize T cell exhaustion status.

RESULTS

CD8⁺ T cell infiltration increased from 3 to 14 days after ablative irradiation while terminally exhausted populations significantly predominated CD8⁺ T cells during late course of this infiltrating period, indicating that 3-7 days post-irradiation might be a potential appropriate window for delivering ICI treatment. A PET radiomics approach was established to differentiate T cell exhaustion status, which fitted well in both ICI and irradiation settings. We also visualized the underlying association of more heterogeneous texture on PET images with progressed T cell exhaustion.

CONCLUSIONS

We proposed a non-invasive imaging predictor which accurately assessed heterogeneous T cell exhaustion status relevant to ICI treatment and irradiation, and might serve as a promising solution to timely estimate immune-responsiveness of tumor microenvironment and the optimal timing of combined therapy.

Advanced Techniques in Head and Neck Cancer Imaging: Guide to Precision Cancer Management

Precision treatment requires precision imaging. With the advent of various advanced techniques in head and neck cancer treatment, imaging has become an integral part of the multidisciplinary approach to head and neck cancer care from diagnosis to staging and also plays a vital role in response evaluation in various tumors. Conventional anatomic imaging (CT scan, MRI, ultrasound) remains basic and focuses on defining the anatomical extent of the disease and its spread. Accurate assessment of the biological behavior of tumors, including tumor cellularity, growth, and response evaluation, is evolving with recent advances in molecular, functional, and hybrid/multiplex imaging. Integration of these various advanced diagnostic imaging and nonimaging methods aids understanding of cancer pathophysiology and provides a more comprehensive evaluation in this era of precision treatment. Here we discuss the current status of various advanced imaging techniques and their applications in head and neck cancer imaging.

Radiomic and clinical nomogram for cognitive impairment prediction in Wilson's disease

Objective

To investigate potential biomarkers for the early detection of cognitive impairment in patients with Wilson's disease (WD), we developed a computer-assisted radiomics model to distinguish between WD and WD cognitive impairment.

Methods

Overall, 136 T1-weighted MR images were retrieved from the First Affiliated Hospital of Anhui University of Chinese Medicine, including 77 from patients with WD and 59 from patients with WD cognitive impairment. The images were divided into training and test groups at a ratio of 70:30. The radiomic features of each T1-weighted image were extracted using 3D Slicer software. R software was used to establish clinical and radiomic models based on clinical characteristics and radiomic features, respectively. The receiver operating characteristic profiles of the three models were evaluated to assess their diagnostic accuracy and reliability in distinguishing between WD and WD cognitive impairment. We combined relevant neuropsychological test scores of prospective memory to construct an integrated predictive model and visual nomogram to effectively assess the risk of cognitive decline in patients with WD.

Results

The area under the curve values for distinguishing WD and WD cognitive impairment for the clinical, radiomic, and integrated models were 0.863, 0.922, and 0.935 respectively, indicative of excellent performance. The nomogram based on the integrated model successfully differentiated between WD and WD cognitive impairment.

Conclusion

The nomogram developed in the current study may assist clinicians in the early identification of cognitive impairment in patients with WD. Early intervention following such identification may help improve long-term prognosis and quality of life of these patients.

MRI-Based Computer-Aided Diagnostic Model to Predict Tumor Grading and Clinical Outcomes in Patients With Soft Tissue Sarcoma

BACKGROUND

MRI acts as a potential resource for exploration and interpretation to identify tumor characterization by advanced computer-aided diagnostic (CAD) methods.

PURPOSE

To evaluate and validate the performance of MRI-based CAD models for identifying low-grade and high-grade soft tissue sarcoma (STS) and for investigating survival prognostication.

STUDY TYPE

Retrospective.

SUBJECTS

A total of 540 patients (295 male/female: 295/245, median age: 42 years) with STSs.

FIELD SEQUENCE

5-T MRI with T₁ WI sequence and fat-suppressed T₂ -weighted (T₂ FS) sequence.

ASSESSMENT

Manual regions of interests (ROIs) were delineated for generation of radiomic features. Automatic segmentation and pretrained convolutional neural networks (CNNs) were performed for deep learning (DL) analysis. The last fully connected layer at the top of CNNs was removed, and the global max pooling was added to transform feature maps to numeric values. Tumor grade was determined on histological specimens.

STATISTICAL TESTS

The support vector machine was adopted as the classifier for all MRI-based models. The DL signature was derived from the DL-MRI model with the highest area under the curve (AUC). The significant clinical variables, tumor location and size, integrated with radiomics and DL signatures were ready for construction of clinical-MRI nomogram to identify tumor grading. The prognostic value of clinical variables and these MRI-based signatures for overall survival (OS) was evaluated via Cox proportional hazard.

RESULTS

The clinical-MRI differentiation nomogram represented an AUC of 0.870 in the training cohort, and an AUC of 0.855, accuracy of 79.01%, sensitivity of 79.03%, and specificity of 78.95% in the validation cohort. The prognostic model showed good performance for OS with 3-year C-index of 0.681 and 0.642 and 5-year C-index of 0.722 and 0.676 in the training and validation cohorts.

