Non-operative management of biliary complications after Liver Transplantation in pediatric patients: A 30-year experience.

BACKGROUND: To evaluate the efficacy of percutaneous and endoscopic therapeutic interventions for biliary strictures and leaks following LT in children. METHODS: Retrospective analysis of 49 consecutive pediatric liver transplant recipients (27 girls, 22 boys, mean age at transplant 3.9 years) treated at our institution from 1989 to 2019 for biliary leak and/or biliary stricture was performed. Minimally invasive approach was considered clinically successful if it resulted in patency of the narrowed biliary segment and/or correction of the biliary leak. RESULTS: Forty-two patients had a stricture at the biliary anastomosis; seven had a biliary leak. After an average 13.8 years of follow-up, long-term clinical success with minimally invasive treatment (no surgery or re-transplant) was achieved for 24 children (57%) with biliary stricture and 4 (57%) with biliary leaks. Eight patients required re-transplant; however, only one was due to failure of both percutaneous and surgical management. For biliary strictures, failure of non-surgical management was associated with younger age at stricture diagnosis (p < .02). CONCLUSIONS: Percutaneous and endoscopic management of biliary strictures and leaks after LT in children is associated with a durable result in >50% of children.

Intermediate to Long-Term Clinical Outcomes of Percutaneous Cryoablation for Renal Masses.

PURPOSE: The purpose of this study was to evaluate the effectiveness and adverse outcomes of percutaneous cryoablation (CA) for treatment of renal masses in a large cohort of patients. MATERIALS AND METHODS: This retrospective analysis included 299 CA procedures (297 masses in 277 patients) performed between July 2007 and May 2018 at a single institution. The mean patient age was 66.1 years (range, 30-93 years) with 65.8% being male. A total of 234 (78.8%) masses were biopsy-proven renal cell carcinoma (RCC). The mean maximal tumor diameter was 2.5 cm (range, 0.7-6.6 cm). Efficacy was assessed only for ablations of biopsy-proven RCC, whereas the evaluation of adverse events and renal function included all masses. Complications were graded according to the Society of International Radiology classification. RESULTS: Major complications occurred in 3.0% of procedures (n = 9), none of which resulted in death or permanent disability. The mean imaging follow-up period was 27.4 months (range, 1-115) for the 199 RCC patients (204 ablated tumors) with follow-up imaging available. Complete response on initial follow-up imaging at mean 4.2 months (range, 0.3-75.6) was achieved in 195 of 204 tumors (95.6%) after a single session and in 200 of 204 tumors (98.0%) after 1 or 2 sessions. Of the RCC patients achieving complete response initially, local recurrence during the follow-up period occurred in 3 of 200 tumors (1.5%). Metastatic progression occurred in 10 of 193 (5.2%) RCC patients without prior metastatic disease during follow-up. CONCLUSIONS: CA for renal masses is safe and remains efficacious through intermediate- and long-term follow-up.

Uterine Artery Embolization following Cesarean Delivery but prior to Hysterectomy in the Management of Patients with Invasive Placenta.

PURPOSE: To evaluate outcomes of patients with placenta accreta spectrum (PAS) disorders who underwent uterine artery embolization (UAE) following cesarean delivery but before hysterectomy. MATERIALS AND METHODS: A retrospective review of patients with PAS treated with cesarean-hysterectomy (C-hyst) was performed. Patients in the UAE group underwent UAE after cesarean delivery but before hysterectomy; patients in the control group underwent C-hyst alone. Estimated blood loss (EBL), transfusion requirements, length of intensive care unit (ICU) stay, and adverse events were evaluated. RESULTS: The study included 31 patients, 7 in the UAE group and 24 in the control group. Median EBL, transfusion requirements, and length of ICU stay in the UAE group compared with control group were 1,500 mL (range, 500-2,000 mL) vs 2,000 mL (range, 1,000-4,500 mL) (P = .04), 150 mL (range, 0-650 mL) vs 550 mL (range, 0-3,125 mL) (P = .10), and 0 d (range, 0-1 d) vs 0.5 d (range, 0-2 d) (P = .07). All patients in the UAE group had placenta increta; patients in the control group had placenta accreta (29%), increta (54%), and percreta (17%) (P = .10). Subgroup analysis of patients with placenta increta demonstrated that the UAE group had a significant decrease in median EBL (P = .004), transfusion requirements (P = .009), and length of ICU stay (P = .04). No adverse events following UAE were noted. CONCLUSIONS: UAE following cesarean delivery but before hysterectomy in patients with placenta increta appears to be safe and effective in decreasing EBL, transfusion requirements, and length of ICU stay compared with C-hyst alone.

