PURPOSE:To evaluate retrievable inferior vena cava (IVC) filter outcomes with a prospectively maintained database and active patient management by interventional radiology (IR). MATERIALS & METHODS:Patients with retrievable IVC filters placed by IR from 2012 to 2023 at a single, tertiary institution were tracked in a prospective registry, and follow-up was organized by a designated IR physician assistant. Patients were contacted after the filter was placed by IR and a clinic visit arranged; filter removal was scheduled when deemed appropriate. Retrospective review of filter outcomes, including retrieval, patient death, and need for permanent filtration, was performed. RESULTS:Over the 12-year study period, 607 retrievable IVC filters were placed: 516 Denali, 63 Eclipse, 19 Günther Tulip, and 9 Celect Platinum. In total, 43% (260) were retrieved, 12% (75) were adjudicated to be permanent, and 42% (253) died with the filter in place. The remaining 3% (19) comprised patients alive with the filter not yet retrieved at study endpoint, 42% (8/19) of which were placed in 2023. Of this cohort, 8 still needed the filter and were being monitored to determine follow-up timing, 2 needed a follow-up appointment, and 9 were lost to follow-up due to repeated no-shows and/or inability to reach the patient despite multiple attempts. Overall, 1.5% (9/607) of all filters placed were not accounted for. CONCLUSION:This study demonstrates high accountability (98.5%) of retrievable IVC filters when using a prospective registry actively managed by an IR PA, providing an effective and feasible model for facilitating appropriate follow-up.

This study evaluated the use of the grasp-and-fold technique for complex forceps retrieval of inferior vena cava (IVC) filters. A retrospective study of 14 patients (12 women and 2 men) who had either deeply tip-embedded or severely distorted IVC filters was performed at a single institution over 10 years. In this technique, endobronchial forceps were used to fold the filter in half to remove it through the sheath because the filter tip could not be accessed by dissection. The grasp-and-fold technique successfully removed all 14 filters. One patient had retained filter struts, which were present before the procedure. One mild and 5 moderate adverse events (AEs), including fracture fragment embolization requiring retrieval and self-limited IVC extravasation, occurred. No severe AEs occurred. In this small patient cohort, the grasp-and-fold forceps technique successfully retrieved deeply tip-embedded or distorted IVC filters with inaccessible tips.

Purpose: To evaluate differences in arteriographic findings and outcomes after embolization among patients with a sus-pected iatrogenic renal arterial injury (IRAI).Materials and Methods: Patients at the authors' institution who underwent renal arteriography for suspected IRAIs after partial nephrectomy, biopsy, or percutaneous access over a 20-year period were included. Records, imaging, and outcomes were reviewed. Data analysis was performed using the Fisher exact or Kruskal-Wallis test.Results: Ninety arteriograms were performed on 83 patients after partial nephrectomy (n = 32), biopsy (n = 27), or percutaneous access (n = 24), including for nephrostomy/ureterostomy and stone removal. The median number of days between the index procedure and arteriogram was highest (15 days) after partial nephrectomy and lowest (5 days) after biopsy (P = .0001). Embolization was performed during 76% of arteriograms. If prearteriographic imaging showed positive results for IRAIs, embolization was performed in 67% versus 33% if imaging showed negative results (P = .005). The transfusion rate was higher after biopsy than after partial nephrectomy or percutaneous access (P = .002). Acute kidney injury after arteriogram occurred in 7% of patients; however, all returned to baseline by 1 week.Conclusions: Despite the different mechanism of IRAIs in partial nephrectomy, biopsy, and percutaneous access, arte-riographic findings and outcomes were overall similar among groups. Prearteriographic imaging can help identify IRAIs but cannot supersede the clinical judgment regarding indication for embolization. IRAIs can present acutely or after a long interim, although patients who underwent biopsy presented earlier and more frequently required a blood transfusion. IRAIs can be treated with embolization without permanent deleterious effects on renal function.

Routine Removal of Inferior Vena Cava Filters This interactive feature about an incidentally discovered inferior vena cava filter offers a case vignette accompanied by two essays, one supporting removal of the inferior vena cava filter and the other supporting deferring removal unless complications arise.

