Opportunistic Screening: Radiology Scientific Expert Panel.

Radiologic tests often contain rich imaging data not relevant to the clinical indication. Opportunistic screening refers to the practice of systematically leveraging these incidental imaging findings. Although opportunistic screening can apply to imaging modalities such as conventional radiography, US, and MRI, most attention to date has focused on body CT by using artificial intelligence (AI)-assisted methods. Body CT represents an ideal high-volume modality whereby a quantitative assessment of tissue composition (eg, bone, muscle, fat, and vascular calcium) can provide valuable risk stratification and help detect unsuspected presymptomatic disease. The emergence of "explainable" AI algorithms that fully automate these measurements could eventually lead to their routine clinical use. Potential barriers to widespread implementation of opportunistic CT screening include the need for buy-in from radiologists, referring providers, and patients. Standardization of acquiring and reporting measures is needed, in addition to expanded normative data according to age, sex, and race and ethnicity. Regulatory and reimbursement hurdles are not insurmountable but pose substantial challenges to commercialization and clinical use. Through demonstration of improved population health outcomes and cost-effectiveness, these opportunistic CT-based measures should be attractive to both payers and health care systems as value-based reimbursement models mature. If highly successful, opportunistic screening could eventually justify a practice of standalone "intended" CT screening.

FDA-regulated AI Algorithms: Trends, Strengths, and Gaps of Validation Studies.

RATIONALE AND OBJECTIVES: To assess key trends, strengths, and gaps in validation studies of the Food and Drug Administration (FDA)-regulated imaging-based artificial intelligence/machine learning (AI/ML) algorithms. MATERIALS AND METHODS: We audited publicly available details of regulated AI/ML algorithms in imaging from 2008 until April 2021. We reviewed 127 regulated software (118 AI/ML) to classify information related to their parent company, subspecialty, body area and specific anatomy type, imaging modality, date of FDA clearance, indications for use, target pathology (such as trauma) and findings (such as fracture), technique (CAD triage, CAD detection and/or characterization, CAD acquisition or improvement, and image processing/quantification), product performance, presence, type, strength and availability of clinical validation data. Pertaining to validation data, where available, we recorded the number of patients or studies included, sensitivity, specificity, accuracy, and/or receiver operating characteristic area under the curve, along with information on ground-truthing of use-cases. Data were analyzed with pivot tables and charts for descriptive statistics and trends. RESULTS: We noted an increasing number of FDA-regulated AI/ML from 2008 to 2021. Seventeen (17/118) regulated AI/ML algorithms posted no validation claims or data. Just 9/118 reviewed AI/ML algorithms had a validation dataset sizes of over 1000 patients. The most common type of AI/ML included image processing/quantification (IPQ; n = 59/118), and triage (CADt; n = 27/118). Brain, breast, and lungs dominated the targeted body regions of interest. CONCLUSION: Insufficient public information on validation datasets in several FDA-regulated AI/ML algorithms makes it difficult to justify clinical applications since their generalizability and presence of bias cannot be inferred.

Evaluation and Real-World Performance Monitoring of Artificial Intelligence Models in Clinical Practice: Try It, Buy It, Check It.

The pace of regulatory clearance of artificial intelligence (AI) algorithms for radiology continues to accelerate, and numerous algorithms are becoming available for use in clinical practice. End users of AI in radiology should be aware that AI algorithms may not work as expected when used beyond the institutions in which they were trained, and model performance may degrade over time. In this article, we discuss why regulatory clearance alone may not be enough to ensure AI will be safe and effective in all radiological practices and review strategies available resources for evaluating before clinical use and monitoring performance of AI models to ensure efficacy and patient safety.

