Academic Neuroradiologists Research Articles

The Influence of Radiologist Practice Setting on Identification of Vascular Compression from Magnetic Resonance Imaging in Trigeminal Neuralgia

Preoperative magnetic resonance imaging (MRI) studies are routinely ordered for trigeminal neuralgia (TN), though with contested reliability in contemporary literature. A potential reason for this disagreement is inconsistency in MRI reading methodologies. Here, we compare the rate of reported neurovascular compression on preoperative MRI by radiologists employed in community or private practice settings and academic neuroradiologists. A retrospective review was conducted on patients who underwent endoscopic microvascular decompression for TN with intraoperatively visualized neurovascular compression and primary read by a non-academic or community radiologist. Patient imaging was then re-read by a board-certified neuroradiologist practicing in an academic setting, who was blinded to the initial read and the side of TN symptoms. Non-academic radiologists reported vascular compression in 26.0% (20/77) of all patients, and mention was rarely made of the non-pathological side (sensitivity= 26.0%). On academic neuroradiologist re-reads, vascular compression was noted in 87.0% (67/77) of patients on the pathological side and in 57.1% (44/77) on the non-pathological side (sensitivity= 87.0%, specificity= 42.9%). Isotropic/near isotropic 3-dimensional steady state or heavily T2-weighted sequences were read with 92.3% sensitivity and 36.9% specificity, compared to 58.3% sensitivity and 66.7% specificity using routine T2 weighted sequences. The frequency of vascular compression reported by non-academic radiologists is much lower than what is reported by academic neuroradiologists reading the same MRI scans. These results highlight the effect of practice setting on the predictive power of neuroimaging. Future studies are indicated to further investigate these relationships, as well as to trial newer imaging modalities.

Brain MRI Deep Learning and Bayesian Inference System Augments Radiology Resident Performance.

Automated quantitative and probabilistic medical image analysis has the potential to improve the accuracy and efficiency of the radiology workflow. We sought to determine whether AI systems for brain MRI diagnosis could be used as a clinical decision support tool to augment radiologist performance. We utilized previously developed AI systems that combine convolutional neural networks and expert-derived Bayesian networks to distinguish among 50 diagnostic entities on multimodal brain MRIs. We tested whether these systems could augment radiologist performance through an interactive clinical decision support tool known as Adaptive Radiology Interpretation and Education System (ARIES) in 194 test cases. Four radiology residents and three academic neuroradiologists viewed half of the cases unassisted and half with the results of the AI system displayed on ARIES. Diagnostic accuracy of radiologists for top diagnosis (TDx) and top three differential diagnosis (T3DDx) was compared with and without ARIES. Radiology resident performance was significantly better with ARIES for both TDx (55% vs 30%; P < .001) and T3DDx (79% vs 52%; P = 0.002), with the largest improvement for rare diseases (39% increase for T3DDx; P < 0.001). There was no significant difference between attending performance with and without ARIES for TDx (72% vs 69%; P = 0.48) or T3DDx (86% vs 89%; P = 0.39). These findings suggest that a hybrid deep learning and Bayesian inference clinical decision support system has the potential to augment diagnostic accuracy of non-specialists to approach the level of subspecialists for a large array of diseases on brain MRI.

Subspecialty-Level Deep Gray Matter Differential Diagnoses with Deep Learning and Bayesian Networks on Clinical Brain MRI: A Pilot Study.