Outcomes after Peripancreatic Fluid Drainage in Patients with Simultaneous Pancreas-Kidney Transplantation.

PURPOSE: To determine the clinical outcomes of patients who underwent image-guided drainage of peripancreatic fluid collections after simultaneous pancreas-kidney (SPK) transplantation. MATERIALS AND METHODS: A retrospective review of all patients who underwent peripancreatic fluid collection drainage after SPK, from January 2000 to August 2017, at a single institution was performed. Patient characteristics, surgical technique, medication regimen, microbial analysis, and clinical outcomes were reviewed. Thirty-one patients requiring a total of 41 drainages were included in this study. The median age was 44 years (range 30-58 years), and median time between SPK and drainage was 28 days (range 8 to 3,401 days). Fisher's exact test, unpaired Student t-tests, and Pearson correlations were used for statistical analysis. RESULTS: Fever (51%) and abdominal pain (31%) were the most common presenting symptoms. The average amount of fluid drained at the time of drain placement was 97 mL (SD 240 mL). The average time spent with a drain in place was 33 days (SD 31 days). Microorganisms were isolated in the fluid of 22 of 41 drainages (54%), with mixed gastrointestinal flora being the most common. No further intervention was needed in 34 of 41 drainages (82%). However, drainage failed in 5 of 31 patients (16%), requiring surgical intervention with removal of the pancreas transplant. CONCLUSIONS: Percutaneous drainage of peripancreatic fluid collections after SPK transplantation is a safe and effective treatment option.

Automated detection of moderate and large pneumothorax on frontal chest X-rays using deep convolutional neural networks: A retrospective study.

BACKGROUND: Pneumothorax can precipitate a life-threatening emergency due to lung collapse and respiratory or circulatory distress. Pneumothorax is typically detected on chest X-ray; however, treatment is reliant on timely review of radiographs. Since current imaging volumes may result in long worklists of radiographs awaiting review, an automated method of prioritizing X-rays with pneumothorax may reduce time to treatment. Our objective was to create a large human-annotated dataset of chest X-rays containing pneumothorax and to train deep convolutional networks to screen for potentially emergent moderate or large pneumothorax at the time of image acquisition. METHODS AND FINDINGS: In all, 13,292 frontal chest X-rays (3,107 with pneumothorax) were visually annotated by radiologists. This dataset was used to train and evaluate multiple network architectures. Images showing large- or moderate-sized pneumothorax were considered positive, and those with trace or no pneumothorax were considered negative. Images showing small pneumothorax were excluded from training. Using an internal validation set (n = 1,993), we selected the 2 top-performing models; these models were then evaluated on a held-out internal test set based on area under the receiver operating characteristic curve (AUC), sensitivity, specificity, and positive predictive value (PPV). The final internal test was performed initially on a subset with small pneumothorax excluded (as in training; n = 1,701), then on the full test set (n = 1,990), with small pneumothorax included as positive. External evaluation was performed using the National Institutes of Health (NIH) ChestX-ray14 set, a public dataset labeled for chest pathology based on text reports. All images labeled with pneumothorax were considered positive, because the NIH set does not classify pneumothorax by size. In internal testing, our "high sensitivity model" produced a sensitivity of 0.84 (95% CI 0.78-0.90), specificity of 0.90 (95% CI 0.89-0.92), and AUC of 0.94 for the test subset with small pneumothorax excluded. Our "high specificity model" showed sensitivity of 0.80 (95% CI 0.72-0.86), specificity of 0.97 (95% CI 0.96-0.98), and AUC of 0.96 for this set. PPVs were 0.45 (95% CI 0.39-0.51) and 0.71 (95% CI 0.63-0.77), respectively. Internal testing on the full set showed expected decreased performance (sensitivity 0.55, specificity 0.90, and AUC 0.82 for high sensitivity model and sensitivity 0.45, specificity 0.97, and AUC 0.86 for high specificity model). External testing using the NIH dataset showed some further performance decline (sensitivity 0.28-0.49, specificity 0.85-0.97, and AUC 0.75 for both). Due to labeling differences between internal and external datasets, these findings represent a preliminary step towards external validation. CONCLUSIONS: We trained automated classifiers to detect moderate and large pneumothorax in frontal chest X-rays at high levels of performance on held-out test data. These models may provide a high specificity screening solution to detect moderate or large pneumothorax on images collected when human review might be delayed, such as overnight. They are not intended for unsupervised diagnosis of all pneumothoraces, as many small pneumothoraces (and some larger ones) are not detected by the algorithm. Implementation studies are warranted to develop appropriate, effective clinician alerts for the potentially critical finding of pneumothorax, and to assess their impact on reducing time to treatment.