2020 ACR Data Science Institute Artificial Intelligence Survey.

PURPOSE: The ACR Data Science Institute conducted its first annual survey of ACR members to understand how radiologists are using artificial intelligence (AI) in clinical practice and to provide a baseline for monitoring trends in AI use over time. METHODS: The ACR Data Science Institute sent a brief electronic survey to all ACR members via email. Invitees were asked for demographic information about their practice and if and how they were currently using AI as part of their clinical work. They were also asked to evaluate the performance of AI models in their practices and to assess future needs. RESULTS: Approximately 30% of radiologists are currently using AI as part of their practice. Large practices were more likely to use AI than smaller ones, and of those using AI in clinical practice, most were using AI to enhance interpretation, most commonly detection of intracranial hemorrhage, pulmonary emboli, and mammographic abnormalities. Of practices not currently using AI, 20% plan to purchase AI tools in the next 1 to 5 years. CONCLUSION: The survey results indicate a modest penetrance of AI in clinical practice. Information from the survey will help researchers and industry develop AI tools that will enhance radiological practice and improve quality and efficiency in patient care.

Multi-Institutional Assessment and Crowdsourcing Evaluation of Deep Learning for Automated Classification of Breast Density.

OBJECTIVE: We developed deep learning algorithms to automatically assess BI-RADS breast density. METHODS: Using a large multi-institution patient cohort of 108,230 digital screening mammograms from the Digital Mammographic Imaging Screening Trial, we investigated the effect of data, model, and training parameters on overall model performance and provided crowdsourcing evaluation from the attendees of the ACR 2019 Annual Meeting. RESULTS: Our best-performing algorithm achieved good agreement with radiologists who were qualified interpreters of mammograms, with a four-class κ of 0.667. When training was performed with randomly sampled images from the data set versus sampling equal number of images from each density category, the model predictions were biased away from the low-prevalence categories such as extremely dense breasts. The net result was an increase in sensitivity and a decrease in specificity for predicting dense breasts for equal class compared with random sampling. We also found that the performance of the model degrades when we evaluate on digital mammography data formats that differ from the one that we trained on, emphasizing the importance of multi-institutional training sets. Lastly, we showed that crowdsourced annotations, including those from attendees who routinely read mammograms, had higher agreement with our algorithm than with the original interpreting radiologists. CONCLUSION: We demonstrated the possible parameters that can influence the performance of the model and how crowdsourcing can be used for evaluation. This study was performed in tandem with the development of the ACR AI-LAB, a platform for democratizing artificial intelligence.

Lesion detection performance of an abbreviated gadoxetic acid-enhanced MRI protocol for colorectal liver metastasis surveillance.

OBJECTIVE: To assess the lesion detection performance of an abbreviated MRI (AMRI-M) protocol consisting of ultrafast SE T2W, DWI, and T1W-HBP at 20 min for colorectal liver metastasis (CRLM) surveillance. METHODS: In this Institutional Review Board (IRB)-approved retrospective study, gadoxetic acid-enhanced MRI scans of 57 patients (43 with pathologically proven CRLMs) were assessed. Two readers independently evaluated two sets of images per patient and commented on the number, location, and size of liver lesions. Set 1 included ultrafast spin-echo (SE) T2-weighted (T2W) + T1-weighted (T1W) hepatobiliary phase (HBP) at 20 min sequences + diffusion-weighted imaging (DWI), and set 2 consisted of the standard MRI protocol. A maximum of 10 lesions per patient were recorded. Cohen's kappa analysis, sensitivity, areas under the curve (AUCs), and the MRI cost analysis of the AMRI-M protocol were assessed. RESULTS: Between 198 and 209 lesions were assessed with each set of images. The inter-observer agreement for the abbreviated protocol was reported excellent (κ = 0.91). The sensitivity and AUCs for the lesion characterization of AMRI-M protocol were very high (over 90%) for both readers. No statistically significant differences in sensitivity (assessed by mixed-effects logistic regression) and AUCs for lesion characterization (by ROC regression) were found between both protocols. The AMRI-M acquisition time was estimated to be less than 10 min, which translated into 59% cost of standard MRI. CONCLUSION: Our proposed AMRI-M protocol (ultrafast SE T2W, DWI, and T1W-HBP at 20 min) is fast, low-cost alternative to the standard MRI protocol and has a high lesion detection performance. KEY POINTS: • Gadoxetic acid-enhanced protocol has increased the accuracy, sensitivity, and specificity of MRI for detecting colorectal liver metastases. • Our proposed abbreviated MRI protocol is fast, low-cost alternative compared with the standard MRI protocol and has a high lesion detection performance. • Adoption of our protocol may translate to substantial savings for patients and payers.