Background:Pseudocirrhosis is defined by radiologic changes of the liver parenchyma secondary to metastatic disease and/or cancer treatments, and portends a high rate of morbidity and mortality from sequelae of portal hypertension. Transjugular intrahepatic portosystemic shunt (TIPS) is an effective treatment for portal hypertension; however, TIPS is relatively contraindicated in the setting of hepatic metastases. The study aims to determine the technical efficacy and clinical outcomes of patients undergoing TIPS for symptomatic pseudocirrhosis.Methods:Retrospective analysis of patients with hepatic malignancy who underwent TIPS between 2008 and 2020 at a single tertiary care center. Patients with imaging findings of pseudocirrhosis and without history of primary liver malignancy or confounding causes of cirrhosis were included. West Haven scores assessing hepatic encephalopathy were obtained from chart review. Technical success was defined as successful TIPS creation with reduction in the portosystemic gradient (PSG). Clinical success was defined as resolution of variceal bleeding and/or ascites.Results:Nine patients (4 female/5 male), average (± SD) age 61.2±9.5 years with metastatic pseudocirrhosis were included for analysis. Primary malignancy was colorectal adenocarcinoma (n=5), neuroendocrine tumor (n=3), and malignant endothelial hemangioendothelioma (n=1). Average Model for End Stage Liver Disease (MELD-Na) score was 15.7±3.7. Technical success was 8/9 (89%) with average PSG reduced from 23.5±11.0 to 6.5±2.8 mmHg (P=0.001). Clinical success was 6/9 (67%). Two patients required TIPS revision after initial clinical success. Mild-moderate HE occurred in 6/9 patients post TIPS (67%), with a highest West Haven score of 2. Time from TIPS to death for acute variceal bleeding and ascites was 4.9±4.2 and 12±16.5 months, respectively. Cause of death was disease progression (n=5), variceal bleeding (n=1), or unavailable (n=2).Conclusions:TIPS in the setting of malignant pseudocirrhosis can be created safely with similar clinical outcomes to TIPS performed for benign disease. Rates of low-grade hepatic encephalopathy may be higher amongst patients undergoing TIPS for pseudocirrhosis.

Purpose: To report results of 16 years of using the endobronchial forceps technique to remove embedded inferior vena cava (IVC) filters.Materials and Methods: Between January 2005 and June 2021, 534 patients (310 women and 224 men) with a mean age of 52 years (standard deviation [SD] +/- 16 years) presented for complex filter retrieval of 535 tip-or strut-embedded IVC filters. Tip-embedded filters were diagnosed on rotational venography. Filters were considered strut-embedded if they were closed-cell filters with dwell times of >6 months. The filter was dissected from the IVC using rigid bronchoscopy forceps and removed through a vascular sheath. Results: The endobronchial forceps technique was successful in 530 of 537 retrieval attempts on an intention-to-treat basis (98.7%); a total of 530 filters were retrieved. There were 7 failures: (a) 5 failed retrieval attempts (2 that were retrieved successfully in subsequent procedures) and (b) 2 for which retrieval was not attempted. The mean filter dwell time was 1,459 days (SD +/- 1,617 days). Laser sheaths were not used for any removal. Filters included herein were 137 Celect (94 Celect and 43 Celect Platinum), 99 Gunther Tulip, 72 Option (48 Option and 24 Option Elite), 68 G2, 45 G2X/Eclipse, 42 Denali, 30 OptEase, 29 Recovery, 7 Meridian, and 6 ALN with Hook filters. Thirty-four minor (6.3%) and 11 major (2%) adverse events (AEs) occurred, which did not result in permanent sequelae.Conclusions: Use of endobronchial forceps for removal of tip-and strut-embedded retrievable IVC filters is effective and has low AE rates.