DATA CONCLUSION

MRI-based CAD nomogram represents effective abilities in classification of low-grade and high-grade STSs. The MRI-based prognostic model yields favorable preoperative capacities to identify long-term survivals for STSs.

EVIDENCE LEVEL

3 TECHNICAL EFFICACY: Stage 4.

Quantification of lung function on CT images based on pulmonary radiomic filtering

PURPOSE

To develop a radiomics filtering technique for characterizing spatial-encoded regional pulmonary ventilation information on lung computed tomography (CT).

METHODS

The lung volume was segmented on 46 CT images, and a 3D sliding window kernel was implemented across the lung volume to capture the spatial-encoded image information. Fifty-three radiomic features were extracted within the kernel, resulting in a fourth-order tensor object. As such, each voxel coordinate of the original lung was represented as a 53-dimensional feature vector, such that radiomic features could be viewed as feature maps within the lungs. To test the technique as a potential pulmonary ventilation biomarker, the radiomic feature maps were compared to paired functional images (Galligas PET or DTPA-SPECT) based on the Spearman correlation (ρ) analysis.

RESULTS

The radiomic feature maps GLRLM-based Run-Length Non-Uniformity and GLCOM-based Sum Average are found to be highly correlated with the functional imaging. The achieved ρ (median [range]) for the two features are 0.46 [0.05, 0.67] and 0.45 [0.21, 0.65] across 46 patients and 2 functional imaging modalities, respectively.

CONCLUSIONS

The results provide evidence that local regions of sparsely encoded heterogeneous lung parenchyma on CT are associated with diminished radiotracer uptake and measured lung ventilation defects on PET/SPECT imaging. These findings demonstrate the potential of radiomics to serve as a complementary tool to the current lung quantification techniques and provide hypothesis-generating data for future studies.

Investigation of the Inter- and Intrascanner Reproducibility and Repeatability of Radiomics Features in T1-Weighted Brain MRI

BACKGROUND

Radiomics is the high throughput analysis of medical images using computer algorithms, which specifically assess textural features. It has increasingly been proposed as a tool for the development of imaging biomarkers. However, an important acknowledged limitation of radiomics is the lack of reproducibility of features produced.

PURPOSE

To assess reproducibility and repeatability of radiomics variables in brain MRI through a multivisit, multicenter study.

STUDY TYPE

Retrospective.

POPULATION

Fourteen individuals visiting three institutions twice, 10 males with the mean age of 36.3 years and age range 25-51.

FIELD STRENGTH

3D T1W inversion recovery on three 1.5-T General Electric scanners.

ASSESSMENT

Radiomics analysis by a consultant radiologist performed on the T1W images of the whole brain on all visits. All possible radiomics features were generated.

STATISTICAL TEST

Concordance correlation coefficient (CCC) and dynamic range (DR) for all variables were calculated to assess the test-retest repeatability. Intraclass correlation coefficients (ICCs) were calculated to investigate the reproducibility of features across centers.

RESULTS

Of 1596 features generated, 57 from center 1, 15 from center 2, and 22 from center 3 had a CCC > 0.9 and DR > 0.9. Eight variables had CCC > 0.9 and DR > 0.9 in all centers. Forty-one variables had an ICC of >0.9. No variables had CCC > 0.9, DR > 0.9, and ICC > 0.9.

DATA CONCLUSION

Repeatability and reproducibility of variables is a significant limitation of radiomics analysis in 3DT1W brain MRI. Careful selection of radiomic features is required.

LEVEL OF EVIDENCE

4 TECHNICAL EFFICACY STAGE: 2.

Combining multiparametric MRI features-based transfer learning and clinical parameters: application of machine learning for the differentiation of uterine sarcomas from atypical leiomyomas

OBJECTIVES

To explore the feasibility and effectiveness of machine learning (ML) based on multiparametric magnetic resonance imaging (mp-MRI) features extracted from transfer learning combined with clinical parameters to differentiate uterine sarcomas from atypical leiomyomas (ALMs).

METHODS

The data of 86 uterine sarcomas between July 2011 and December 2019 and 86 ALMs between June 2013 and June 2017 were retrospectively reviewed. We extracted deep-learning features and radiomics features from T2-weighted imaging (T2WI) and diffusion weighted imaging (DWI). The two feature extraction methods, transfer learning and radiomics, were compared. Random forest was adopted as the classifier. T2WI features, DWI features, combined T2WI and DWI (mp-MRI) features, and combined clinical parameters and mp-MRI features were applied to establish T2, DWI, T2-DWI, and complex multiparameter (mp) models, respectively. Predictive performance was assessed with the area under the receiver operating characteristic curve (AUC).

RESULTS

In the test set, the T2, DWI, T2-DWI and complex mp models based on transfer learning (AUCs range from 0.76 to 0.81, 0.80 to 0.88, 0.85 to 0.92, and 0.94 to 0.96, respectively) outperformed the models based on radiomics (AUCs of 0.73, 0.76, 0.79, and 0.92, respectively). Moreover, the complex mp model showed the best prediction performance, with the Resnet50-complex mp model achieving the highest AUC (0.96) and accuracy (0.87).