10 Years of Nephrogenic Systemic Fibrosis: A Comprehensive Analysis of Nephrogenic Systemic Fibrosis Reports Received by a Pharmaceutical Company from 2006 to 2016.

OBJECTIVES: The aim of this study was to critically assess the evaluation and categorization process for nephrogenic systemic fibrosis (NSF) based on reports received by Bayer from 2006 to 2016. MATERIALS AND METHODS: A total of 779 NSF reports received by Bayer globally from 2006 to 2016 were included in the analysis. Arlington Medical Resources provided gadolinium-based contrast agent (GBCA) market share. Reports were conservatively categorized based on the Cowper/Girardi criteria. A statistical model simulated the impact of market share and market introduction on the number of unconfounded reports. RESULTS: For all reports, reported onset of disease ranged from 1996 and 2012. Of 779 reports, 325 involved a Bayer product only, 208 involved only products from other companies (or unknown GBCA), and 246 involved both Bayer and non-Bayer products. Most of all reports (86%) originated from the United States.Through 2006, Magnevist and Omniscan dominated the US market (>80% combined market share). All other GBCAs with fewer NSF reports comprised the remaining combined market share of less than 20% or were introduced after May 2007, after safety recommendations came into effect.A total of 563 reports (220 single-agent and 343 multiagent reports) involved Magnevist. In at least 150 of the 343 reports, a different GBCA (Omniscan, 118; OptiMARK, 15; MultiHance, 6; and macrocyclic agent, 11) showed the closest temporal relationship to onset of NSF-like symptoms.The simulation model demonstrated that patients receiving a GBCA with lower market share and late market introduction are less likely to be observed in an unconfounded setting. CONCLUSIONS: Year of market introduction, as well as US market share in 2000 to 2007, greatly influenced the absolute number of NSF reports for each GBCA, their a priori probability to cause NSF, as well as their a priori probability to be associated with unconfounded cases of NSF. Variability in case interpretation and pharmacovigilance approaches also influence the absolute number of unconfounded cases and should therefore not be used for comparative risk assessments. This should be primarily based on objective product parameters such as structure, stability, pharmacokinetics, and dose.

Efficacy and safety of gadobutrol-enhanced MRA of the renal arteries: Results from GRAMS (Gadobutrol-enhanced renal artery MRA study), a prospective, intraindividual multicenter phase 3 blinded study.

PURPOSE: To compare the performance of magnetic resonance angiography (MRA) with 1M gadobutrol, a high relaxivity macrocyclic contrast agent, to 2D time-of-flight MRA (ToF-MRA) using computed tomographic angiography (CTA) as the standard of reference. Primary objectives were evaluation for superiority of structural delineation and noninferiority for detection and exclusion of clinically significant disease. MATERIALS AND METHODS: In all, 315 subjects underwent unenhanced and contrast-enhanced MRA with 1M gadobutrol (CE-MRA) and were scanned with 1.5T MRI equipped with an at least 6-element body coil. Evaluations were based on both centralized blinded read (BR) performed by six readers as well as investigator site interpretations for the 292 subjects who completed the study. Quantitative evaluations including percent stenosis and normal vessel measurements were also performed. Secondary endpoints included identification of accessory renal arteries, diagnosis of fibromuscular dysplasia (FMD), diagnostic confidence, and need for additional imaging. RESULTS: A total of 292 patients suspected of renal artery disease completed the study. CE-MRA demonstrated statistically significant improvement in assessability of vascular segments compared to ToF: 95.9% vs. 77.6% (P < 0.0001). In the BR, the sensitivity and specificity of CE-MRA were noninferior to ToF-MRA (53.4% vs. 46.6% and 95.1% vs. 85.7%, respectively). There was less error in the CE-MRA stenosis measurements (0.15 mm gadobutrol vs. 0.41 mm ToF, P < 0.05). FMD was correctly diagnosed more frequently, 10% more accessory renal arteries were identified (P < 0.01), diagnostic confidence increased (P < 0.01), and fewer additional imaging studies were recommended (P < 0.01). CONCLUSION: Gadobutrol-enhanced MRA of the renal arteries has superior visualization, more accurate vessel measurements, and may serve as a CTA alternative without any ionizing radiation. LEVEL OF EVIDENCE: 1 Technical Efficacy: Stage 2 J. Magn. Reson. Imaging 2018;47:572-581.