Operator Dose Reduction during Transjugular Liver Biopsy Using a Radiation-Attenuating Drape: A Prospective, Randomized Study.

PURPOSE: To assess the effectiveness of disposable radiation-absorbing surgical drapes on operator radiation dose during transjugular liver biopsy (TJLB). MATERIALS AND METHODS: This dual-arm prospective, randomized study was conducted between May 2017 and January 2018 at a single institution. TJLB procedures (N = 62; patient age range, 19-80 y) were assigned at a 1:1 ratio to the use of radiation-absorbing surgical drapes or standard surgical draping. The primary outcome was cumulative radiation equivalent dose incident on the operator, as determined by an electronic personal dosimeter worn at the chest during each procedure. Cumulative kerma-area product (KAP), total fluoroscopy time, and total number of exposures used during each liver biopsy procedure were also determined. RESULTS: Mean radiation dose incident on the operator decreased by 56% with the use of radiation-absorbing drapes (37 µSv ± 35; range, 4-183 µSv) compared with standard draping (84 µSv ± 58; range, 11-220 µSv). Radiation incident on the patient was similar between groups, with no significant differences in mean KAP, total fluoroscopy time, and number of exposures acquired during the procedures. CONCLUSIONS: Use of disposable radiation-absorbing drapes reduces scatter radiation to interventionalists performing TJLB.

Radiation Dose Reduction during Uterine Fibroid Embolization Using an Optimized Imaging Platform.

PURPOSE: To assess radiation dose reduction during uterine fibroid embolization (UFE) using an optimized angiographic processing and acquisition platform. MATERIALS AND METHODS: Radiation dose data for 70 women (mean age, 46 y; range, 34-67 y) who underwent UFE were retrospectively analyzed. Twenty-one patients underwent UFE using the baseline fluoroscopic and angiographic image acquisition platform, and 49 underwent UFE after implementing an optimized imaging platform in otherwise identical angiography suites. Cumulative kerma-area product (CKAP), cumulative air kerma (CAK), total fluoroscopy time, and image exposure number were collected for each procedure. Image quality was assessed by 3 interventional radiologists blinded to the platform used for image acquisition and processing. RESULTS: Patients undergoing UFE using the new x-ray fluoroscopy platform had significantly lower CKAP and CAK indicators than patients for whom baseline settings were used. Mean CKAP decreased by 60% from 438.5 Gy · cm2 (range, 180.3-1,081.1 Gy · cm2) to 175.2 Gy · cm2 (range, 47.1-757.0 Gy · cm2; P < .0001). Mean CAK decreased by 45% from 2,034.2 mGy (range, 699.3-5,056.0 mGy) to 1,109.8 mGy (range, 256.6-4,513.6 mGy; P = .001). No degradation of image quality was identified through qualitative evaluation. CONCLUSIONS: Significant reduction in patient radiation dose indicators can be achieved with use of an optimized image acquisition and processing platform.

High-Throughput Classification of Radiographs Using Deep Convolutional Neural Networks.

The study aimed to determine if computer vision techniques rooted in deep learning can use a small set of radiographs to perform clinically relevant image classification with high fidelity. One thousand eight hundred eighty-five chest radiographs on 909 patients obtained between January 2013 and July 2015 at our institution were retrieved and anonymized. The source images were manually annotated as frontal or lateral and randomly divided into training, validation, and test sets. Training and validation sets were augmented to over 150,000 images using standard image manipulations. We then pre-trained a series of deep convolutional networks based on the open-source GoogLeNet with various transformations of the open-source ImageNet (non-radiology) images. These trained networks were then fine-tuned using the original and augmented radiology images. The model with highest validation accuracy was applied to our institutional test set and a publicly available set. Accuracy was assessed by using the Youden Index to set a binary cutoff for frontal or lateral classification. This retrospective study was IRB approved prior to initiation. A network pre-trained on 1.2 million greyscale ImageNet images and fine-tuned on augmented radiographs was chosen. The binary classification method correctly classified 100 % (95 % CI 99.73-100 %) of both our test set and the publicly available images. Classification was rapid, at 38 images per second. A deep convolutional neural network created using non-radiological images, and an augmented set of radiographs is effective in highly accurate classification of chest radiograph view type and is a feasible, rapid method for high-throughput annotation.