CONCLUSIONS

Transfer learning is feasible and superior to radiomics in the differential diagnosis of uterine sarcomas and ALMs in our dataset. ML models based on deep learning features of nonenhanced mp-MRI and clinical parameters can achieve good diagnostic efficacy.

KEY POINTS

• The ML model combining nonenhanced mp-MRI features and clinical parameters can distinguish uterine sarcomas from ALMs. • Transfer learning can be applied to differentiate uterine sarcomas from ALMs and outperform radiomics. • The most accurate prediction model was Resnet50-based transfer learning, built with the deep-learning features of mp-MRI and clinical parameters.

Robustness of radiomics to variations in segmentation methods in multimodal brain MRI

Radiomics in neuroimaging uses fully automatic segmentation to delineate the anatomical areas for which radiomic features are computed. However, differences among these segmentation methods affect radiomic features to an unknown extent. A scan-rescan dataset (n = 46) of T1-weighted and diffusion tensor images was used. Subjects were split into a sleep-deprivation and a control group. Scans were segmented using four segmentation methods from which radiomic features were computed. First, we measured segmentation agreement using the Dice-coefficient. Second, robustness and reproducibility of radiomic features were measured using the intraclass correlation coefficient (ICC). Last, difference in predictive power was assessed using the Friedman-test on performance in a radiomics-based sleep deprivation classification application. Segmentation agreement was generally high (interquartile range = 0.77–0.90) and median feature robustness to segmentation method variation was higher (ICC > 0.7) than scan-rescan reproducibility (ICC 0.3–0.8). However, classification performance differed significantly among segmentation methods (p < 0.001) ranging from 77 to 84%. Accuracy was higher for more recent deep learning-based segmentation methods. Despite high agreement among segmentation methods, subtle differences significantly affected radiomic features and their predictive power. Consequently, the effect of differences in segmentation methods should be taken into account when designing and evaluating radiomics-based research methods.

Computed Tomography-Based Deep Learning Nomogram Can Accurately Predict Lymph Node Metastasis in Gastric Cancer

BACKGROUND

Computed tomography is the most commonly used imaging modality for preoperative assessment of lymph node status, but the reported accuracy is unsatisfactory.

AIMS

To evaluate and verify the predictive performance of computed tomography deep learning on the presurgical evaluation of lymph node metastasis in patients with gastric cancer.

METHODS

347 patients were retrospectively selected (training cohort: 242, test cohort: 105). The enhanced computed tomography arterial phase images of gastric cancer were used for lesion segmentation, radiomics and deep learning feature extraction. Three methods were used for feature selection. Support vector machine (SVM) or random forest (RF) was used to build models. The classification performance of the models was evaluated using the area under the receiver operating characteristic curve (AUC). We also established a nomogram that included clinical predictors.

RESULTS

The model based on ResNet50-RF showed favorable classification performance and was verified in the test cohort (AUC = 0.9803). The nomogram based on deep learning feature scores and the lymph node status reported by computed tomography showed excellent discrimination. AUC of 0.9978 was achieved in the training cohort and verified in the test cohort (AUC = 0.9914). Decision analysis curve showed the value of nomogram in clinical application.

CONCLUSION

The computed tomography-based deep learning nomogram can accurately and effectively evaluate lymph node metastasis in patients with gastric cancer before surgery.

Evaluation of Epidermal Growth Factor Receptor 2 Status in Gastric Cancer by CT-Based Deep Learning Radiomics Nomogram

Purpose

To explore the role of computed tomography (CT)-based deep learning and radiomics in preoperative evaluation of epidermal growth factor receptor 2 (HER2) status in gastric cancer.

Materials and methods

The clinical data on gastric cancer patients were evaluated retrospectively, and 357 patients were chosen for this study (training cohort: 249; test cohort: 108). The preprocessed enhanced CT arterial phase images were selected for lesion segmentation, radiomics and deep learning feature extraction. We integrated deep learning features and radiomic features (Inte). Four methods were used for feature selection. We constructed models with support vector machine (SVM) or random forest (RF), respectively. The area under the receiver operating characteristics curve (AUC) was used to assess the performance of these models. We also constructed a nomogram including Inte-feature scores and clinical factors.

Results

The radiomics-SVM model showed good classification performance (AUC, training cohort: 0.8069; test cohort: 0.7869). The AUC of the ResNet50-SVM model and the Inte-SVM model in the test cohort were 0.8955 and 0.9055. The nomogram also showed excellent discrimination achieving greater AUC (training cohort, 0.9207; test cohort, 0.9224).

Conclusion

CT-based deep learning radiomics nomogram can accurately and effectively assess the HER2 status in patients with gastric cancer before surgery and it is expected to assist physicians in clinical decision-making and facilitates individualized treatment planning.