PURPOSE:To evaluate the utility of computed tomography (CT) angiography before transarterial embolization (TAE) in predicting TAE's technical success for type II endoleaks (T2ELs).MATERIALS AND METHODS:Fifty-eight patients (mean age, 74.4 years; range, 46-89 years) who underwent attempted TAE for T2EL from July 2014 to August 2019 and underwent CT angiography before the procedure were included. Each CT angiography result was assessed for a feeding artery that was traceable over its entire course from either the superior mesenteric artery or the internal iliac artery to the endoleak cavity. TAE was performed using coils and was considered technically successful if embolization of the endoleak cavity and feeding artery was performed. The technical success rates were compared between patients with and without traceable feeding arteries.RESULTS:A fully traceable feeding artery supplying 75% (44/59) of endoleaks in the cohort was identified. TAE was technically successful in 95% (42/44) of these cases but only in 13% (2/15) of the cases without a fully traceable feeding artery (P < .001). When the inferior mesenteric artery was the feeding artery, it was always fully traceable, and TAE was technically successful in 100% (33/33) of the cases. When a lumbar artery was the feeding artery, it was fully traceable in only 42% (11/26) of the cases. When the lumbar artery was not fully traceable, TAE was technically successful in only 13% (2/15) of the cases.CONCLUSIONS:The traceability of a feeding artery over its entire course to an endoleak cavity using CT angiography was associated with TAE's technical success. Lumbar feeding arteries were less likely to be fully traceable. TAE's high failure rate when the feeding artery was not fully traceable suggests that translumbar embolization can be considered as an initial approach for theses patients.

Purpose: To investigate the incidence of infection in patients with prior biliary interventions undergoing hepatic embolotherapy following extended antibiotic prophylaxis using moxifloxacin monotherapy or a multidrug regimen. Material and methods: Under an Institutional Review Board-approved protocol, retrospective review of a quality assurance database identified all liver-directed therapies (LDTs) at a tertiary care center between 2010 and 2019 with biliary intervention prior to LDT Records were reviewed for infectious complications within 3 months of chemo- or radioembolization. Patients were categorized based on extended antibiotic prophylaxis regimen: oral moxifloxacin monotherapy or multidrug regimen of levofloxacin and metroniodazole plus preprocedural neomycin and erythromycin. Procedures without at least 2 months of clinical follow-up, hepatic ablation, and procedures without extended antibiotic prophylaxis were excluded Regression analysis was used to analyze multivariate data to detect a difference in infection rate. Results: Twenty-four chemoembolization and 58 radioembolization procedures were performed on 55 patients with prior biliary interventions. Forty-four used monotherapy and 38 used multidrug regimen. The incidence of infection was 16.7% (4/24) after chemoembolization and 13.8% (8/58) after radioembolization The incidence of infection in patients did not differ between antibiotic prophylaxis regimens (18.2% [8/44] with moxifloxacin monotherapy and 10.5% [4/38] multidrug regimen, P = .3) or between types of biliary interventions (24.1% [7/29] with bilioenteric anastomosis and 23.8% [5/21] biliary stenting, P = .3). Conclusions: The types of extended antibiotic prophylaxis (moxifloxacin monotherapy vs multitherapy), prior biliary intervention, and embolotherapy were not found to be associated with differences in the incidence of infectious complications in this population.

In the era of precision medicine, biopsies are playing an increasingly central role in cancer research and treatment paradigms; however, patient outcomes and analyses of biopsy quality, as well as impact on downstream clinical and research applications, remain underreported. Herein, we report biopsy safety and quality outcomes for percutaneous core biopsies of hepatocellular carcinoma (HCC) performed as part of a prospective clinical trial. Patients with a clinical diagnosis of HCC were enrolled in a prospective cohort study for the genetic, proteomic, and metabolomic profiling of HCC at two academic medical centers from April 2016 to July 2020. Under image guidance, 18G core biopsies were obtained using coaxial technique at the time of locoregional therapy. The primary outcome was biopsy quality, defined as tumor fraction in the core biopsy. 56 HCC lesions from 50 patients underwent 60 biopsy events with a median of 8 core biopsies per procedure (interquartile range, IQR, 7–10). Malignancy was identified in 45/56 (80.4%, 4 without pathology) biopsy events, including HCC (40/56, 71.4%) and cholangiocarcinoma (CCA) or combined HCC-CCA (5/56, 8.9%). Biopsy quality was highly variable with a median of 40% tumor in each biopsy core (IQR 10–75). Only 43/56 (76.8%) and 23/56 (41.1%) samples met quality thresholds for genomic or metabolomic/proteomic profiling, respectively, requiring expansion of the clinical trial. Overall and major complication rates were 5/60 (8.3%) and 3/60 (5.0%), respectively. Despite uniform biopsy protocol, biopsy quality varied widely with up to 59% of samples to be inadequate for intended purpose. This finding has important consequences for clinical trial design and highlights the need for quality control prior to applications in which the presence of benign cell types may substantially alter findings.