To develop and validate a system that could perform automated diagnosis of common and rare neurologic diseases involving deep gray matter on clinical brain MRI studies. In this retrospective study, multimodal brain MRI scans from 212 patients (mean age, 55 years ± 17 [standard deviation]; 113 women) with 35 neurologic diseases and normal brain MRI scans obtained between January 2008 and January 2018 were included (110 patients in the training set, 102 patients in the test set). MRI scans from 178 patients (mean age, 48 years ± 17; 106 women) were used to supplement training of the neural networks. Three-dimensional convolutional neural networks and atlas-based image processing were used for extraction of 11 imaging features. Expert-derived Bayesian networks incorporating domain knowledge were used for differential diagnosis generation. The performance of the artificial intelligence (AI) system was assessed by comparing diagnostic accuracy with that of radiologists of varying levels of specialization by using the generalized estimating equation with robust variance estimator for the top three differential diagnoses (T3DDx) and the correct top diagnosis (TDx), as well as with receiver operating characteristic analyses. In the held-out test set, the imaging pipeline detected 11 key features on brain MRI scans with 89% accuracy (sensitivity, 81%; specificity, 95%) relative to academic neuroradiologists. The Bayesian network, integrating imaging features with clinical information, had an accuracy of 85% for T3DDx and 64% for TDx, which was better than that of radiology residents (n = 4; 56% for T3DDx, 36% for TDx; P < .001 for both) and general radiologists (n = 2; 53% for T3DDx, 31% for TDx; P < .001 for both). The accuracy of the Bayesian network was better than that of neuroradiology fellows (n = 2) for T3DDx (72%; P = .003) but not for TDx (59%; P = .19) and was not different from that of academic neuroradiologists (n = 2; 84% T3DDx, 65% TDx; P > .09 for both). A hybrid AI system was developed that simultaneously provides a quantitative assessment of disease burden, explainable intermediate imaging features, and a probabilistic differential diagnosis that performed at the level of academic neuroradiologists. This type of approach has the potential to improve clinical decision making for common and rare diseases.Supplemental material is available for this article.© RSNA, 2020.

Artificial Intelligence System Approaching Neuroradiologist-level Differential Diagnosis Accuracy at Brain MRI.

Background Although artificial intelligence (AI) shows promise across many aspects of radiology, the use of AI to create differential diagnoses for rare and common diseases at brain MRI has not been demonstrated. Purpose To evaluate an AI system for generation of differential diagnoses at brain MRI compared with radiologists. Materials and Methods This retrospective study tested performance of an AI system for probabilistic diagnosis in patients with 19 common and rare diagnoses at brain MRI acquired between January 2008 and January 2018. The AI system combines data-driven and domain-expertise methodologies, including deep learning and Bayesian networks. First, lesions were detected by using deep learning. Then, 18 quantitative imaging features were extracted by using atlas-based coregistration and segmentation. Third, these image features were combined with five clinical features by using Bayesian inference to develop probability-ranked differential diagnoses. Quantitative feature extraction algorithms and conditional probabilities were fine-tuned on a training set of 86 patients (mean age, 49 years ± 16 [standard deviation]; 53 women). Accuracy was compared with radiology residents, general radiologists, neuroradiology fellows, and academic neuroradiologists by using accuracy of top one, top two, and top three differential diagnoses in 92 independent test set patients (mean age, 47 years ± 18; 52 women). Results For accuracy of top three differential diagnoses, the AI system (91% correct) performed similarly to academic neuroradiologists (86% correct; P = .20), and better than radiology residents (56%; P < .001), general radiologists (57%; P < .001), and neuroradiology fellows (77%; P = .003). The performance of the AI system was not affected by disease prevalence (93% accuracy for common vs 85% for rare diseases; P = .26). Radiologists were more accurate at diagnosing common versus rare diagnoses (78% vs 47% across all radiologists; P < .001). Conclusion An artificial intelligence system for brain MRI approached overall top one, top two, and top three differential diagnoses accuracy of neuroradiologists and exceeded that of less-specialized radiologists. © RSNA, 2020 Online supplemental material is available for this article. See also the editorial by Zaharchuk in this issue.