Robustness of radiomic features in magnetic resonance imaging for patients with glioblastoma: multi-center study

Background and purpose

Radiomics offers great potential in improving diagnosis and treatment for patients with glioblastoma multiforme. However, in order to implement radiomics in clinical routine, the features used for prognostic modelling need to be stable. This comprises significant challenge in multi-center studies. The aim of this study was to evaluate the impact of different image normalization methods on MRI features robustness in multi-center study.

Methods

Radiomics stability was checked on magnetic resonance images of eleven patients. The images were acquired in two different hospitals using contrast-enhanced T1 sequences. The images were normalized using one of five investigated approaches including grey-level discretization, histogram matching and z-score. Then, radiomic features were extracted and features stability was evaluated using intra-class correlation coefficients. In the second part of the study, improvement in the prognostic performance of features was tested on 60 patients derived from publicly available dataset.

Results

Depending on the normalization scheme, the percentage of stable features varied from 3.4% to 8%. The histogram matching based on the tumor region showed the highest amount of the stable features (113/1404); while normalization using fixed bin size resulted in 48 stable features. The histogram matching also led to better prognostic value (median c-index increase of 0.065) comparing to non-normalized images.

Conclusions

MRI normalization plays an important role in radiomics. Appropriate normalization helps to select robust features, which can be used for prognostic modelling in multicenter studies. In our study, histogram matching based on tumor region improved both stability of radiomic features and their prognostic value.

Novel computer aided diagnostic models on multimodality medical images to differentiate well differentiated liposarcomas from lipomas approached by deep learning methods

BACKGROUND

Deep learning methods have great potential to predict tumor characterization, such as histological diagnosis and genetic aberration. The objective of this study was to evaluate and validate the predictive performance of multimodality imaging-derived models using computer-aided diagnostic (CAD) methods for prediction of MDM2 gene amplification to identify well-differentiated liposarcoma (WDLPS) and lipoma.

MATERIALS AND METHODS

All 127 patients from two institutions were included with 89 patients in one institution for model training and 38 patients in the other institution for external validation between January 2012 and December 2018. For each modality, handcrafted radiomics analysis with manual segmentation was applied to extract 851 features for each modality, and six pretrained convolutional neural networks (CNNs) extracted 512-2048 deep learning features automatically. Extracted imaging-based features were selected via univariate filter selection methods and the recursive feature elimination algorithm, which were then classified by support vector machine for model construction. Integrated with two significant clinical variables, age and LDH level, a clinical-radiological model was constructed for identification WDLPS and lipoma. All differentiation models were evaluated using the area under the receiver operating characteristics curve (AUC) and their 95% confidence interval (CI).

RESULTS

The multimodality model on deep learning features extracted from ResNet50 algorithm (RN-DL model) performed great differentiation performance with an AUC of 0.995 (95% CI 0.987-1.000) for the training cohort, and an AUC of 0.950 (95% CI 0.886-1.000), accuracy of 92.11%, sensitivity of 95.00% (95% CI 73.06-99.74%), specificity of 88.89% (95% CI 63.93-98.05%) in external validation. The integrated clinical-radiological model represented an AUC of 0.996 (95% CI 0.989-1.000) for the training cohort, and an AUC of 0.942 (95% CI 0.867-1.000), accuracy of 86.84%, sensitivity of 95.00% (95% CI 73.06-99.74%), and specificity of 77.78% (95% CI 51.92-92.63%) in external validation.

CONCLUSIONS

Imaging-based multimodality models represent effective discrimination abilities between WDLPS and lipoma via CAD methods, and might be a practicable approach in assistance of treatment decision.

Repeatability and Reproducibility of Computed Tomography Radiomics for Pulmonary Nodules: A Multicenter Phantom Study

BACKGROUND

Radiomics can yield minable information from medical images, which can facilitate computer-aided diagnosis. However, the lack of repeatability and reproducibility of radiomic features (RFs) may hinder their generalizability in clinical applications.

OBJECTIVES

The aims of this study were to explore 3 main sources of variability in RFs, investigate their influencing magnitudes and patterns, and identify a subset of robust RFs for further studies.

MATERIALS AND METHODS

A chest phantom with nodules was scanned with different computed tomography (CT) scanners repeatedly with varying acquisition and reconstruction parameters (April-May 2019) to evaluate 3 sources of variability: test-retest, inter-CT, and intra-CT protocol variability. The robustness of the RFs was measured using the concordance correlation coefficient, dynamic range, and intraclass correlation coefficient (ICC). The influencing magnitudes and patterns were analyzed using the Friedman test and Spearman rank correlation coefficient. Stable and informative RFs were selected, and their redundancy was eliminated using hierarchical clustering. Clinical validation was also performed to verify the clinical effectiveness and potential enhancement of the generalizability of radiomics research.

RESULTS

A total of 1295 RFs that showed all 3 sources of variability were included. The reconstruction kernel and the iteration level showed the greatest (ICC, 0.35 ± 0.31) and the least (ICC, 0.63 ± 0.27) influence on magnitudes. The different sources of variability showed relatively consistent patterns of influence (false discovery rate <0.001). Finally, we obtained a subset of 19 stable, informative, and nonredundant RFs under all 3 sources of variability. These RFs exhibited clinical effectiveness and showed better prediction performance than unstable RFs in the validation dataset (P = 0.017, Delong test).