Implications of discordant findings between hepatic angiography and cross-sectional imaging in transplant candidates with hepatocellular carcinoma.

The goal of this study was to determine whether the detection of discordant numbers of hypervascular foci at hepatic angiography versus contrast-enhanced (CE) cross-sectional imaging [computed tomography (CT) or magnetic resonance imaging (MRI)] is associated with adverse clinical outcomes in patients with hepatocellular carcinoma (HCC) who are listed for liver transplantation. We retrospectively reviewed the records of 218 consecutive patients with HCC who were listed for a liver transplant and who underwent transarterial chemoembolization at our institution between January 1, 2006 and December 31, 2010. Patients were grouped into 3 categories: (1) the number of nodules at CT/MRI was concordant with the number of hypervascular foci detected at angiography (n=136), (2) the number of nodules at CT/MRI was greater than the number of hypervascular foci at angiography (n=45), and (3) the number of nodules at CT/MRI was fewer than the number of hypervascular foci at angiography (n=37). The study outcomes were liver transplantation and tumor recurrence after transplantation. The detection of at least 3 more hypervascular foci at angiography versus the number of HCC nodules on CT/MRI was associated with a significantly lower rate of transplantation [multivariate subhazard ratio (SHR), 0.39; 95% confidence interval (CI), 0.17-0.92]. The detection of fewer hypervascular foci at angiography versus the number of HCC nodules on CT/MRI was associated with a significantly higher rate of tumor recurrence after transplantation (multivariate SHR, 3.49; 95% CI, 1.27-9.56). In conclusion, liver transplant candidates with HCC who demonstrate discordant findings between angiography and CE CT or MRI may be at a higher risk for dropout from the transplant list and for tumor recurrence after transplantation.

Imaging Predictors of Elevated Lung Shunt Fraction in Patients Being Considered for Yttrium-90 Radioembolization.

PURPOSE: To identify imaging findings associated with elevated lung shunt fraction (LSF) in patients being considered for yttrium-90 ((90)Y) radioembolization. MATERIALS AND METHODS: During the period 2009-2014, 152 consecutive patients underwent planning hepatic arteriography with technetium-99m ((99m)Tc) macroaggregated albumin (MAA) injection. Computed tomography (CT) or magnetic resonance imaging performed before the procedure for each patient was assessed for hepatic vein (HV) tumor thrombus or occlusion from external compression by tumor. When imaging was a multiphase CT scan (117 patients), the arterial phase was evaluated for evidence of early HV opacification (relative to unaffected HVs), indicating hepatic venous shunting. These factors were correlated with LSF determined by (99m)Tc-MAA imaging. RESULTS: Median LSF was 6.7% (range, < 0.1%-71%), significantly higher for HCC (8.0% vs 6.3% for other tumors, P = .048). Larger tumor size was associated with higher LSF in univariate analysis (P = .001). There was high interobserver agreement for determining hepatic venous shunting (97%, κ = 0.847), which was associated with higher LSF (P < .001; 78% sensitivity, 93% specificity). Of 5 cases of HV tumor thrombus, all had high (> 20%) LSF (P < .001). HV occlusion was also associated with higher LSF (P = .039). Multivariate analysis confirmed that early HV opacification and either HV tumor thrombus or occlusion were associated with higher LSF. CONCLUSIONS: Early HV opacification and HV tumor thrombus or occlusion on cross-sectional imaging performed before radioembolization are associated with elevated LSF, which may contraindicate or limit the dose delivered in (90)Y radioembolization. This information could be helpful during patient counseling for anticipating the most appropriate mode of liver-directed therapy.

Patient Radiation Dose Reduction during Transarterial Chemoembolization Using a Novel X-Ray Imaging Platform.

