Apparent Variation in Measurement Size of Colorectal Cancer Metastases to the Liver on Dual Contrast MRI.

PURPOSE: The aim of this study was to formally investigate the apparent variation in lesion size of hepatic metastatic lesions from colorectal cancer on hepatobiliary phase (HBP) and dual contrast images of magnetic resonance imaging performed with both hepatobiliary and extracellular contrast agents. METHODS: Patients with known colorectal carcinoma who had undergone dual contrast liver magnetic resonance imaging were identified in our institutional database. Metastatic lesions were measured semiautomatically on both HBP and dual contrast images with a custom software tool that automatically identifies the lesion edge and thereby the lesion diameter. Lesion measurements from both sets of images were compared with a Student t test and Bland-Altman analysis. Lesions were also measured on both HBP and dual contrast images by 2 fellowship-trained abdominal radiologists. Measurements from the software and radiologists were compared with a Student t test and Bland-Altman analysis; interreader agreement was evaluated with the intraclass correlation coefficient. RESULTS: A total of 70 liver lesions in 39 patients was identified. Software-based measurements were significantly larger on HBP than dual contrast images ( P < 0.001), with a mean lesion size of 10.9 ± 4.2 mm for HBP and 10.5 ± 4.2 mm for dual contrast measurements. Radiologist-based measurements showed a similar trend, with HBP measurements being significantly larger than dual contrast measurements ( P < 0.001). Bland-Altman analysis indicated a mean bias ± 2 SD of +0.4 ± 1.6 mm for software-based measurements and +0.9 ± 2.9 mm and +0.7 ± 2.1 mm for readers 1 and 2, respectively. The intraclass correlation coefficient for interreader agreement was 0.9. CONCLUSIONS: Both software-based and radiologist-based measurements of colorectal cancer liver metastases are significantly larger on HBP than dual contrast images. Based on these findings, we recommend that longitudinal assessment be performed consistently on either HBP or dual contrast phases to avoid introduction of avoidable variability.

ouR Space Commentary: Wellness Perspective From SABI's Early Career Committee.

ABSTRACT: Radiologists and members-in-training are experiencing higher (and escalating) rates of burnout, resulting in a profound impact on the health of physicians, patients, and the community. Lately, the radiology community has demonstrated a growing awareness of this phenomenon, which has led to emphasis on practicing and promoting wellness. With a myriad of factors contributing to burnout in radiology, a multifaceted approach is pivotal for counteracting burnout and fostering overall well-being, including efforts driven at both organizational and individual levels. This article discusses perspectives from the members of the Early Career Committee at the Society for Advanced Body Imaging (SABI); it explores their beliefs and practical strategies for maintaining personal well-being.

Consensus-Based Technical Recommendations for Clinical Translation of Renal Phase Contrast MRI.

BACKGROUND: Phase-contrast (PC) MRI is a feasible and valid noninvasive technique to measure renal artery blood flow, showing potential to support diagnosis and monitoring of renal diseases. However, the variability in measured renal blood flow values across studies is large, most likely due to differences in PC-MRI acquisition and processing. Standardized acquisition and processing protocols are therefore needed to minimize this variability and maximize the potential of renal PC-MRI as a clinically useful tool. PURPOSE: To build technical recommendations for the acquisition, processing, and analysis of renal 2D PC-MRI data in human subjects to promote standardization of renal blood flow measurements and facilitate the comparability of results across scanners and in multicenter clinical studies. STUDY TYPE: Systematic consensus process using a modified Delphi method. POPULATION: Not applicable. SEQUENCE FIELD/STRENGTH: Renal fast gradient echo-based 2D PC-MRI. ASSESSMENT: An international panel of 27 experts from Europe, the USA, Australia, and Japan with 6 (interquartile range 4-10) years of experience in 2D PC-MRI formulated consensus statements on renal 2D PC-MRI in two rounds of surveys. Starting from a recently published systematic review article, literature-based and data-driven statements regarding patient preparation, hardware, acquisition protocol, analysis steps, and data reporting were formulated. STATISTICAL TESTS: Consensus was defined as ≥75% unanimity in response, and a clear preference was defined as 60-74% agreement among the experts. RESULTS: Among 60 statements, 57 (95%) achieved consensus after the second-round survey, while the remaining three showed a clear preference. Consensus statements resulted in specific recommendations for subject preparation, 2D renal PC-MRI data acquisition, processing, and reporting. DATA CONCLUSION: These recommendations might promote a widespread adoption of renal PC-MRI, and may help foster the set-up of multicenter studies aimed at defining reference values and building larger and more definitive evidence, and will facilitate clinical translation of PC-MRI. LEVEL OF EVIDENCE: 1 TECHNICAL EFFICACY STAGE: 1.