Abstract Background Up to 30% of breast cancer patients will eventually relapse with metastatic disease. With an increasing array of therapeutic options, there is an ongoing need for predictive biomarkers to help guide treatment strategies including sequencing of therapies in the metastatic setting. We sought to evaluate the prognostic and predictive potential of a panel-specific tumor mutational burden (TMB) in metastatic breast cancer patients. Methods METAMORPH is a prospective, longitudinal cohort study. Eligible patients (pts) had newly diagnosed or progressive metastatic breast cancer and enrolled prior to starting a new line of therapy (physician’s choice) at the University of Pennsylvania. Pts underwent tissue biopsy of a suspected metastatic site. Tumor samples were analyzed for mutations and copy number alterations (CNA’s) using our institution’s CLIA-certified Center for Personalized Diagnostics (CPD) targeted gene panel, which evolved over the course of the study from 20 genes to 152 genes. TMB-high (TMB-H) was defined as ≥3 mutations and/or copy-number gains (CNG) among 18 genes shared across all panel versions. Pts were followed for time to progression (TTP), progression-free survival (PFS), and overall survival (OS). The frequency of rapid progressors and rapid death (defined as having progressed or died within 3 months of enrollment, respectively) was assessed. Results Three hundred pts enrolled from 2013-2020, of whom 200 pts had CPD reports generated. Of these, 12 pts were excluded due to either no treatment change on enrollment (n=11) or different primary cancer on biopsy (n=1). Thus 188 pts were included in this analysis. The median age was 55 years (range 28-79). 77% of pts identified as white, 18% as Black or African American, and 3.2% as Asian. Pts had a median of 1 line (range 0-12) of prior systemic therapy in the metastatic setting. 46.8% had no prior therapies for MBC, while 31% had ≥3 prior lines of therapy. 74.4% were HR+, 22.8% TNBC, and 2.7% HR-/HER2+. 6.9% of the cohort were classified as TMB-H. The average mutation/CNG rate was 2.2/sample, and 22.5% had no mutations or CNA’s. The most common mutations were TP53 (35%) and PIK3CA (26%). While TMB-H patients showed a statistically non-significant trend towards shorter median TTP and PFS compared with TMB-L, they comprised a significantly greater proportion of rapid progressors (54.5% vs 24.1%, p=0.027), with an odds ratio for rapid progression of 3.8 (95% CI 1.08-13.2). In a multivariate logistic regression analysis, TMB-H remained independently associated with rapid progression when adjusted for receptor subtype and next line of therapy. Receptor subtype analysis revealed that ER- (including ER-/PR+) patients with TMB-H had a shorter median TTP compared to ER- TMB-L (147 vs 68 days, p=0.03). TMB-H was also associated with significantly shorter OS compared with TMB-L (587 vs 648 days, p=0.02; HR 2.2 [95% CI 1.11-4.41]). 44.4% of TMB-H pts died within 3 months of enrollment, as compared to 11.0% of TMB-L pts (p=0.005), with an odds ratio for rapid death, adjusted for number of previous lines of therapy and receptor subtype, of 6.7 (95% CI 1.5-31.0). Conclusion MBC pts who are TMB-H represent a population who are highly resistant to standard therapies, progress rapidly, and have significantly shorter overall survival with more rapid time to death. Our data support further studies investigating the utility of TMB as a predictive biomarker in directing patients away from standard treatment options and towards novel approaches e.g. immunotherapy. Citation Format: Igor Makhlin, Amy S Clark, Paul Wileyto, Noah Goodman, John Ndicu, Shannon DeLuca, Candace Clark, S. William Stavropoulos, Natalie Shih, Michael D Feldman, Susan M Domchek, Jennifer M Matro, Payal D Shah, Hayley M Knollman, Kevin R Fox, Kara N Maxwell, Lewis A Chodosh, Angela DeMichele. Investigating the clinical utility of tumor mutational burden in predicting rapid progression and death in patients with metastatic breast cancer [abstract]. In: Proceedings of the 2020 San Antonio Breast Cancer Virtual Symposium; 2020 Dec 8-11; San Antonio, TX. Philadelphia (PA): AACR; Cancer Res 2021;81(4 Suppl):Abstract nr PD9-10.