Magnetic resonance imaging of the lumbar spine in adult: Evaluation of spinal incidental findings in patients with low back pain

Aim: We aimed to incidental findings discovered in patients with suspected low back pain disease undergoing MRI of the lumbar spine and the clinical importance of those findings. Material and Methods: One thousand seven hundred fourteen patients included who were admitted to our hospital with a preliminary diagnosis of low back pain were examined. The duration of low back pain complaints of the patients was ranging from 1 week to 4 years. MRI examinations were interpreted by academic neuroradiologists who had 7 years of experience, at the time of the study. Results: Overall, 358 patients (20.9%) had incidental findings. Transitional vertebrae was the most common (134 cases, 7.8%) incidental finding, followed by vertebral hemangioma (65 cases, 3.8%) and Tarlov cyst (54 cases, 3.2%). The incidence of fatty filum terminale, compact bone island, spina bifida, synovial cyst and sacral meningocele was 2.1%, 1.5%, 1%, 0.8% and 0.7% respectively. Of these 1714 patients, 329 patients (92.06%) had one finding, 22 patients (5.95%) had two findings, and 7 patient (1.98%) had three findings. The conus medullaris was terminated in 1710 (99.7%) of the patients (L1-2) .This level were considered normal. In four cases, the spinal cord was determined to be below the required level. The prevalence of transitional vertebra was 7.8%. The prevalence of sacralization and lumbalization was 5.1% and 2.7%, respectively. Conclusion: In this study we showed that incidental findings are mostly benign but interestingly there were no malignant or potentially malignant incidental lesions.

Reporting of Critical Findings in Neuroradiology

The objective of our study was to assess compliance among academic neuroradiologists in reporting institutionally derived critical findings. We analyzed 3054 neuroradiology CT and MRI reports generated in 1 month. Reports were categorized by whether or not they contained a critical finding based on a previously established list. The reports were subcategorized by whether the reporting neuroradiologist flagged the report as containing a critical finding and whether the radiologist verbally communicated the critical finding to the referring clinician. Reports were divided into day or night categories and the frequency of critical findings for each time period was calculated. Of the 3054 reports included in this study, 301 (9.9%) had critical findings. Of those 301 reports, 233 (77.4%) were flagged and the referring clinician was called. Of the remaining 68 reports with critical findings, the reporting radiologist did not call the clinician about 35.3% of them (24/68). Of the 2753 reports without critical findings, 2658 (96.5%) were appropriately not flagged and the clinician was not called. However, radiologists called clinicians about 3.5% (95/2753) of the reports without critical findings and erroneously flagged 68.4% (65/95) of those reports as critical. A majority of the cases with critical findings were reported at night (55.1%) despite the fact that 67.2% of the studies occurred during the day. Compliance with reporting and communicating critical findings must be monitored. Calling clinicians to report noncritical findings may result in unnecessary interruptions in work flow for radiologists and referring health care providers.

Quality Control in Neuroradiology: Impact of Trainees on Discrepancy Rates

Prior studies have found a 2%-8% clinically significant error rate in radiology practice. We compared discrepancy rates of studies interpreted by subspecialty-trained neuroradiologists working with and without trainees. Subspecialty-trained neuroradiologists reviewed 2162 studies during 41 months. Discrepancies between the original and "second opinion" reports were scored: 1, no change; 2, clinically insignificant detection discrepancy; 3, clinically insignificant interpretation discrepancy; 4, clinically significant detection discrepancy; and 5, clinically significant interpretation discrepancy. Faculty alone versus faculty and trainee discrepancy rates were calculated. In 87.6% (1894/2162), there were no discrepancies with the original report. The neuroradiology division had a 1.8% (39/2162; 95% CI, 1.3%-2.5%) rate of clinically significant discrepancies. In cases reviewed solely by faculty neuroradiologists (16.2% = 350/2162 of the total), the rate of discrepancy was 1.7% (6/350). With fellows (1232/2162, 57.0% of total) and residents (580/2162, 26.8% of total), the rates of discrepancy were 1.6% (20/1232) and 2.2% (13/580), respectively. The odds of a discrepant result were 26% greater (OR = 1.26; 95% CI, 0.38-4.20) when reading with a resident and 8% less (OR = 0.92; 95% CI, 0.35-2.44) when reading with a fellow than when reading alone. There was a 1.8% rate of clinically significant detection or interpretation discrepancy among academic neuroradiologists. The difference in the discrepancy rates between faculty only (1.7%), fellows and faculty (1.6%), and residents and faculty (2.2%) was not statistically significant but showed a trend indicating that reading with a resident increased the odds of a discrepant result.

