Systematic Analysis of Common Factors Impacting Deep Learning Model Generalizability in Liver Segmentation.

Purpose: To investigate the effect of training data type on generalizability of deep learning liver segmentation models. Materials and Methods: This Health Insurance Portability and Accountability Act-compliant retrospective study included 860 MRI and CT abdominal scans obtained between February 2013 and March 2018 and 210 volumes from public datasets. Five single-source models were trained on 100 scans each of T1-weighted fat-suppressed portal venous (dynportal), T1-weighted fat-suppressed precontrast (dynpre), proton density opposed-phase (opposed), single-shot fast spin-echo (ssfse), and T1-weighted non-fat-suppressed (t1nfs) sequence types. A sixth multisource (DeepAll) model was trained on 100 scans consisting of 20 randomly selected scans from each of the five source domains. All models were tested against 18 target domains from unseen vendors, MRI types, and modality (CT). The Dice-Sørensen coefficient (DSC) was used to quantify similarity between manual and model segmentations. Results: Single-source model performance did not degrade significantly against unseen vendor data. Models trained on T1-weighted dynamic data generally performed well on other T1-weighted dynamic data (DSC = 0.848 ± 0.183 [SD]). The opposed model generalized moderately well to all unseen MRI types (DSC = 0.703 ± 0.229). The ssfse model failed to generalize well to any other MRI type (DSC = 0.089 ± 0.153). Dynamic and opposed models generalized moderately well to CT data (DSC = 0.744 ± 0.206), whereas other single-source models performed poorly (DSC = 0.181 ± 0.192). The DeepAll model generalized well across vendor, modality, and MRI type and against externally sourced data. Conclusion: Domain shift in liver segmentation appears to be tied to variations in soft-tissue contrast and can be effectively bridged with diversification of soft-tissue representation in training data.Keywords: Convolutional Neural Network (CNN), Deep Learning Algorithms, Machine Learning Algorithms, Supervised Learning, CT, MRI, Liver Segmentation Supplemental material is available for this article. © RSNA, 2023.

Use of Skeletal Muscle Index as a Predictor of Wait-List Mortality in Patients With End-Stage Liver Disease.

The aim of this study is to validate a proposed definition of sarcopenia in predicting wait-list mortality. We retrospectively evaluated 355 adults (age ≥18 years) with cirrhosis listed for first-time LT from January 1, 2010, to April 1, 2018 from our center. Demographic, laboratory, and outcome data were collected in conjunction with computed tomography scans performed within 3 months of listing. Using imaging analysis software, the skeletal muscle index (SMI), which is a marker for sarcopenia-related mortality, was calculated. A survival analysis was performed to evaluate the association of the proposed sarcopenia definition of SMI <50 cm2 /m2 for men or <39 cm2 /m2 for women with wait-list mortality or delisting. Median SMI was 54.1 cm2 /m2 (range, 47-60 cm2 /m2 ). A total of 61 (17.2%) patients exhibited sarcopenia according to the proposed threshold, and 24.6% (57/232) of men were sarcopenic compared with 3.3% (4/123) of women (P < 0.001). Mean (standard deviation [SD]) SMI was also higher for men (56.6 ± 9.6 cm2 /m2 ) than for women (50.7 ± 8.0 cm2 /m2 ; P < 0.001). Median follow-up time among patients was 2.1 months (0-12 months), and 30 events were observed (hazard ratio, 0.98; 95% confidence interval, 0.95-1.02; P = 0.41). There was no statistically significant difference in time on the waiting list between patients with and without sarcopenia (P = 0.89) as defined at the threshold. Using the prespecified definitions of sarcopenia based on SMI, there was no statistically significant difference in mortality and delisting from the transplant waiting list between patients with and without sarcopenia in this population. Practice and region-specific patterns for pretransplant selection and median Model for End-Stage Liver Disease at transplant may affect SMI as a predictor of wait-list mortality.

DIfferential Subsampling With Cartesian Ordering With Respiratory Triggering Versus Conventional Liver Acquisition With Volume Acquisition: A Multiple Reader Preference Study.