Clear cell renal cell carcinoma (ccRCC) is the most common subtype of renal cell carcinoma. Currently, there is a lack of noninvasive methods to stratify ccRCC prognosis prior to any invasive therapies. The purpose of this study was to preoperatively predict the tumor stage, size, grade, and necrosis (SSIGN) score of ccRCC using MRI-based radiomics. A multicenter cohort of 364 histopathologically confirmed ccRCC patients (272 low [< 4] and 92 high [≥ 4] SSIGN score) with preoperative T2-weighted and T1-contrast-enhanced MRI were retrospectively identified and divided into training (254 patients) and testing sets (110 patients). The performance of a manually optimized radiomics model was assessed by measuring accuracy, sensitivity, specificity, area under receiver operating characteristic curve (AUROC), and area under precision-recall curve (AUPRC) on an independent test set, which was not included in model training. Lastly, its performance was compared to that of a machine learning pipeline, Tree-Based Pipeline Optimization Tool (TPOT). The manually optimized radiomics model using Random Forest classification and Analysis of Variance feature selection methods achieved an AUROC of 0.89, AUPRC of 0.81, accuracy of 0.89 (95% CI 0.816–0.937), specificity of 0.95 (95% CI 0.875–0.984), and sensitivity of 0.72 (95% CI 0.537–0.852) on the test set. The TPOT using Extra Trees Classifier achieved an AUROC of 0.94, AUPRC of 0.83, accuracy of 0.89 (95% CI 0.816–0.937), specificity of 0.95 (95% CI 0.875–0.984), and sensitivity of 0.72 (95% CI 0.537–0.852) on the test set. Preoperative MR radiomics can accurately predict SSIGN score of ccRCC, suggesting its promise as a prognostic tool that can be used in conjunction with diagnostic markers.

Purpose To examine spinal interaction types and prevalence of inferior vena cava (IVC) filters in patients presenting for complex filter removal. Materials and Methods The records of 447 patients presenting for complex removal of IVC filters were reviewed, including patient demographics, IVC filter dwell time, filter fracture status, and computed tomography (CT) evidence of filter interaction with the spine. Spinal interaction was defined as a filter strut touching or penetrating into the vertebral body or disc. Patients with evidence of filter penetration and spinal interaction had abdominal CT that preceded filter removal assessed by 2 interventional radiologists to categorize the type of spinal interaction, including bony reaction and osteophyte formation. Results CT evidence of spinal interaction by the filter was found in 18% of patients (80/447). Interaction with the spine was more common in single point of fixation filters than filters with rails (P = .007) and was more likely in filters with round wires than flat wires (P = .0007). Patients with interaction had longer dwell times (mean [SD] 5.7 [4.46] y) compared with patients without interaction (mean [SD] 3.2 [3.85] y); this relationship was significant (P < .0001). Women were more likely than men to experience filter/spine interaction (P = .04). Filters with spinal interaction were more likely to be fractured (P = .001). Filter interaction was found in 38% (30/78) of patients with symptoms, including chest and back pain, compared with 14% (50/369) of patients without symptoms (P < .0001, odds ratio 3.99). Conclusions Retrievable IVC filters may interact with the spine. These interactions are associated with longer filter dwell times, female sex, and round wire filter construction.

Nearly all breast cancer deaths result from metastatic disease. Despite this, the genomic events that drive metastatic recurrence are poorly understood. We performed whole-exome and shallow whole-genome sequencing to identify genes and pathways preferentially mutated or copy-number altered in metastases compared with the paired primary tumors from which they arose. Seven genes were preferentially mutated in metastases - MYLK, PEAK1, SLC2A4RG, EVC2, XIRP2, PALB2, and ESR1 - 5 of which are not significantly mutated in any type of human primary cancer. Four regions were preferentially copy-number altered: loss of STK11 and CDKN2A/B, as well as gain of PTK6 and the membrane-bound progesterone receptor, PAQR8. PAQR8 gain was mutually exclusive with mutations in the nuclear estrogen and progesterone receptors, suggesting a role in treatment resistance. Several pathways were preferentially mutated or altered in metastases, including mTOR, CDK/RB, cAMP/PKA, WNT, HKMT, and focal adhesion. Immunohistochemical analyses revealed that metastases preferentially inactivate pRB, upregulate the mTORC1 and WNT signaling pathways, and exhibit nuclear localization of activated PKA. Our findings identify multiple therapeutic targets in metastatic recurrence that are not significantly mutated in primary cancers, implicate membrane progesterone signaling and nuclear PKA in metastatic recurrence, and provide genomic bases for the efficacy of mTORC1, CDK4/6, and PARP inhibitors in metastatic breast cancer.