Quality Control in Neuroradiology: Discrepancies in Image Interpretation among Academic Neuroradiologists

Prior studies have found a 3%-6% clinically significant error rate in radiology practice. We set out to assess discrepancy rates between subspecialty-trained university-based neuroradiologists. Over 17 months, university neuroradiologists randomly reviewed 1000 studies and reports of previously read examinations of patients in whom follow-up studies were read. The discrepancies between the original and "second opinion" reports were scored according to a 5-point scale: 1, no change; 2, clinically insignificant detection discrepancy; 3, clinically insignificant interpretation discrepancy; 4, clinically significant detection discrepancy; and 5, clinically significant interpretation discrepancy. Of the 1000 studies, 876 (87.6%) showed agreements with the original report. The neuroradiology division had a 2.0% (20/1000; 95% CI, 1.1%-2.9%) rate of clinically significant discrepancies involving 8 CTs and 12 MR images. Discrepancies were classified as vascular (n = 7), neoplastic (n = 9), congenital (n = 2), and artifacts (n = 2). Individual neuroradiologist's scores ranged from 0% to 7.7% ± 2.3% (n = 18). Both CT and MR imaging studies had a discrepancy rate of 2.0%. Our quality assessment study could serve as initial data before intervention as part of a PQI project.

Variability in Study Withdrawal Rates Among Academic Neuroradiologists Participating in a Radiology Utilization Management Program

The aim of this study was to analyze variability in study withdrawal rates among academic neuroradiologists participating in a utilization management (UM) program. The activities of 5 academic neuroradiologists participating in a UM program were assessed. The project period spanned 42 months. Participating neuroradiologists received identical training and used identical rule sets to evaluate the appropriateness of outpatient CT and MRI studies. Neuroradiologists could approve studies on the basis of available electronic data or contact referring physicians for further information. The rate of withdrawals (defined as "study not performed by consensus" or "study changed by consensus" with the referring physician) was compared among the neuroradiologists. A total of 5,256 studies were reviewed (mean, 1,051; median, 1,043). There were 573 studies (10.9%) not performed by consensus and 298 studies (5.7%) changed by consensus, resulting in a total of 871 withdrawals (16.6%). Among the neuroradiologists, withdrawal rates varied by approximately a factor of 2 (mean, 16.5%; median, 15.2%; range, 12.8%-23.5%). Although rate of studies not performed by consensus varied by approximately a factor of 2 (mean, 10.8%; median, 8.1%; range, 7.6%-18.0%), the rate of studies changed by consensus showed little variation (mean, 5.7%; median, 5.5%; range, 4.8%-6.6%). Variability in withdrawal rates was related to cases in which referring physicians were asked not to perform studies; there was little variation when referring physicians were asked to change studies. These data imply agreement among the neuroradiologists regarding appropriateness of rule sets but suggest that personality characteristics such as persuasiveness, persistence, and desire for conflict avoidance may play an important role in outcomes.

The tail wagging the dog

The tail wagging the dog

Experience of an Academic Neuroradiology Division Participating in a Utilization Management Program

There has been a steady increase in the utilization of expensive outpatient imaging studies, with a resultant increase in health care costs. To reduce unnecessary and inappropriate studies, the authors' department has participated in a utilization management (UM) program; in this article, the authors report the experience of the division of neuroradiology. Using evidence-based guidelines, a commercial UM program provides real-time decision support for physicians ordering expensive outpatient imaging studies. After consultation between UM personnel and a referring physician's staff, studies not meeting appropriateness criteria are referred to an academic radiologist for review. The radiologist can approve the study after reviewing the electronic chart or call the referring physician for further information. Studies are not denied by radiologists. For health plans of one large payer, 2,032 neuroradiologic studies were reviewed by 7 neuroradiologists during a 1-year period, of which 1,622 (80%) were approved. The remaining 410 studies (20%) were initially withdrawn by consensus or by no callback. Overall, 293 of the 410 (71%) studies (208 magnetic resonance, 85 computed tomography) initially not performed were not reordered (duration of follow-up, 3-15 months). Hence, 293 of 2,032 (14%) of all studies reviewed were not performed. The estimated cost of scans not performed was about $150,000. Approximately 6% of requests were changed to more appropriate studies. The participation of academic neuroradiologists in a UM program affected the performance of many expensive outpatient imaging studies. There was a durable reduction in unnecessary and inappropriate studies. These results demonstrate the utility of evidence-based appropriateness criteria in radiology UM programs and the power of the sentinel effect.