Spleen volume on CT and the effect of abdominal trauma.

The aim of this study is to determine the magnitude of change in spleen volume on CT in subjects sustaining blunt abdominal trauma without hemorrhage relative to patients without disease and how the spleen volumes are distributed. Sixty-seven subjects with blunt abdominal trauma and 101 control subjects were included in this retrospective single-center, IRB-approved, and HIPAA-compliant study. Patients with an injured spleen were excluded. Using a semiautomatic segmentation program, two readers computed spleen volumes from CT. Spleen volume distribution in male and female trauma and control cohorts were compared nonparametrically. Spleen volume plotted against height, weight, and age were analyzed by linear regression. The number of females and males are, respectively, 35 and 32 in trauma subjects and 69 and 32 among controls. Female trauma patients (49.6 years) were older than males (39.8 years) (p = 0.02). Distributions of spleen volume were not normal, skewed above their means, requiring a nonparametric comparison. Spleen volumes in trauma patients were smaller than those in controls with medians of 230 vs 294 mL in males(p < 0.006) and 163 vs 191 mL in females(p < 0.04). Spleen volume correlated positively with weight in females and with height in male controls, and negatively with age in male controls (p < 0.01). Variation in reproducibility and repeatability was acceptable at 1.5 and 4.9 %, respectively. Reader variation was 1.7 and 4.6 % for readers 1 and 2, respectively. The mean spleen volume in controls was 245 mL, the largest ever reported. Spleen volume decreases in response to blunt abdominal trauma. Spleen volumes are not normally distributed. Our population has the largest spleen volume reported in the literature, perhaps a consequence of the obesity epidemic.

Distinguishing hemangiomas from metastases on liver MRI performed with gadoxetate disodium: Value of the extended washout sign.

OBJECTIVE: To describe the enhancement pattern of hemangiomas with gadoxetate disodium and propose a new sign - the "extended washout sign" - to diagnose hemangiomas on hepatobiliary MR imaging. MATERIALS AND METHODS: In this retrospective IRB approved study, quantitative and qualitative image analysis of 45 hemangiomas and 37 metastases in 77 patients was performed. Gadoxetate disodium-enhanced MR imaging was obtained during arterial and portal venous phases as well as with delays of 3, 8, and 20min. Lesion signal intensity was measured at each phase. Quantitatively, extended washout was defined as a 10% or greater decrease in signal intensity from 8 to 20min on 3D gradient echo images. Statistical analysis was performed using unpaired Student's t-test. Qualitative analysis was also performed to assess observer confidence based on T2-weighted images, dynamic images, and combined early (8min) and late (20min) hepatobiliary phases. Extended washout was defined as a perceptible change in signal from 8 to 20min. RESULTS: On quantitative analysis, 84% of hemangiomas demonstrated a positive extended washout sign while only 18% of metastases did. Qualitatively, 78% of hemangiomas demonstrated a perceptible change in signal from 8 to 20min, while only 5.4% of metastases did. When extended washout was used in combination with T2 hyperintensity, specificity increased to 100%, with a sensitivity of 87% and area under the curve of 0.99. CONCLUSIONS: When combined with T2 signal intensity, the extended washout sign can be used to increase accuracy of differentiating hemangiomas from metastases on gadoxetate disodium-enhanced MRI.

Management of Hepatocellular Adenoma: Recent Advances.