Pre-treatment determination of renal cell carcinoma aggressiveness may help guide clinical decision-making. We aimed to differentiate low-grade (Fuhrman I–II) from high-grade (Fuhrman III–IV) renal cell carcinoma using radiomics features extracted from routine MRI. 482 pathologically confirmed renal cell carcinoma lesions from 2008 to 2019 in a multicenter cohort were retrospectively identified. 439 lesions with information on Fuhrman grade from 4 institutions were divided into training and test sets with an 8:2 split for model development and internal validation. Another 43 lesions from a separate institution were set aside for independent external validation. The performance of TPOT (Tree-Based Pipeline Optimization Tool), an automatic machine learning pipeline optimizer, was compared to hand-optimized machine learning pipeline. The best-performing hand-optimized pipeline was a Bayesian classifier with Fischer Score feature selection, achieving an external validation ROC AUC of 0.59 (95% CI 0.49–0.68), accuracy of 0.77 (95% CI 0.68–0.84), sensitivity of 0.38 (95% CI 0.29–0.48), and specificity of 0.86 (95% CI 0.78–0.92). The best-performing TPOT pipeline achieved an external validation ROC AUC of 0.60 (95% CI 0.50–0.69), accuracy of 0.81 (95% CI 0.72–0.88), sensitivity of 0.12 (95% CI 0.14–0.30), and specificity of 0.97 (95% CI 0.87–0.97). Automated machine learning pipelines can perform equivalent to or better than hand-optimized pipeline on an external validation test non-invasively predicting Fuhrman grade of renal cell carcinoma using conventional MRI.

The objectives of this study were to determine the effects of expanded polytetrafluoroethylene (PTFE)-covered stent location and TIPS extension on primary patency. This retrospective cohort study examined patients with PTFE-covered TIPS creation between 07/2002 and 06/2016. Clinical information and patency outcomes at 24 months were extracted. At TIPS creation, extension was performed at the discretion of the operator. Kaplan–Meier curves of primary patency were generated with conditional variables of HVO-HCJ distance, extension status, and covered versus uncovered extensions. Additional logistic regression analyses of distances were performed. Of 393 patients, 115 patients (29%) underwent stent extension, 79 (20% of total cohort) of which were at the HVO end alone. Primary patency for all TIPS was 75%, 68%, and 54% at 3, 6, and 12 months. The data endpoint were transplant or death in 92 (23%) and 116 (30%). Kaplan–Meier curves showed no statistically significant difference between the variables and primary patency at 12 and 24 months: distance up to versus greater than 10 mm (p = 0.32, 0.81); extension versus no extension (p = 0.83, 0.85); uncovered versus covered extensions (p = 0.58, 0.70). Logistic regression analyses showed a trend toward statistical significance. In the setting of PTFE-covered TIPS creation, extended TIPSs and unextended well-positioned TIPSs have no difference in primary patency rates. Stent position and extension length may have an effect on primary patency, but were likely obscured by “user recognition” effects.

AbstractPurposeThis study assesses the feasibility of deep learning detection and classification of 3 retrievable inferior vena cava filters with similar radiographic appearances and emphasizes the importance of visualization methods to confirm proper detection and classification.Materials and MethodsThe fast.ai library with ResNet-34 architecture was used to train a deep learning classification model. A total of 442 fluoroscopic images (N=144 patients) from inferior vena cava filter placement or removal were collected. Following image preprocessing, the training set included 382 images (110 Celect, 149 Denali, 123 Günther Tulip), of which 80% were used for training and 20% for validation. Data augmentation was performed for regularization. A random test set of 60 images (20 images of each filter type), not included in the training or validation set, was used for evaluation. Total accuracy and receiver operating characteristic area under the curve were used to evaluate performance. Feature heatmaps were visualized using guided backpropagation and gradient-weighted class activation mapping.ResultsThe overall accuracy was 80.2% with mean receiver operating characteristic area under the curve of 0.96 for the validation set (N=76), and 85.0% with mean receiver operating characteristic area under the curve of 0.94 for the test set (N=60). Two visualization methods were used to assess correct filter detection and classification.ConclusionsA deep learning model can be used to automatically detect and accurately classify inferior vena cava filters on radiographic images. Visualization techniques should be utilized to ensure deep learning models function as intended.