The Journal Formerly Known as the American Journal of Neuroradiology

I was most interested in the commentary in the April issue of AJNR (“How American is the American Journal of Neuroradiology ?”) because it is hard not to notice the decreasing proportion of US articles. This commentary deserves careful reading by all academic neuroradiologists. The chart shown

American Society of Neuroradiology Research Survey 2001

Rationale and Objectives Research is a critical component of the mission of academic radiology, and success in research is necessary for the future of neuroradiology. Thus, the authors set out to establish a baseline of research activities of American Society of Neuroradiology (ASNR) members. Materials and Methods The authors surveyed 100 fellowship program directors. The survey was Web based, with recruitment from the ASNR Fellowship Database survey site and with e-mail and fax solicitations to the Web site. Questions focused on neuroradiologist and neuroradiology fellow involvement in research. Results Forty-eight of the 100 program directors (48%) responded. Several key findings emerged: (a) About one-third of fellowship programs require all fellows to do some research, with most fellows receiving less than 1 day per week of academic time; (b) just over half of the programs expect fellows to publish a paper; (c) about two-thirds of academic neuroradiologists get at least 1 academic day per week; (d) most academic neuroradiologists perform research, but most of this research is unfunded; and (e) about nine of 10 academic sections have at least one neuroradiologist with some extramural funding. Conclusion The relative lack of extramural funding among academic neuroradiologists is a reality that is probably multifactorial; however, there may be a direct relationship between amount of academic time free from clinical duties and successful competition for funding. The time (and, thus, financial) support of research-oriented fellows and faculty should be increased.

Detection of early CT signs of >1/3 middle cerebral artery infarctions : interrater reliability and sensitivity of CT interpretation by physicians involved in acute stroke care.

This study had 2 goals: (1) to assess interrater reliability of academic neuroradiologists when classifying acute infarction by CT scan as >1/3 middle cerebral artery (MCA) involvement, <1/3 MCA involvement, or no infarction and (2) to determine the sensitivity of physicians potentially involved in acute stroke treatment in detecting >1/3 MCA acute infarctions. Studies of tissue plasminogen activator show an association between early signs of major infarction and poor outcome. The American Academy of Neurology and the American Heart Association recommend avoiding thrombolysis if early signs of major infarction are present. We presented 25 scans (normals, acute infarctions, and old infarctions) to 3 academic neuroradiologists. A scoring sheet based on Alteplase Thrombolysis for Acute Noninterventional Therapy in Ischemic Stroke (ATLANTIS)/CT Summit criteria was used to determine >1/3 MCA territory involvement. Nine of the 25 scans were presented again to assess intrarater reliability. We recalculated results of our previous study in which physicians interpreted infarction scans, now designating the scans as >1/3 MCA, <1/3 MCA, or normal, as determined by the neuroradiologists. All 3 neuroradiologists agreed on no infarction, <1/3 MCA, and >1/3 MCA on 64% of the scans. Neuroradiologist test-retest agreement was 96% for >1/3 MCA territory. Overall sensitivity for emergency physicians, neurologists, and general radiologists for detecting the presence of infarction in scans rated as >1/3 MCA was 78%. Neuroradiologists can achieve moderate agreement in detecting >1/3 MCA infarction. The emergency physicians, neurologists, and general radiologists tested were reasonably skilled at detecting >1/3 MCA acute infarction. However, their performance did not reliably identify all patients who have early CT infarct signs that place them at increased risk for cerebral hemorrhage after thrombolytic therapy.