Fully automated intracardiac 4D flow MRI post-processing using deep learning for biventricular segmentation.

OBJECTIVES: 4D flow MRI allows for a comprehensive assessment of intracardiac blood flow, useful for assessing cardiovascular diseases, but post-processing requires time-consuming ventricular segmentation throughout the cardiac cycle and is prone to subjective errors. Here, we evaluate the use of automatic left and right ventricular (LV and RV) segmentation based on deep learning (DL) network that operates on short-axis cine bSSFP images. METHODS: A previously published DL network was fine-tuned via retraining on a local database of 106 subjects scanned at our institution. In 26 test subjects, the ventricles were segmented automatically by the network and manually by 3 human observers on bSSFP MRI. The bSSFP images were then registered to the corresponding 4D flow images to apply the segmentation to 4D flow velocity data. Dice coefficients and the relative deviation between measurements (automatic vs. manual and interobserver manual) of various hemodynamic parameters were assessed. RESULTS: The automated segmentation resulted in similar Dice scores (LV: 0.92, RV: 0.86) and lower relative deviations from manual segmentation in left ventricular (LV) average kinetic energy (KE) (8%) and RV KE (15%) than the Dice scores (LV: 0.91, RV: 0.87) and relative deviations between manual segmentation observers (LV KE: 11%, p = 0.01; RV KE: 19%, p = 0.03). CONCLUSIONS: The automated post-processing method using deep learning resulted in hemodynamic measurements that differ from a manual observer's measurements equally or less than the variation between manual observers. This approach can be used to decrease post-processing time on intraventricular 4D flow data and mitigate interobserver variability. KEY POINTS: • Our proposed method allows for fully automated post-processing of intraventricular 4D flow MRI data. • Our method resulted in hemodynamic measurements that matched those derived from manual segmentation equally as well as interobserver variability. • Our method can be used to greatly accelerate intraventricular 4D flow post-processing and improve interobserver repeatability.

Muscle velocity and inertial force from phase contrast MRI.

PURPOSE: To evaluate velocity waveforms in muscle and to create a tool and algorithm for computing and analyzing muscle inertial forces derived from 2D phase contrast (PC) magnetic resonance imaging (MRI). MATERIALS AND METHODS: PC MRI was performed in the forearm of four healthy volunteers during 1 Hz cycles of wrist flexion-extension as well as in the lower leg of six healthy volunteers during 1 Hz cycles of plantarflexion-dorsiflexion. Inertial forces (F) were derived via the equation F = ma. The mass, m, was derived by multiplying voxel volume by voxel-by-voxel estimates of density via fat-water separation techniques. Acceleration, a, was obtained via the derivative of the PC MRI velocity waveform. RESULTS: Mean velocities in the flexors of the forearm and lower leg were 1.94 ± 0.97 cm/s and 5.57 ± 2.72 cm/s, respectively, as averaged across all subjects; the inertial forces in the flexors of the forearm and lower leg were 1.9 × 10(-3) ± 1.3 × 10(-3) N and 1.1 × 10(-2) ± 6.1 × 10(-3) N, respectively, as averaged across all subjects. CONCLUSION: PC MRI provided a promising means of computing muscle velocities and inertial forces-providing the first method for quantifying inertial forces.

Simultaneous MRI of lung structure and perfusion in a single breathhold.

PURPOSE: To develop and demonstrate a breathheld 3D radial ultrashort echo time (UTE) acquisition to visualize co-registered lung perfusion and vascular structure. MATERIALS AND METHODS: Nine healthy dogs were scanned twice at 3 Tesla (T). Contrast-enhanced pulmonary perfusion scans were acquired with a temporally interleaved three-dimensional (3D) radial UTE (TE = 0.08 ms) sequence in a breathhold (1 s time frames over a 33 s breathhold). The 3D breathheld volume was reconstructed into time-resolved perfusion datasets, and a composite vascular structure dataset. For structural comparison, a 5 min respiratory-gated 3D radial UTE scan was acquired. Data were analyzed by quantitative metrics and radiologist scoring. RESULTS: Appropriate time-course of contrast was seen in all subjects. Right ventricle to aorta transit times were 7.4 ± 2.0 s. Relative lung enhancement was a factor of 8.4 ± 1.5. Radiologist scoring showed similarly excellent visualization of the pulmonary arteries to the subsegmental level in breathheld (94% of cases) and respiratory-gated (100% of cases) acquisitions (P = 0.33) despite the aggressive under sampling in the breathheld scan. Similarly, differentiation of lung tissue and airways was achieved by both acquisition methods. CONCLUSION: A time-resolved 3D radial UTE sequence for simultaneous imaging of pulmonary perfusion and co-registered vascular structure is feasible.