PURPOSE: To evaluate radiation dose reduction in patients undergoing transarterial chemoembolization with the use of a new image acquisition and processing platform. MATERIALS AND METHODS: Radiation-dose data were obtained from 176 consecutive chemoembolization procedures in 135 patients performed in a single angiography suite. From January 2013 through October 2013, 85 procedures were performed by using our institution's standard fluoroscopic settings. After upgrading the x-ray fluoroscopy system with an image acquisition and processing platform designed to reduce image noise and reduce skin entrance dose, 91 chemoembolization procedures were performed from November 2013 through December 2014. Cumulative dose-area product (CDAP), cumulative air kerma (CAK), and total fluoroscopy time were recorded for each procedure. Image quality was assessed by three interventional radiologists blinded to the x-ray acquisition platform used. RESULTS: Patient radiation dose indicators were significantly lower for chemoembolization procedures performed with the novel imaging platform. Mean CDAP decreased from 3,033.2 dGy·cm(2) (range, 600.3-9,404.1 dGy·cm(2)) to 1,640.1 dGy·cm(2) (range, 278.6-6,779.9 dGy·cm(2); 45.9% reduction; P < .00001). Mean CAK decreased from 1,445.4 mGy (range, 303.6-5,233.7 mGy) to 971.7 mGy (range, 144.2-3,512.0 mGy; 32.8% reduction; P < .0001). A 20.3% increase in mean total fluoroscopy time was noted after upgrading the imaging platform, but blinded analysis of the image quality revealed no significant degradation. CONCLUSIONS: Although a small increase in fluoroscopy time was observed, a significant reduction in patient radiation dose was achieved by using the optimized imaging platform, without image quality degradation.

Periprocedural Patient Care.

Periprocedural care of patients who undergo image-guided interventions is a task of monumental importance. As physicians who perform procedures, radiologists rely on their noninterpretive skills to optimize patient care. At the center of periprocedural care is proper patient identification. It is imperative to perform the indicated procedure for the correct patient. It is also of great importance to discuss with the patient the nature of the procedure. This conversation should include the indications, risks, benefits, alternatives, and potential complications of the procedure. Once the patient agrees to the procedure and grants informed consent, it is imperative to stop and confirm that the correct procedure is being performed on the correct patient. This universal time-out policy helps decrease errors and improves patient care. To optimize our interpretative and procedural skills, it may be necessary to provide the patient with sedation or anesthesia. However, it is important to understand the continuum of sedation and be able to appropriately monitor the patient and manage the sedation in these patients. To minimize the risks of infection, periprocedural care of patients relies on aseptic or, at times, sterile techniques. Before the procedure, it is important to evaluate the patient's coagulation parameters and bleeding risks and correct the coagulopathy, if needed. During the procedure, the patient's blood pressure and at times the patient's glucose levels will also require monitoring and management. After the procedure, patients must be observed in a recovery unit and deemed safe for discharge. The fundamental components of periprocedural care necessary to enhance patient safety, satisfaction, and care are reviewed to familiarize the reader with the important noninterpretive skills necessary to optimize periprocedural care.

The inferior emissary vein: a reliable landmark for right adrenal vein sampling.

BACKGROUND: Right adrenal vein (RAV) catheterization can be a very challenging step in adrenal venous sampling (AVS). Visualization of the inferior emissary vein (IEV) may be an indication of successful RAV catheterization. PURPOSE: To compare the rate of successful RAV sampling in the presence of the IEV. MATERIAL AND METHODS: Retrospective review of all consecutive patients with PA who underwent AVS between April 2009 and April 2012 was performed. A total of 30 patients were identified. Procedural images, cortisol, and aldosterone values obtained from sampling of the RAV and inferior vena cava (IVC) were reviewed. Cortisol measurements obtained from RAV samples were divided by measurements from the infra-renal IVC blood samples in order to calculate the selectivity index (SI). An SI >3 was considered indicative of technically successful RAV sampling. RESULTS: RAV sampling was considered technically successful in 29 out of 30 cases (97%). In cases of successful RAV sampling (29 patients), the IEV was identified in 25 patients (86%). The IEV was visualized in isolation in 16 patients (64%), and in conjunction with visualization of the RAV or right adrenal gland stain in nine patients (36%). The IEV was not visualized in the one case of unsuccessful RAV sampling. Visualizing the IEV had a sensitivity of 86.2% for successful RAV sampling. CONCLUSION: The IEV may serve as a reliable landmark for the RAV during RAV sampling.

Caval penetration by retrievable inferior vena cava filters: a retrospective comparison of Option and Günther Tulip filters.