OBJECTIVE: The aim of this study was to compare respiratory-triggered DIfferential Subsampling with Cartesian Ordering (rtDISCO) and breath-held Liver Acquisition with Volume Acquisition (LAVA) image quality. METHODS: In this institutional review board-approved, Health Insurance Portability and Accountability Act-compliant prospective study, 25 subjects underwent T1 imaging with rtDISCO and LAVA before and after intravenous contrast. Three readers scored individual series and side-by-side comparisons for motion and noise. Eight clinical tasks were qualitatively assessed. RESULTS: As individual series, readers rated rtDISCO images as more degraded by motion on both precontrast (mean rtDISCO score, 2.7; LAVA, 1.6; P < 0.001) and postcontrast images (rtDISCO, 2.4; LAVA, 1.8; P < 0.001). Readers preferred LAVA images based on motion on both precontrast (mean preference, -1.2; P < 0.001) and postcontrast images (mean preference, -0.7; P < 0.001) on side-by-side assessment. There was no preference between sequences for 6 of 8 clinical tasks on postcontrast images. CONCLUSIONS: Readers preferred LAVA with respect to motion but not noise; there was no preference in most of the tested clinical tasks.

LI-RADS Treatment Response Algorithm: Performance and Diagnostic Accuracy.

Background In 2017, the Liver Imaging Reporting and Data System (LI-RADS) included an algorithm for the assessment of hepatocellular carcinoma (HCC) treated with local-regional therapy. The aim of the algorithm was to enable standardized evaluation of treatment response to guide subsequent therapy. However, the performance of the algorithm has not yet been validated in the literature. Purpose To evaluate the performance of the LI-RADS 2017 Treatment Response algorithm for assessing the histopathologic viability of HCC treated with bland arterial embolization. Materials and Methods This retrospective study included patients who underwent bland arterial embolization for HCC between 2006 and 2016 and subsequent liver transplantation. Three radiologists independently assessed all treated lesions by using the CT/MRI LI-RADS 2017 Treatment Response algorithm. Radiology and posttransplant histopathology reports were then compared. Lesions were categorized on the basis of explant pathologic findings as either completely (100%) or incompletely (<100%) necrotic, and performance characteristics and predictive values for the LI-RADS Treatment Response (LR-TR) Viable and Nonviable categories were calculated for each reader. Interreader association was calculated by using the Fleiss κ. Results A total of 45 adults (mean age, 57.1 years ± 8.2; 13 women) with 63 total lesions were included. For predicting incomplete histopathologic tumor necrosis, the accuracy of the LR-TR Viable category for the three readers was 60%-65%, and the positive predictive value was 86%-96%. For predicting complete histopathologic tumor necrosis, the accuracy of the LR-TR Nonviable category was 67%-71%, and the negative predictive value was 81%-87%. By consensus, 17 (27%) of 63 lesions were categorized as LR-TR Equivocal, and 12 of these lesions were incompletely necrotic. Interreader association for the LR-TR category was moderate (κ = 0.55; 95% confidence interval: 0.47, 0.67). Conclusion The Liver Imaging Reporting and Data System 2017 Treatment Response algorithm had high predictive value and moderate interreader association for the histopathologic viability of hepatocellular carcinoma treated with bland arterial embolization when lesions were assessed as Viable or Nonviable. © RSNA, 2019 Online supplemental material is available for this article. See also the editorial by Gervais in this issue.

Imaging-guided percutaneous thrombin injection for the treatment of iatrogenic femoral artery pseudoaneurysms.