CONCLUSIONS

The stability of RFs was affected to different degrees by test-retest and differences in CT manufacturers and models and CT acquisition and reconstruction parameters, but the influences of these factors showed relatively consistent patterns. We also obtained a subset of 19 stable, informative, and nonredundant RFs that should be preferably used to enhance the generalizability of further radiomics research.

Transient Enlargement in Meningiomas Treated with Stereotactic Radiotherapy

Simple Summary

Accurate assessment of treatment efficacy is a prerequisite for the improvement in therapeutic outcomes in clinical trials. However, it is very challenging to accurately track the size of meningiomas after radiotherapy, because of their complex shapes and often slow growth. Measuring the whole tumor volume as opposed to simple diameter measurements to assess treatment efficacy, therefore, is very promising but little is known on expected volumetric changes of meningiomas following radiotherapy. Therefore, in this study, we meticulously investigated volumetric changes in meningiomas following radiotherapy incorporating volumetric measurements from 468 MRI studies and evaluated newly proposed RANO volumetric criteria in the context of radiotherapy. We found that temporary tumor enlargement after radiotherapy overall was rare but occurred significantly more frequently after high than after low single doses of radiation, represented an important differential diagnosis to tumor progression and would have skewed results in a clinical trial if not accounted for.

Abstract

To investigate the occurrence of pseudoprogression/transient enlargement in meningiomas after stereotactic radiotherapy (RT) and to evaluate recently proposed volumetric RANO meningioma criteria for response assessment in the context of RT. Sixty-nine meningiomas (benign: 90%, atypical: 10%) received stereotactic RT from January 2005–May 2018. A total of 468 MRI studies were segmented longitudinally during a median follow-up of 42.3 months. Best response and local control were evaluated according to recently proposed volumetric RANO criteria. Transient enlargement was defined as volumetric increase ≥20% followed by a subsequent regression ≥20%. The mean best volumetric response was −23% change from baseline (range, −86% to +19%). According to RANO, the best volumetric response was SD in 81% (56/69), MR in 13% (9/69) and PR in 6% (4/69). Transient enlargement occurred in only 6% (4/69) post RT but would have represented 60% (3/5) of cases with progressive disease if not accounted for. Transient enlargement was characterized by a mean maximum volumetric increase of +181% (range, +24% to +389 %) with all cases occurring in the first year post-RT (range, 4.1–10.3 months). Transient enlargement was significantly more frequent with SRS or hypofractionation than with conventional fractionation (25% vs. 2%, p = 0.015). Five-year volumetric control was 97.8% if transient enlargement was recognized but 92.9% if not accounted for. Transient enlargement/pseudoprogression in the first year following SRS and hypofractionated RT represents an important differential diagnosis, especially because of the high volumetric control achieved with stereotactic RT. Meningioma enlargement during subsequent post-RT follow-up and after conventional fractionation should raise suspicion for tumor progression.

Convolutional Neural Network Addresses the Confounding Impact of CT Reconstruction Kernels on Radiomics Studies

Achieving high feature reproducibility while preserving biological information is one of the main challenges for the generalizability of current radiomics studies. Non-clinical imaging variables, such as reconstruction kernels, have shown to significantly impact radiomics features. In this study, we retrain an open-source convolutional neural network (CNN) to harmonize computerized tomography (CT) images with various reconstruction kernels to improve feature reproducibility and radiomic model performance using epidermal growth factor receptor (EGFR) mutation prediction in lung cancer as a paradigm. In the training phase, the CNN was retrained and tested on 32 lung cancer patients’ CT images between two different groups of reconstruction kernels (smooth and sharp). In the validation phase, the retrained CNN was validated on an external cohort of 223 lung cancer patients’ CT images acquired using different CT scanners and kernels. The results showed that the retrained CNN could be successfully applied to external datasets with different CT scanner parameters, and harmonization of reconstruction kernels from sharp to smooth could significantly improve the performance of radiomics model in predicting EGFR mutation status in lung cancer. In conclusion, the CNN based method showed great potential in improving feature reproducibility and generalizability by harmonizing medical images with heterogeneous reconstruction kernels.

Impact of Interobserver Variability in Manual Segmentation of Non-Small Cell Lung Cancer (NSCLC) Applying Low-Rank Radiomic Representation on Computed Tomography

Simple Summary

Discovery of predictive and prognostic radiomic features in cancer is currently of great interest to the radiologic and oncologic community. Tumor phenotypic and prognostic information can be obtained by extracting features on tumor segmentations, and it is typically imaging analysts, physician trainees, and attending physicians who provide these labeled datasets for analysis. The potential impact of level and type of specialty training on interobserver variability in manual segmentation of NSCLC was examined. Although there was some variability in segmentation between readers, the subsequently extracted radiomic features were overall well correlated. High fidelity radiomic feature extraction relies on accurate feature extraction from imaging that produce robust prognostic and predictive radiomic NSCLC biomarkers. This study concludes that this goal can be obtained using segmenters of different levels of training and clinical experience.