Hepatocellular adenoma (HCA) is a rare benign liver cell neoplasm that occurs more frequently in young women with a history of prolonged use of oral contraceptives. Surgical resection is considered because of the risk of hemorrhage in 25% and of malignant transformation in 5% of patients with HCA. HCA is a heterogeneous disease comprising 3 subtypes with distinct molecular and complication profiles. The inflammatory or telangiectatic subtype is at increased risk for hemorrhage, the ß-catenin-activated subtype is at increased risk for malignant transformation, and the hepatocyte nuclear factor-1α-inactivated or steatotic subtype is at the least risk for complications. One-third of the patients with HCA have multiple tumors on imaging with no increased risk of complications. Magnetic resonance imaging is the modality of choice for the diagnosis and subtype characterization of HCA. Systematic resection of HCA is recommended in male patients owing to the higher incidence of malignant transformation, and surgical excision in women should be reserved for tumors 5 cm or larger associated with an increased risk of complications. Cessation of hormonal therapy and radiologic surveillance in women with HCA tumors smaller than 5 cm shows that the vast majority of HCA remain stable or undergo spontaneous regression. Percutaneous core needle biopsy is of limited value because the therapeutic strategy is based primarily on patient sex and tumor size. Transarterial embolization is the initial treatment for HCA complicated by hemorrhage. Pregnancy should not be discouraged in the presence of HCA, however, frequent sonographic surveillance is recommended.

Inflammatory hepatocellular adenomas can mimic focal nodular hyperplasia on gadoxetic acid-enhanced MRI.

OBJECTIVE: Inflammatory hepatocellular adenoma (HCA) is a recently categorized entity of hepatocellular neoplasms. We investigated whether gadoxetic acid-enhanced MRI can distinguish inflammatory HCA from focal nodular hyperplasia (FNH). MATERIALS AND METHODS: From January 1, 2009, through January 1, 2013, gadoxetic acid-enhanced MRI examinations from two institutions were reviewed for HCA, with specific histologic features of inflammatory HCA. Biopsy and resection slides were reviewed, and immunohistochemistry for glutamine synthetase was performed in a subset to confirm the initial diagnosis. RESULTS: A total of 10 possible cases of inflammatory HCA were identified in the pathology database. On the basis of glutamine synthetase staining performed for this study, three cases were rediagnosed as FNH and thus were excluded from the study. Therefore, a total of seven patients with inflammatory HCA were identified. On gadoxetic acid-enhanced MRI, four of these patients had classic features of FNH (group A, FNH mimics), and three had imaging features suggestive of HCA (group B, typical inflammatory HCA). Imaging features that were considered diagnostic of FNH included isointense or minimal T2 hyperintensity, arterial enhancement, and diffuse hyperintensity on hepatobiliary phase. Three of the four patients with FNH mimics had slides available for pathologic rereview, and the diagnosis of inflammatory HCA was supported by glutamine synthetase immunohistochemistry findings. The pathology reports of the remaining four cases were rereviewed and were also found to have features consistent with inflammatory HCA. CONCLUSION: Inflammatory HCA can mimic FNH on MRI, including hepatobiliary phase hyperintensity. Moreover, conventional pathology using histopathology alone may lead to misclassification of inflammatory HCA.

US FDA draft Guidance Standard for Clinical Trial Imaging Endpoints: more than just imaging?

The current cost of developing a successful drug is typically over a billion dollars, with the registration trial(s) determining the success or failure of the entire development program. Often the primary endpoint of these trials is a subjective assessment. For registration trials with subjective endpoints, a regulatory agency may require a blinded independent central review (BICR) of the trial data. The BICR is a mechanism to reduce bias in open-labeled trials and to potentially increase accuracy and precision. A decision tree algorithm has been developed that can be used to determine when and what type of a BICR is needed. The US FDA draft Guidance Standard for Clinical Trial Imaging Endpoints can be used as an effective process map in exploring the value and use of BICRs in imaging, and in any hard to interpret variable subjective assessment in general.

Retail venue based screening mammography: assessment of women's preferences.