PURPOSE: To evaluate factors that may affect successful ultrasound-guided percutaneous thrombin injection of iatrogenic femoral artery pseudoaneurysms (PSA). MATERIALS AND METHODS: This was an IRB-approved, HIPAA-compliant retrospective study of 326 consecutive subjects (138 males, 188 females; mean age 68 years, range 18-95) who underwent thrombin injection for treatment of femoral PSA; follow-up ultrasound was available in 145 subjects. The number of PSA lobes and dimensions, pre-procedure laboratory values (international normalized ratio [INR], activated partial thromboplastin time [aPTT], platelet count), and concomitant anticoagulation therapy were recorded. RESULTS: Technical success was achieved in 98.2% (320/326) of subjects. Primary effectiveness (complete thrombosis at 24 h) was achieved in 74.5% (108/145). Twenty-five subjects underwent repeat thrombin injection, successful in 21 subjects, for a total effectiveness rate of 97.0% (129/133). No imaging factor was associated with technique failure, including number of lobes (p = 0.898), largest dimension (p = 0.344), or volume (p = 0.697). No statistically significant difference in pre-procedure INR, aPTT, or platelet count was found between subjects with CT and those with IT (p > 0.138). Anticoagulation therapy was associated with incomplete thrombosis (35.5% [38/107] for CT vs. 63.9% [23/26] for IT; p = 0.002). CONCLUSION: Imaging-guided percutaneous thrombin injection has high technical success and effectiveness rates for the treatment of iatrogenic femoral artery PSA. Anticoagulation therapy was the only factor associated with incomplete thrombosis.

Accurate fatty acid composition estimation of adipose tissue in the abdomen based on bipolar multi-echo MRI.

PURPOSE: To develop a bipolar multi-echo MRI method for the accurate estimation of the adipose tissue fatty acid composition (FAC) using clinically relevant protocols at clinical field strength. METHODS: The proposed technique jointly estimates confounding factors (field map, R2* , eddy-current phase) and triglyceride saturation state parameters by fitting multi-echo gradient echo acquisitions to a complex signal model. The noise propagation behavior was improved by applying a low-rank enforcing denoising technique and by addressing eddy-current-induced phase discrepancies analytically. The impact of the total echo train duration on the FAC parameter map accuracy was analyzed in an oil phantom at 3T. Accuracy and reproducibility assessment was based on in vitro oil phantom measurements at two field strengths (3T and 1.5T) and with two different protocols. Repeatability was assessed in vivo in patients (n = 8) with suspected fatty liver disease using test-retest acquisitions in the abdominal subcutaneous, perirenal and mesenteric fat depots. RESULTS: Echo train readout durations of at least five times the conventional in-phase time were required for accurate FAC estimation in areas of high fat content. In vitro, linear regression and Bland-Altman analyses demonstrated strong (r > 0.94) and significant (P â‰ª 0.01) correlations between measured and reference FACs for all acquisitions, with smaller overall intercepts and biases at 3T compared to 1.5T. In vivo, reported mean absolute differences between test and retest were 1.54%, 3.31%, and 2.63% for the saturated, mono-unsaturated, and poly-unsaturated fat component, respectively. CONCLUSIONS: Accurate and reproducible MRI-based FAC quantification within a breath-hold is possible at clinical field strengths.

Pilot Evaluation of Angiogenesis Signaling Factor Response after Transcatheter Arterial Embolization for Hepatocellular Carcinoma.

Purpose To identify changes in a broad panel of circulating angiogenesis factors after bland transcatheter arterial embolization (TAE), a purely ischemic treatment for hepatocellular carcinoma (HCC). Materials and Methods This prospective HIPAA-compliant study was approved by the institutional review board. Informed written consent was obtained from all participants prior to entry into the study. Twenty-five patients (21 men; mean age, 61 years; range, 30-81 years) with Liver Imaging Reporting and Data System category 5 or biopsy-proven HCC and who were undergoing TAE were enrolled from October 15, 2014, through December 2, 2015. Nineteen plasma angiogenesis factors (angiopoietin 2; hepatocyte growth factor; platelet-derived growth factor AA and BB; placental growth factor; vascular endothelial growth factor A and D; vascular endothelial growth factor receptor 1, 2, and 3; osteopontin; transforming growth factor ß1 and ß2; thrombospondin 2; intercellular adhesion molecule 1; interleukin 6 [IL-6]; stromal cell-derived factor 1; tissue inhibitor of metalloproteinases 1; and vascular cell adhesion molecule 1 [VCAM-1]) were measured by using enzyme-linked immunosorbent assays at 1 day, 2 weeks, and 5 weeks after TAE and were compared with baseline levels by using paired Wilcoxon tests. Tumor response was assessed according to modified Response Evaluation Criteria in Solid Tumors (mRECIST). Angiogenesis factor levels were compared between responders and nonresponders by mRECIST criteria by using unpaired Wilcoxon tests. Results All procedures were technically successful with no complications. Fourteen angiogenesis factors showed statistically significant changes following TAE, but most changes were transient. IL-6 was upregulated only 1 day after the procedure, but showed the largest increases of any factor. Osteopontin and VCAM-1 demonstrated sustained upregulation at all time points following TAE. At 3-month follow-up imaging, 11 patients had responses to TAE (complete response, n = 6; partial response, n = 5) and 11 patients were nonresponders (stable disease, n = 9; progressive disease, n = 2). In nonresponders, the percent change in IL-6 on the day after TAE (P = .033) and the mean percent change in osteopontin after TAE (P = .024) were significantly greater compared with those of responders. Conclusion Multiple angiogenesis factors demonstrated significant upregulation after TAE. VCAM-1 and osteopontin demonstrated sustained upregulation, whereas the rest were transient. IL-6 and osteopontin correlated significantly with radiologic response after TAE. © RSNA, 2017.