PURPOSE: To compare the frequency of vena caval penetration by the struts of the Option and Günther Tulip cone filters on postplacement computed tomography (CT) imaging. MATERIALS AND METHODS: All patients who had an Option or Günther Tulip inferior vena cava (IVC) filter placed between January 2010 and May 2012 were identified retrospectively from medical records. Of the 208 IVC filters placed, the positions of 58 devices (21 Option filters, 37 Günther Tulip filters [GTFs]) were documented on follow-up CT examinations obtained for reasons unrelated to filter placement. In cases when multiple CT studies were obtained after placement, each study was reviewed, for a total of 80 examinations. Images were assessed for evidence of caval wall penetration by filter components, noting the number of penetrating struts and any effect on pericaval tissues. RESULTS: Penetration of at least one strut was observed in 17% of all filters imaged by CT between 1 and 447 days following placement. Although there was no significant difference in the overall prevalence of penetration when comparing the Option filter and GTF (Option, 10%; GTF, 22%), only GTFs showed time-dependent penetration, with penetration becoming more likely after prolonged indwelling times. No patient had damage to pericaval tissues or documented symptoms attributed to penetration. CONCLUSIONS: Although the Günther Tulip and Option filters exhibit caval penetration at CT imaging, only the GTF exhibits progressive penetration over time.

Clinical outcomes of percutaneous drainage of breast fluid collections after mastectomy with expander-based breast reconstruction.

PURPOSE: To determine clinical outcomes of patients who underwent imaging-guided percutaneous drainage of breast fluid collections after mastectomy and breast reconstruction. MATERIALS AND METHODS: A retrospective review was performed including all consecutive patients who underwent percutaneous drainage of fluid collections after mastectomy with tissue expander-based reconstruction between January 2007 and September 2012. During this period, 879 mastectomies (563 patients) with expander-based breast reconstruction were performed. Fluid collections developed in 28 patients (5%), which led to 30 imaging-guided percutaneous drainage procedures. The median follow-up time was 533 days. Patient characteristics, surgical technique, microbiology analysis, and clinical outcomes were reviewed. RESULTS: The mean age of patients was 51.5 years (range, 30.9-69.4 y), and the median time between breast reconstruction and drainage was 35 days (range, 4-235 d). Erythema and swelling were the most common presenting symptoms. The median volume of fluid evacuated at the time of drain placement was 70 mL. Drains were left in place for a median 14 days (range, 6-34 d). Microorganisms were detected in the fluid in 12 of 30 drainage procedures, with Staphylococcus aureus being the most common microorganism. No further intervention was needed in 21 of 30 drainage procedures (70%). However, surgical intervention (removal of expanders) was needed after 6 (20%) drainage procedures, and additional percutaneous drainage procedures were performed after 3 (10%) drainage procedures. CONCLUSIONS: Percutaneous drainage is an effective means of treating postoperative fluid collections after expander-based breast reconstruction and can obviate the need for repeat surgery in most cases.

Creation of a percutaneous mesocaval shunt to control variceal bleeding in a child.

Although transjugular intrahepatic portosystemic shunt (TIPS) placement is the standard procedure for the treatment of portal hypertension, it is often impossible to perform in patients with extrahepatic portal vein occlusion. In these patients, options for decompressing the liver are few. In this report, we present a novel solution for managing gastro-esophageal hemorrhage in a child with portal hypertension and extrahepatic portal vein occlusion, through the creation of a percutaneous mesocaval shunt.

Automated detection of IVC filters on radiographs with deep convolutional neural networks.

PURPOSE: To create an algorithm able to accurately detect IVC filters on radiographs without human assistance, capable of being used to screen radiographs to identify patients needing IVC filter retrieval. METHODS: A primary dataset of 5225 images, 30% of which included IVC filters, was assembled and annotated. 85% of the data was used to train a Cascade R-CNN (Region Based Convolutional Neural Network) object detection network incorporating a pre-trained ResNet-50 backbone. The remaining 15% of the data, independently annotated by three radiologists, was used as a test set to assess performance. The algorithm was also assessed on an independently constructed 1424-image dataset, drawn from a different institution than the primary dataset. RESULTS: On the primary test set, the algorithm achieved a sensitivity of 96.2% (95% CI 92.7-98.1%) and a specificity of 98.9% (95% CI 97.4-99.5%). Results were similar on the external test set: sensitivity 97.9% (95% CI 96.2-98.9%), specificity 99.6 (95% CI 98.9-99.9%). CONCLUSION: Fully automated detection of IVC filters on radiographs with high sensitivity and excellent specificity required for an automated screening system can be achieved using object detection neural networks. Further work will develop a system for identifying patients for IVC filter retrieval based on this algorithm.