RATIONALE AND OBJECTIVES: The aim of this study was to explore women's interest and preferences in undergoing screening mammography in a retail health care setting. MATERIALS AND METHODS: Self-administered surveys were distributed to 400 mammography patients in May to June 2009. All of the women who were asked were eligible for screening (age >40 years, no abnormal mammographic findings in the recent past). Three hundred eighty-six screening-eligible women filled out and returned the self-administered survey. RESULTS: The average respondent age was 57 years. Three hundred ten of the patients (80.3%) had college or postgraduate educations. Two hundred three (52.6%) reported annual incomes >$60,000. Two hundred forty-one respondents (62.4%) had been undergoing screening mammography for >10 years, while this was the first examination for eight patients (2%). More than half of the patients (n = 215 [55.7%]) affirmed their interest in undergoing annual screening mammography in a private area within a retail shopping facility. Most preferred a pharmacy (77%) over Wal-Mart or a grocery store. Appealing factors about a retail setting were proximity to home (90%), free parking (62%), and operating hours (48.8%). CONCLUSIONS: There is interest among women in undergoing screening mammography at retail health care clinics, preferably pharmacies. The provision of services at a convenient location can increase adherence to guidelines for screening mammography.

Appearance of hepatocellular adenomas on gadoxetic acid-enhanced MRI.

OBJECTIVE: The purpose of this study was to evaluate enhancement characteristics of hepatocellular adenomas (HCAs) using gadoxetic acid as a hepatocyte-specific MR contrast agent. METHODS: Twenty-four patients with histopathologically proven HCAs were retrospectively identified. MRI consisted of T1- and T2-weighted (w) sequences with and without fat saturation (fs), multiphase dynamic T1-w images, and fs T1-w images during the hepatobiliary phase. Standard of reference was surgical resection (n = 19) or biopsy (n = 5). Images were analysed for morphology and contrast behaviour including signal intensity (SI) measurement on T1-w images normalised to the pre-contrast base line. RESULTS: In total 34 HCAs were evaluated. All HCAs showed enhancement in the arterial phase; 38 % of HCAs showed reduced contrast enhancement ("wash-out") in the venous phase. All HCAs showed enhancement (SI increase, 56 ± 53 %; P <0.001) in the hepatobiliary phase, although liver uptake was stronger (96 ± 58 %). Thus, 31 of all HCAs (91 %) appeared hypointense to the surrounding liver in the hepatobiliary phase, while 3 out of 34 lesions were iso-/hyperintense. CONCLUSIONS: Gadoxetic acid accumulates in HCAs in the hepatobiliary phase, although significantly less than in surrounding liver. Thus, HCA appears in the vast majority of cases as a hypointense lesion on hepatobiliary phase images. KEY POINTS: • Magnetic resonance-specific contrast agents are now available for hepatic imaging. • Hepatocellular adenomas enhance with gadoxetic acid as in previous CT/MRI experience. • Enhancement during the hepatobiliary phase is less in HCAs than in liver. • Typical HCAs appear as hypointense lesions on T1-w hepatobiliary phase images. • True hyperintense HCA enhancement can occasionally occur during the hepatobiliary phase.

Retrospective assessment of prevalence of nephrogenic systemic fibrosis (NSF) after implementation of a new guideline for the use of gadobenate dimeglumine as a sole contrast agent for magnetic resonance examination in renally impaired patients.

From May 2007 to January 2008, patients with Stage 3-5 chronic kidney disease (CKD) undergoing gadobenate dimeglumine (GBD)-enhanced magnetic resonance (MR) examinations were included in the retrospective investigation. The electronic medical records were reviewed to assess the prevalence of nephrogenic systemic fibrosis (NSF) in renally impaired patients underwent GBD-enhanced MR examinations. In all, 250 patients (98 men, mean age 72.6 years) were included: 97% of the patients had Stage 3 CKD (estimated GFR 30-59 mL/min/1.73 m(2)); 37% had been exclusively exposed to GBD. The remaining were exposed to GBD and other gadolinium-based contrast agents (GBCAs). The mean dose of GBD was 22 mL (standard deviation [SD], 11.2). Including exposure to other GBCAs, the mean cumulative dose of gadolinium was 61 mL (SD, 62.3). A total of 206 patients (82%) had skin examinations following the last GBD administration (mean duration, 108 days). No evidence of suspected or diagnosed NSF was found. In conclusion, on the basis of a retrospective chart review there was no skin evidence of NSF in predominantly Stage 3 CKD patients who were exposed to GBD at an average follow-up of 108 days, either solely or in combination with other GBCAs. J. Magn. Reson. Imaging 2009;30:1335-1340. (c) 2009 Wiley-Liss, Inc.