PURPOSE:With increasing incidence of renal mass, it is important to make a pretreatment differentiation between benign renal mass and malignant tumor. We aimed to develop a deep learning model that distinguishes benign renal tumors from renal cell carcinoma (RCC) by applying a residual convolutional neural network (ResNet) on routine MR imaging.EXPERIMENTAL DESIGN:Preoperative MR images (T2-weighted and T1-postcontrast sequences) of 1,162 renal lesions definitely diagnosed on pathology or imaging in a multicenter cohort were divided into training, validation, and test sets (70:20:10 split). An ensemble model based on ResNet was built combining clinical variables and T1C and T2WI MR images using a bagging classifier to predict renal tumor pathology. Final model performance was compared with expert interpretation and the most optimized radiomics model.RESULTS:Among the 1,162 renal lesions, 655 were malignant and 507 were benign. Compared with a baseline zero rule algorithm, the ensemble deep learning model had a statistically significant higher test accuracy (0.70 vs. 0.56, P = 0.004). Compared with all experts averaged, the ensemble deep learning model had higher test accuracy (0.70 vs. 0.60, P = 0.053), sensitivity (0.92 vs. 0.80, P = 0.017), and specificity (0.41 vs. 0.35, P = 0.450). Compared with the radiomics model, the ensemble deep learning model had higher test accuracy (0.70 vs. 0.62, P = 0.081), sensitivity (0.92 vs. 0.79, P = 0.012), and specificity (0.41 vs. 0.39, P = 0.770).CONCLUSIONS:Deep learning can noninvasively distinguish benign renal tumors from RCC using conventional MR imaging in a multi-institutional dataset with good accuracy, sensitivity, and specificity comparable with experts and radiomics.

BackgroundPretreatment determination of renal cell carcinoma aggressiveness may help to guide clinical decision‐making.PurposeTo evaluate the efficacy of residual convolutional neural network using routine MRI in differentiating low‐grade (grade I–II) from high‐grade (grade III–IV) in stage I and II renal cell carcinoma.Study TypeRetrospective.PopulationIn all, 376 patients with 430 renal cell carcinoma lesions from 2008–2019 in a multicenter cohort were acquired. The 353 Fuhrman‐graded renal cell carcinomas were divided into a training, validation, and test set with a 7:2:1 split. The 77 WHO/ISUP graded renal cell carcinomas were used as a separate WHO/ISUP test set.Field Strength/Sequence1.5T and 3.0T/T2‐weighted and T1 contrast‐enhanced sequences.AssessmentThe accuracy, sensitivity, and specificity of the final model were assessed. The receiver operating characteristic (ROC) curve and precision‐recall curve were plotted to measure the performance of the binary classifier. A confusion matrix was drawn to show the true positive, true negative, false positive, and false negative of the model.Statistical TestsMann–Whitney U‐test for continuous data and the chi‐square test or Fisher's exact test for categorical data were used to compare the difference of clinicopathologic characteristics between the low‐ and high‐grade groups. The adjusted Wald method was used to calculate the 95% confidence interval (CI) of accuracy, sensitivity, and specificity.ResultsThe final deep‐learning model achieved a test accuracy of 0.88 (95% CI: 0.73–0.96), sensitivity of 0.89 (95% CI: 0.74–0.96), and specificity of 0.88 (95% CI: 0.73–0.96) in the Fuhrman test set and a test accuracy of 0.83 (95% CI: 0.73–0.90), sensitivity of 0.92 (95% CI: 0.84–0.97), and specificity of 0.78 (95% CI: 0.68–0.86) in the WHO/ISUP test set.Data ConclusionDeep learning can noninvasively predict the histological grade of stage I and II renal cell carcinoma using conventional MRI in a multiinstitutional dataset with high accuracy.Level of Evidence3Technical Efficacy Stage2