Measurements of wall shear stress and aortic pulse wave velocity in swine with familial hypercholesterolemia.

PURPOSE: To assess measurements of pulse wave velocity (PWV) and wall shear stress (WSS) in a swine model of atherosclerosis. MATERIALS AND METHODS: Nine familial hypercholesterolemic (FH) swine with angioplasty balloon catheter-induced atherosclerotic lesions to the abdominal aorta (injured group) and 10 uninjured FH swine were evaluated with a 4D phase contrast (PC) magnetic resonance imaging (MRI) acquisition, as well as with radial and Cartesian 2D PC acquisitions, on a 3T MR scanner. PWV values were computed from the 2D and 4D PC techniques, compared between the injured and uninjured swine, and validated against reference standard pressure probe-based PWV measurements. WSS values were also computed from the 4D PC MRI technique and compared between injured and uninjured groups. RESULTS: PWV values were significantly greater in the injured than in the uninjured groups with the 4D PC MRI technique (P = 0.03) and pressure probes (P = 0.02). No significant differences were found in PWV between groups using the 2D PC techniques (P = 0.75-0.83). No significant differences were found for WSS values between the injured and uninjured groups. CONCLUSION: The 4D PC MRI technique provides a promising means of evaluating PWV and WSS in a swine model of atherosclerosis, providing a potential platform for developing the technique for the early detection of atherosclerosis.

Aortic pulse wave velocity measurements with undersampled 4D flow-sensitive MRI: comparison with 2D and algorithm determination.

PURPOSE: To compare pulse wave velocity (PWV) measurements obtained from radially undersampled 4D phase-contrast magnetic resonance imaging (PC-MRI) with 2D PC measurements and to evaluate four PWV algorithms. MATERIALS AND METHODS: PWV was computed from radially undersampled 3D, 3-directionally velocity-encoded PC-MRI (4D) acquisitions performed on a 3T MR scanner in 18 volunteers. High temporal resolution 2D PC scans serving as a reference standard were available in 14 volunteers. Four PWV algorithms were tested: time-to-upstroke (TTU), time-to-peak (TTP), time-to-foot (TTF), and cross-correlation (XCorr). Bland-Altman analysis was used to determine inter- and intraobserver reproducibility and to compare differences between algorithms. Differences in age and PWV measurements were analyzed with Student's t-tests. The variability of age-corrected data was assessed with a Brown-Forsythe analysis of variance (ANOVA) test. RESULTS: 2D (4.6-5.3 m/s) and 4D (3.8-4.8 m/s) PWV results were in agreement with previously reported values in healthy subjects. Of the four PWV algorithms, the TTU, TTF, and XCorr algorithms gave similar and reliable results. Average biases of +0.30 m/s and -0.01 m/s were determined for intra- and interobserver variability, respectively. The Brown-Forsythe test revealed that no differences in variability could be found between 2D and 4D PWV measurements. CONCLUSION: 4D PC-MRI with radial undersampling provides reliable and reproducible measurements of PWV. TTU, TTF, and XCorr were the preferred PWV algorithms.

Synthetic dual-energy CT reconstruction from single-energy CT Using artificial intelligence.