Abstract

This study tackles interobserver variability with respect to specialty training in manual segmentation of non-small cell lung cancer (NSCLC). Four readers included for segmentation are: a data scientist (BY), a medical student (LS), a radiology trainee (MH), and a specialty-trained radiologist (SK) for a total of 293 patients from two publicly available databases. Sørensen–Dice (SD) coefficients and low rank Pearson correlation coefficients (CC) of 429 radiomics were calculated to assess interobserver variability. Cox proportional hazard (CPH) models and Kaplan-Meier (KM) curves of overall survival (OS) prediction for each dataset were also generated. SD and CC for segmentations demonstrated high similarities, yielding, SD: 0.79 and CC: 0.92 (BY-SK), SD: 0.81 and CC: 0.83 (LS-SK), and SD: 0.84 and CC: 0.91 (MH-SK) in average for both databases, respectively. OS through the maximal CPH model for the two datasets yielded c-statistics of 0.7 (95% CI) and 0.69 (95% CI), while adding radiomic and clinical variables (sex, stage/morphological status, and histology) together. KM curves also showed significant discrimination between high- and low-risk patients (p-value < 0.005). This supports that readers’ level of training and clinical experience may not significantly influence the ability to extract accurate radiomic features for NSCLC on CT. This potentially allows flexibility in the training required to produce robust prognostic imaging biomarkers for potential clinical translation.

Computed tomography-based deep-learning prediction of induction chemotherapy treatment response in locally advanced nasopharyngeal carcinoma

BACKGROUND

Deep learning methods have great potential to predict treatment response. The objective of this study was to evaluate and validate the predictive performance of the computed tomography (CT)-based model using deep learning features for identification of responders and nonresponders to induction chemotherapy (IC) in nasopharyngeal carcinoma (NPC).

MATERIALS AND METHODS

All eligible patients were included retrospectively between January 2012 and December 2018, and assigned to the training (n = 208) or the testing cohort (n = 89). We extracted deep learning features of six pretrained convolutional neural networks (CNNs) via transfer learning method, and handcrafted radiomics features manually. Support vector machine (SVM) was adopted as the classifier. All predictive models were evaluated using the area under the receiver operating characteristics curve (AUC), by which an optimal model was selected. We also built clinical and clinical-radiological models for comparison.

RESULTS

The model with features extracted from ResNet50 (RN-SVM) had optimal performance among all models with features extracted from pretrained CNNs with an AUC of 0.811, accuracy of 68.54%, sensitivity of 61.54%, specificity of 87.50%, positive predictive value (PPV) of 93.02%, and negative predictive value (NPV) of 45.65% in the testing cohort. The handcrafted radiomics model was slightly inferior to the RN-SVM model with an AUC of 0.663 and accuracy of 60.67% in the testing cohort. All the imaging-derived models had better predictive performance than the clinical model.

CONCLUSION

The noninvasive deep learning method could provide efficient prediction of treatment response to IC in locally advanced NPC and might be a practicable approach in therapeutic strategy decision-making.

Incorporating Robustness to Imaging Physics into Radiomic Feature Selection for Breast Cancer Risk Estimation

Digital mammography has seen an explosion in the number of radiomic features used for risk-assessment modeling. However, having more features is not necessarily beneficial, as some features may be overly sensitive to imaging physics (contrast, noise, and image sharpness). To measure the effects of imaging physics, we analyzed the feature variation across imaging acquisition settings (kV, mAs) using an anthropomorphic phantom. We also analyzed the intra-woman variation (IWV), a measure of how much a feature varies between breasts with similar parenchymal patterns-a woman's left and right breasts. From 341 features, we identified "robust" features that minimized the effects of imaging physics and IWV. We also investigated whether robust features offered better case-control classification in an independent data set of 575 images, all with an overall BI-RADS® assessment of 1 (negative) or 2 (benign); 115 images (cases) were of women who developed cancer at least one year after that screening image, matched to 460 controls. We modeled cancer occurrence via logistic regression, using cross-validated area under the receiver-operating-characteristic curve (AUC) to measure model performance. Models using features from the most-robust quartile of features yielded an AUC = 0.59, versus 0.54 for the least-robust, with p < 0.005 for the difference among the quartiles.

Quantification of Minimum Detectable Difference in Radiomics Features Across Lesions and CT Imaging Conditions

RATIONALE AND OBJECTIVES

The 3-fold purpose of this study was to (1) develop a method to relate measured differences in radiomics features in different computed tomography (CT) scans to one another and to true feature differences; (2) quantify minimum detectable change in radiomics features based on measured radiomics features from pairs of synthesized CT images acquired under variable CT scan settings, and (3) ascertain and inform the recommendations of the Quantitative Imaging Biomarkers Alliance (QIBA) for nodule volumetry.