Randomized Controlled Trial of Octyl Cyanoacrylate Skin Adhesive versus Subcuticular Suture for Skin Closure after Implantable Venous Port Placement.

PURPOSE: To compare early outcomes of skin closure with octyl cyanoacrylate skin adhesive versus subcuticular suture closure. MATERIALS AND METHODS: Over a 7-month period, 109 subjects (28 men and 81 women; mean age, 58.6 y) scheduled to undergo single-lumen implantable venous port insertion for chemotherapy were randomly assigned to skin closure with either octyl cyanoacrylate skin adhesive or absorbable subcuticular suture after suturing the deep dermal layer. Subjects were followed for episodes of infection or dehiscence within 3 months of port implantation. At 3 months, photographs of the healed incision were obtained and reviewed by a plastic surgeon in a blinded fashion who rated cosmetic scar appearance based on a validated 10-point cosmesis score. RESULTS: Of subjects, 54 were randomly assigned to skin adhesive, and 55 were randomly assigned to subcuticular suture. No subjects had incision dehiscence. Infection rates at 3 months were similar between groups (2.1% vs 4.0%; P = 1.0). The mean cosmesis scores were 4.40 for skin adhesive and 4.46 for subcuticular suture (P = .898). The superficial skin closure time was 8.6 minutes for suture versus 1.4 minutes for skin adhesive (P < .001). CONCLUSIONS: Scar cosmesis and patient outcomes did not significantly vary between skin adhesive versus subcuticular suture, although skin closure time was significantly less with skin adhesive.

Hepatic venous pressure measurements: comparison of end-hole and balloon catheter methods.

PURPOSE: To determine the difference in hepatic venous pressures measured with the use of an end-hole diagnostic catheter versus a balloon catheter. MATERIALS AND METHODS: A total of 92 patients underwent transjugular hepatic venous pressure measurements with a 5-F diagnostic end-hole catheter and a balloon catheter, with the catheter type used initially determined randomly. With both catheters, free and wedged systolic, diastolic, and mean pressures were collected. Hepatic venous pressure gradients were calculated from each pressure set. Eighty-five patients (92%) also underwent concurrent transjugular biopsy after pressures were recorded. Demographic, histologic, and specific procedural information were also collected. RESULTS: The study included 47 men and 45 women, with a mean age of 52.7 years (range, 19-84 y). For the entire population, there were statistically significant differences in mean measurements between the two catheters in wedged systolic (P = .004), diastolic (P = .021), and mean (P = .036) pressures. However, the differences between the means were only 0.783, 0.609, and 0.207 mm Hg, respectively. A subanalysis based on histologic stage revealed no difference between catheter types for normal or cirrhotic livers, but a significant (P = .017) difference in systolic wedged pressure (absolute difference of 0.67 mm Hg) in patients with mild to moderate fibrosis (stages 1-3). In all differences, the balloon catheter had the greater pressure reading. CONCLUSIONS: There was a significant difference in wedged pressure measurements between the two catheter systems in the overall population and among patients with a histologic grade indicating fibrosis. However, the absolute value differences between the two systems were comparatively small (< 1 mm Hg).