PURPOSE: To develop and assess the utility of synthetic dual-energy CT (sDECT) images generated from single-energy CT (SECT) using two state-of-the-art generative adversarial network (GAN) architectures for artificial intelligence-based image translation. METHODS: In this retrospective study, 734 patients (389F; 62.8 years ± 14.9) who underwent enhanced DECT of the chest, abdomen, and pelvis between January 2018 and June 2019 were included. Using 70-keV as the input images (n = 141,009) and 50-keV, iodine, and virtual unenhanced (VUE) images as outputs, separate models were trained using Pix2PixHD and CycleGAN. Model performance on the test set (n = 17,839) was evaluated using mean squared error, structural similarity index, and peak signal-to-noise ratio. To objectively test the utility of these models, synthetic iodine material density and 50-keV images were generated from SECT images of 16 patients with gastrointestinal bleeding performed at another institution. The conspicuity of gastrointestinal bleeding using sDECT was compared to portal venous phase SECT. Synthetic VUE images were generated from 37 patients who underwent a CT urogram at another institution and model performance was compared to true unenhanced images. RESULTS: sDECT from both Pix2PixHD and CycleGAN were qualitatively indistinguishable from true DECT by a board-certified radiologist (avg accuracy 64.5%). Pix2PixHD had better quantitative performance compared to CycleGAN (e.g., structural similarity index for iodine: 87% vs. 46%, p-value < 0.001). sDECT using Pix2PixHD showed increased bleeding conspicuity for gastrointestinal bleeding and better removal of iodine on synthetic VUE compared to CycleGAN. CONCLUSIONS: sDECT from SECT using Pix2PixHD may afford some of the advantages of DECT.

Differentiation of benign from malignant solid renal lesions using CT-based radiomics and machine learning: comparison with radiologist interpretation.

PURPOSE: To assess the performance of a machine learning model trained with contrast-enhanced CT-based radiomics features in distinguishing benign from malignant solid renal masses and to compare model performance with three abdominal radiologists. METHODS: Patients who underwent intra-operative ultrasound during a partial nephrectomy were identified within our institutional database, and those who had pre-operative contrast-enhanced CT examinations were selected. The renal masses were segmented from the CT images and radiomics features were derived from the segmentations. The pathology of each mass was identified; masses were labeled as either benign [oncocytoma or angiomyolipoma (AML)] or malignant [clear cell, papillary, or chromophobe renal cell carcinoma (RCC)] depending on the pathology. The data were parsed into a 70/30 train/test split and a random forest machine learning model was developed to distinguish benign from malignant lesions. Three radiologists assessed the cohort of masses and labeled cases as benign or malignant. RESULTS: 148 masses were identified from the cohort, including 50 benign lesions (23 AMLs, 27 oncocytomas) and 98 malignant lesions (23 clear cell RCC, 44 papillary RCC, and 31 chromophobe RCCs). The machine learning algorithm yielded an overall accuracy of 0.82 for distinguishing benign from malignant lesions, with an area under the receiver operating curve of 0.80. In comparison, the three radiologists had significantly lower accuracies (p = 0.02) ranging from 0.67 to 0.75. CONCLUSION: A machine learning model trained with CT-based radiomics features can provide superior accuracy for distinguishing benign from malignant solid renal masses compared to abdominal radiologists.

Diagnostic performance of the "drooping" sign in CT diagnosis of exophytic renal angiomyolipoma.

OBJECTIVE: To evaluate the prevalence of angular interface and the "drooping" sign in exophytic renal angiomyolipomas (AMLs) and the diagnostic performance in differentiating exophytic lipid-poor AMLs from other solid renal masses. METHODS: This IRB-approved, two-center study included 185 patients with 188 exophytic solid renal masses < 4 cm with histopathology and pre-operative CT within 30 days of surgical resection or biopsy. Images were reviewed for the presence of angular interface and the "drooping" sign qualitatively by three readers blinded to the final diagnosis, with majority rules applied. Both features were assessed quantitatively by cohort creators (who are not readers) independently. Free-marginal kappa was used to assess inter-reader agreement and agreement between two methods assessing each feature. Fisher's exact test, Mann-Whitney test, and multivariable logistic regression with two-tailed p < 0.05 were used to determine statistical significance. Diagnostic performance was assessed. RESULTS: Ninety-four patients had 96 AMLs, and 91 patients had 92 non-AMLs. Seventy-four (77%) of AMLs were lipid-poor based on quantitative assessment on CT. The presence of angular interface and the "drooping" sign by both qualitative and quantitative assessment were statistically significantly associated with AMLs (39% (qualitative) and 45% (quantitative) vs 15% (qualitative) and 13% (quantitative), and 48% (qualitative) and 43% (quantitative) vs 4% (qualitative) and 1% (quantitative), respectively, all p < 0.001) in univariable analysis. In multivariable analysis, only the "drooping" sign in either qualitative or quantitative assessment was a statistically significant predictor of AMLs (both p < 0.001). Inter-reader agreement for the "drooping" sign was moderate (k = 0.55) and for angular interface was fair (k = 0.33). Agreement between the two methods of assessing the "drooping" sign was substantial (k = 0.84) and of assessing the angular interface was moderate (k = 0.59). The "drooping" sign both qualitatively and quantitatively, alone or in combination of angular interface, had very high specificity (96-100%) and positive predictive value (PPV) (89-100%), moderate negative predictive value (62-68%), but limited sensitivity (23-49%) for lipid-poor AMLs. CONCLUSION: The "drooping" sign by both qualitative and quantitative assessment is highly specific for lipid-rich and lipid-poor AMLs. This feature alone or in combination with angular interface can aid in CT diagnosis of lipid-poor AMLs with very high specificity and PPV.