MATERIALS AND METHODS

Images of anthropomorphic lung nodule models were simulated using resolution and noise properties for 297 unique imaging conditions. Nineteen morphology features were calculated from both the segmentation masks derived from the imaged nodules and from ground truth nodules. Analysis was performed to calculate minimum detectable difference of radiomics features as a function of imaging protocols in comparison to QIBA guidelines.

RESULTS

The minimum detectable differences ranged from 1% to 175% depending on the specific feature and set of imaging protocols. The results showed that QIBA protocol recommendations result in improved minimum detectable difference as compared to the range of possible protocols. The results showed that the minimum detectable differences may be improved from QIBA's current recommendation by further restricting the slice thickness requirement to be between 0.5 mm and 1 mm.

CONCLUSION

Minimum detectable differences of radiomics features were quantified for lung nodules across a wide range of possible protocols. The results can be used prospectively to inform decision-making about imaging protocols to provide superior quantification of radiomics features.

Automated Machine-Learning Framework Integrating Histopathological and Radiological Information for Predicting IDH1 Mutation Status in Glioma

Diffuse gliomas are the most common malignant primary brain tumors. Identification of isocitrate dehydrogenase 1 (IDH1) mutations aids the diagnostic classification of these tumors and the prediction of their clinical outcomes. While histology continues to play a key role in frozen section diagnosis, as a diagnostic reference and as a method for monitoring disease progression, recent research has demonstrated the ability of multi-parametric magnetic resonance imaging (MRI) sequences for predicting IDH genotypes. In this paper, we aim to improve the prediction accuracy of IDH1 genotypes by integrating multi-modal imaging information from digitized histopathological data derived from routine histological slide scans and the MRI sequences including T1-contrast (T1) and Fluid-attenuated inversion recovery imaging (T2-FLAIR). In this research, we have established an automated framework to process, analyze and integrate the histopathological and radiological information from high-resolution pathology slides and multi-sequence MRI scans. Our machine-learning framework comprehensively computed multi-level information including molecular level, cellular level, and texture level information to reflect predictive IDH genotypes. Firstly, an automated pre-processing was developed to select the regions of interest (ROIs) from pathology slides. Secondly, to interactively fuse the multimodal complementary information, comprehensive feature information was extracted from the pathology ROIs and segmented tumor regions (enhanced tumor, edema and non-enhanced tumor) from MRI sequences. Thirdly, a Random Forest (RF)-based algorithm was employed to identify and quantitatively characterize histopathological and radiological imaging origins, respectively. Finally, we integrated multi-modal imaging features with a machine-learning algorithm and tested the performance of the framework for IDH1 genotyping, we also provided visual and statistical explanation to support the understanding on prediction outcomes. The training and testing experiments on 217 pathologically verified IDH1 genotyped glioma cases from multi-resource validated that our fully automated machine-learning model predicted IDH1 genotypes with greater accuracy and reliability than models that were based on radiological imaging data only. The accuracy of IDH1 genotype prediction was 0.90 compared to 0.82 for radiomic result. Thus, the integration of multi-parametric imaging features for automated analysis of cross-modal biomedical data improved the prediction accuracy of glioma IDH1 genotypes.

Robust imaging habitat computation using voxel-wise radiomics features

Tumor heterogeneity has been postulated as a hallmark of treatment resistance and a cure constraint in cancer patients. Conventional quantitative medical imaging (radiomics) can be extended to computing voxel-wise features and aggregating tumor subregions with similar radiological phenotypes (imaging habitats) to elucidate the distribution of tumor heterogeneity within and among tumors. Despite the promising applications of imaging habitats, they may be affected by variability of radiomics features, preventing comparison and generalization of imaging habitats techniques. We performed a comprehensive repeatability and reproducibility analysis of voxel-wise radiomics features in more than 500 lung cancer patients with computed tomography (CT) images and demonstrated the effect of voxel-wise radiomics variability on imaging habitats computation in 30 lung cancer patients with test–retest images. Repeatable voxel-wise features characterized texture heterogeneity and were reproducible regardless of the applied feature extraction parameters. Imaging habitats computed using robust radiomics features were more stable than those computed using all features in test–retest CTs from the same patient. Nine voxel-wise radiomics features (joint energy, joint entropy, sum entropy, maximum probability, difference entropy, Imc1, Imc2, Idn and Idmn) were repeatable and reproducible. This supports their application for computing imaging habitats in lung tumors towards the discovery of previously unseen tumor heterogeneity and the development of novel non-invasive imaging biomarkers for precision medicine.

Variation in textural parameters of hepatic lesions during contrast medium injection

BACKGROUND

Textural parameters extracted using quantitative imaging techniques have been shown to have prognostic value for hepatocellular carcinoma (HCC).

PURPOSE

To evaluate whether the contrast medium timing of the image acquisition affects the reproducibility of textural parameters in HCC and hepatic tissue.

MATERIAL AND METHODS

This retrospective study included 17 patients with 37 HCC lesions. Perfusion computed tomography (CT) was obtained after 50 mL contrast medium injection. HCC lesions were segmented for analysis. The gray-level co-occurrence (GLCM) textural analysis parameters, homogeneity, energy, entropy, inertia, and correlation were calculated. Variation was quantified by calculating the SD of each parameter during respective perfusion series and the inter lesion variation as the SD among the lesions.