Understanding the Tumor Immune Microenvironment in Renal Cell Carcinoma.

Scientific understanding of how the immune microenvironment interacts with renal cell carcinoma (RCC) has substantially increased over the last decade as a result of research investigations and applying immunotherapies, which modulate how the immune system targets and eliminates RCC tumor cells. Clinically, immune checkpoint inhibitor therapy (ICI) has revolutionized the treatment of advanced clear cell RCC because of improved outcomes compared to targeted molecular therapies. From an immunologic perspective, RCC is particularly interesting because tumors are known to be highly inflamed, but the mechanisms underlying the inflammation of the tumor immune microenvironment are atypical and not well described. While technological advances in gene sequencing and cellular imaging have enabled precise characterization of RCC immune cell phenotypes, multiple theories have been suggested regarding the functional significance of immune infiltration in RCC progression. The purpose of this review is to describe the general concepts of the anti-tumor immune response and to provide a detailed summary of the current understanding of the immune response to RCC tumor development and progression. This article describes immune cell phenotypes that have been reported in the RCC microenvironment and discusses the application of RCC immunophenotyping to predict response to ICI therapy and patient survival.

Opportunistic assessment of ischemic heart disease risk using abdominopelvic computed tomography and medical record data: a multimodal explainable artificial intelligence approach.

Current risk scores using clinical risk factors for predicting ischemic heart disease (IHD) events-the leading cause of global mortality-have known limitations and may be improved by imaging biomarkers. While body composition (BC) imaging biomarkers derived from abdominopelvic computed tomography (CT) correlate with IHD risk, they are impractical to measure manually. Here, in a retrospective cohort of 8139 contrast-enhanced abdominopelvic CT examinations undergoing up to 5 years of follow-up, we developed multimodal opportunistic risk assessment models for IHD by automatically extracting BC features from abdominal CT images and integrating these with features from each patient's electronic medical record (EMR). Our predictive methods match and, in some cases, outperform clinical risk scores currently used in IHD risk assessment. We provide clinical interpretability of our model using a new method of determining tissue-level contributions from CT along with weightings of EMR features contributing to IHD risk. We conclude that such a multimodal approach, which automatically integrates BC biomarkers and EMR data, can enhance IHD risk assessment and aid primary prevention efforts for IHD. To further promote research, we release the Opportunistic L3 Ischemic heart disease (OL3I) dataset, the first public multimodal dataset for opportunistic CT prediction of IHD.

MR measures of renal perfusion, oxygen bioavailability and total renal blood flow in a porcine model: noninvasive regional assessment of renal function.

BACKGROUND: Magnetic resonance imaging (MRI) may be a useful adjunct to current methods of evaluating renal function. MRI is a noninvasive imaging modality that has the ability to evaluate the kidneys regionally, which is lacking in current clinical methods. Other investigators have evaluated renal function with MRI-based measurements, such as with techniques to measure cortical and medullary perfusion, oxygen bioavailability and total renal blood flow (TRBF). However, use of all three techniques simultaneously, and therefore the relationships between these MRI-derived functional parameters, have not been reported previously. METHODS: To evaluate the ability of these MRI techniques to track changes in renal function, we scanned 11 swine during a state of hyperperfusion with acetylcholine and a saline bolus and subsequently scanned during a state of hypoperfusion with the prolonged use of isoflurane anesthesia. For each time point, measurements of perfusion, oxygen bioavailability and TRBF were acquired. Measurements of perfusion and oxygen bioavailability were compared with measurements of TRBF for all swine across all time points. RESULTS: Cortical perfusion, cortical oxygen bioavailability, medullary oxygen bioavailability and TRBF significantly increased with the acetylcholine challenge. Cortical perfusion, medullary perfusion, cortical oxygen bioavailability and TRBF significantly decreased during isoflurane anesthesia. Cortical perfusion (Spearman's correlation coefficient = 0.68; P < 1 × 10(-6)) and oxygen bioavailability (Spearman's correlation coefficient = -0.60; P < 0.0001) correlated significantly with TRBF, whereas medullary perfusion and oxygen bioavailability did not correlate with TRBF. CONCLUSIONS: Our results demonstrate expected changes given the pharmacologically induced changes in renal function. Maintenance of the medullary oxygen bioavailability in low blood flow states may reflect the autoregulation particular to this region of the kidney. The ability to non-invasively measure all three parameters of kidney function in a single MRI examination and to evaluate the relationships between these functional parameters is potentially useful for evaluating the state of the human kidneys in situ in future studies.