RESULTS

The average change in texture parameters in both HCC and hepatic tissue per second after injection was 0.01% to 0.3% of the respective texture parameter. In HCC, the average variation in homogeneity, energy, and entropy within each lesion after contrast medium injection was significantly less than the variation observed among the lesions (23% to 74%, P < 0.001). Significant differences in energy, entropy, inertia, and correlation between hepatic tissue and HCC were observed. However, when considering the intra-individual variation of hepatic tissue over time, only the HCC parameter energy was significantly outside that 95% confidence interval (P < 0.02).

CONCLUSION

The contrast medium timing does not affect the reproducibility of textural parameters in HCC and hepatic tissue. Thus, contrast medium timing should not be an issue at CT texture analysis of HCC.

A Radiomics Model for Predicting Early Recurrence in Grade II Gliomas Based on Preoperative Multiparametric Magnetic Resonance Imaging

Objective

This study aimed to develop a radiomics model to predict early recurrence (<1 year) in grade II glioma after the first resection.

Methods

The pathological, clinical, and magnetic resonance imaging (MRI) data of patients diagnosed with grade II glioma who underwent surgery and had a recurrence between 2017 and 2020 in our hospital were retrospectively analyzed. After a rigorous selection, 64 patients were eligible and enrolled in the study. Twenty-two cases had a pathologically confirmed recurrent glioma. The cases were randomly assigned using a ratio of 7:3 to either the training set or validation set. T1-weighted image (T1WI), T2-weighted image (T2WI), and contrast-enhanced T1-weighted image (T1CE) were acquired. The minimum-redundancy-maximum-relevancy (mRMR) method alone or in combination with univariate logistic analysis were used to identify the most optimal predictive feature from the three image sequences. Multivariate logistic regression analysis was then used to develop a predictive model using the screened features. The performance of each model in both training and validation datasets was assessed using a receiver operating characteristic (ROC) curve, calibration curve, and decision curve analysis (DCA).

Results

A total of 396 radiomics features were initially extracted from each image sequence. After running the mRMR and univariate logistic analysis, nine predictive features were identified and used to build the multiparametric radiomics model. The model had a higher AUC when compared with the univariate models in both training and validation data sets with an AUC of 0.966 (95% confidence interval: 0.949–0.99) and 0.930 (95% confidence interval: 0.905–0.973), respectively. The calibration curves indicated a good agreement between the predictable and the actual probability of developing recurrence. The DCA demonstrated that the predictive value of the model improved when combining the three MRI sequences.

Conclusion

Our multiparametric radiomics model could be used as an efficient and accurate tool for predicting the recurrence of grade II glioma.

Virtual Monoenergetic Images of Dual-Energy CT-Impact on Repeatability, Reproducibility, and Classification in Radiomics

Simple Summary

Virtual monoenergetic images from dual-energy CT are incrementally used in routine clinical practice. Thus, radiomic analysis will be more often performed on these images in the future. This study characterized the test–retest repeatability and reproducibility of radiomic features from virtual monoenergetic images and their impact on machine-learning-based lesion classification. The results of this study provide a basis to improve radiomic analyses and identify the role of feature stability in classification tasks when using virtual monoenergetic imaging with different scan or reconstruction parameters in multicenter clinical studies.

Abstract

The purpose of this study was to (i) evaluate the test–retest repeatability and reproducibility of radiomic features in virtual monoenergetic images (VMI) from dual-energy CT (DECT) depending on VMI energy (40, 50, 75, 120, 190 keV), radiation dose (5 and 15 mGy), and DECT approach (dual-source and split-filter DECT) in a phantom (ex vivo), and (ii) to assess the impact of VMI energy and feature repeatability on machine-learning-based classification in vivo in 72 patients with 72 hypodense liver lesions. Feature repeatability and reproducibility were determined by concordance–correlation–coefficient (CCC) and dynamic range (DR) ≥0.9. Test–retest repeatability was high within the same VMI energies and scan conditions (percentage of repeatable features ranging from 74% for SFDE mode at 40 keV and 15 mGy to 86% for DSDE at 190 keV and 15 mGy), while reproducibility varied substantially across different VMI energies and DECTs (percentage of reproducible features ranging from 32.8% for SFDE at 5 mGy comparing 40 with 190 keV to 99.2% for DSDE at 15 mGy comparing 40 with 50 keV). No major differences were observed between the two radiation doses (<10%) in all pair-wise comparisons. In vivo, machine learning classification using penalized regression and random forests resulted in the best discrimination of hemangiomas and metastases at low-energy VMI (40 keV), and for cysts at high-energy VMI (120 keV). Feature selection based on feature repeatability did not improve classification performance. Our results demonstrate the high repeatability of radiomics features when keeping scan and reconstruction conditions constant. Reproducibility diminished when using different VMI energies or DECT approaches. The choice of optimal VMI energy improved lesion classification in vivo and should hence be adapted to the specific task.