Natural history of simple renal cysts: longitudinal CT-based evaluation.

PURPOSE: Simple renal cysts are common benign lesions that arise from the renal parenchyma. Cyst growth can lead to confusion as well as concern from patients and referring providers about the need for imaging follow-up or additional evaluation. The purpose of this study was to evaluate the natural history of simple renal cysts and determine the best metric to characterize cyst evolution. METHODS: 222 simple renal cysts in 182 adults (age = 58.4 ± 6.0 years) were longitudinally evaluated on non-contrast CT examinations over a mean interval of 7.5 ± 2.8 years. Axial long axis, surface area, and volume were evaluated at baseline and follow-up CT examinations. Absolute and percent annualized growth rates were computed between CT studies for each parameter. RESULTS: At baseline CT examinations, mean (± SD) axial long axis, surface area, and volume measurements were 2.5 ± 1.7 cm, 2.5 ± 4.5 cm2, and 17.6 ± 52.5 ml, respectively. On follow-up examinations, measurements were 3.4 ± 2.0 cm, 4.2 ± 5.9 cm2, and 34.4 ± 92.3 ml, respectively. Significant differences (p < 0.01) were found between baseline and follow-up values for each parameter. The absolute growth rate of each parameter was + 0.1 ± 0.1 cm/year, + 2.1 ± 3.4 cm2/year, and + 2.0 ± 5.6 ml/year, respectively. The percent annualized growth rate for each parameter was +6.5 ± 7.3%/year, +18 ± 24%/year, and +46 ± 100%/year, respectively. Overall, 86% (190/222) of cysts increased in size over time; most notably 78% (174/222) increased by ≥ 6% in volume per year. None of the simple cysts developed septations or solid components on follow-up examinations. CONCLUSION: The majority of simple renal cysts increase in size over time, which was not associated with the development of complex features. Surface area and volume are the parameters most indicative of cyst growth or regression over time. In patients with enlarging asymptomatic simple renal cysts, no follow-up imaging is indicated.

Differentiation of benign from malignant solid renal lesions with MRI-based radiomics and machine learning.

BACKGROUND: Solid renal masses are often indeterminate for benignity versus malignancy on magnetic resonance imaging. Such masses are typically evaluated with either percutaneous biopsy or surgical resection. Percutaneous biopsy can be non-diagnostic and some surgically resected lesions are inadvertently benign. PURPOSE: To assess the performance of ten machine learning (ML) algorithms trained with MRI-based radiomics features in distinguishing benign from malignant solid renal masses. METHODS: Patients with solid renal masses identified on pre-intervention MRI were curated from our institutional database. Masses with a definitive diagnosis via imaging (for angiomyolipomas) or via biopsy or surgical resection (for oncocytomas or renal cell carcinomas) were selected. Each mass was segmented for both T2- and post-contrast T1-weighted images. Radiomics features were derived from the segmented masses for each imaging sequence. Ten ML algorithms were trained with the radiomics features gleaned from each MR sequence, as well as the combination of MR sequences. RESULTS: In total, 182 renal masses in 160 patients were included in the study. The support vector machine algorithm trained on radiomics features from T2-weighted images performed superiorly, with an accuracy of 0.80 and an area under the curve (AUC) of 0.79. Linear discriminant analysis (accuracy = 0.84 and AUC = 0.77) and logistic regression (accuracy = 0.78 and AUC = 0.78) algorithms trained on T2-based radiomics features performed similarly. ML algorithms trained on radiomics features from post-contrast T1-weighted images or the combination of radiomics features from T2- and post-contrast T1-weighted images yielded lower performance. CONCLUSION: Machine learning models trained with radiomics features derived from T2-weighted images can provide high accuracy for distinguishing benign from malignant solid renal masses. CLINICAL IMPACT: Machine learning models derived from MRI-based radiomics features may improve the clinical management of solid renal masses and have the potential to reduce the frequency with which benign solid renal masses are biopsied or surgically resected.