Explore the vast range of titles available.

We show that recent (mid-to-late 2024) commercial large language models (LLMs) are capable of good quality metadata extraction and annotation with very little work on the part of investigators for several exemplar real-world annotation tasks in the neuroimaging literature. We investigated the GPT-4o LLM from OpenAI which performed comparably with several groups of specially trained and supervised human annotators. The LLM achieves similar performance to humans, between 0.91 and 0.97 on zero-shot prompts without feedback to the LLM. Reviewing the disagreements between LLM and gold standard human annotations we note that actual LLM errors are comparable to human errors in most cases, and in many cases these disagreements are not errors. Based on the specific types of annotations we tested, with exceptionally reviewed gold-standard correct values, the LLM performance is usable for metadata annotation at scale. We encourage other research groups to develop and make available more specialized “micro-benchmarks,” like the ones we provide here, for testing both LLMs, and more complex agent systems annotation performance in real-world metadata annotation tasks.

LGE was observed in 80% of patients with ischemic cardiomyopathy and 26.3% of patients with nonischemic cardiomyopathy, but only in 3.6% of patients with repaired tetralogy of Fallot and none of patients with simple congenital heart disease (P<0.001). Noninvasive assessment of myocardial fibrosis may provide valuable insights into the pathophysiology of left ventricular diastolic function and therapeutic response.

BackgroundHospital readmissions are costly and reflect negatively on care delivered.ObjectiveTo have a better understanding of unplanned readmissions after carotid revascularization, which might help to prevent them.MethodsThe Nationwide Readmissions Database was used to determine rates and reasons for unplanned readmission following carotid endarterectomy (CEA) and carotid artery stenting (CAS). Trends were assessed by annual percent change, modified Poisson regression was used to estimate risk ratios (RR) for readmission, and propensity scores were used to match cohorts.ResultsAnalysis yielded 522 040 asymptomatic and 55 485 symptomatic admissions for carotid revascularization between 2010 and 2015. Higher 30-day readmission rates were noted after CAS versus CEA in both symptomatic (9.1% vs 7.7%, p<0.001) and asymptomatic (6.8% vs 5.7%, p<0.001) patients. Readmission rates trended lower over time, significantly so for 90-day readmissions in symptomatic patients undergoing CEA. The most common cause for 30-day readmission was stroke in both symptomatic (5.5%) and asymptomatic (3.9%) patients. Factors associated with a higher risk of readmission included age over 80; male gender; Medicaid health insurance; and increases in severity of illness, mortality risk, and comorbidity indices. Analysis of matched cohorts showed that CAS had higher readmission than CEA (RR=1.14 (95% CI 1.06 to 1.22); p<0.001) only in asymptomatic patients. Adverse events during initial admission which predicted 30-day readmission included acute renal failure and acute respiratory failure in asymptomatic patients; hematoma and cardiac events were additional predictive adverse events in symptomatic patients.ConclusionsReadmission is not uncommon after carotid revascularization, occurs more often after CAS, and is predicted by baseline factors and by preventable adverse events at initial admission.

Background: Spinal cavernous malformations usually affect the vertebral bodies and are seldom intradural. Here, we report a rare spinal intradural-extramedullary cavernous malformation associated with extensive superficial siderosis along the neuraxis in a patient with radicular complaints. Case Description: A 60-year-old male presented with subacute headaches, intermittent fever, and acute back and radicular leg pain for 1-2 weeks. Magnetic resonance imaging revealed an intradural-extramedullary lesion just below the conus medullaris (at the L2 level). There was associated subarachnoid hemorrhage in the lumbar cistern and superficial siderosis along the entire spinal neuraxis. Following surgical resection, the patient's symptoms resolved. Histopathology of the lesion was of a cavernous malformation. Conclusions: There are only 56 cases of spinal intradural-extramedullary cavernous malformations published in the literature; however, only 3 described superficial neuraxis siderosis as noted in this case. In the present case, slowly recurring hemorrhages of the lesion located at the conus likely contributed to the complete neuraxis superficial siderosis. Timely evaluation and treatment of these lesions is warranted to avoid further compressive and/or hemorrhagic complications.

Kidney transplantation is the treatment of choice for children with end-stage renal disease. In Iran, a kidney transplantation program was started in the Labfi Nejad Hospital, Tehran in 1985. From 1985 to 2003, 278 children (mean age 11.6 years, 59.7% males) received their first renal transplant. All transplants were donations from live donors (12.5% live-related donors); 30.8% of patients were preemptively transplanted. The overall 1-year patient survival rate was 92% and the 5-year survival rate 74%. The median graft survival time was 7.2 years. The rate of graft survival was 88.8% at 1 year, 77% at 3 years, 67% at 5 years, 50% at 7 years, and 43% at 10 years after transplantation. The survival rate of patients and transplants improved significantly with time (p<0.05). In patients transplanted before 1997, the 5-year graft survival was 50% and 82% in patients transplanted after 1997. At the same time intervals, the frequency of acute rejection episodes was 66.6 versus 40.8% and of chronic rejection 50.5 versus 28.7%. The outcome in children below the age of 6 years was poor. Graft survival was negatively correlated with the frequency and an early time point of acute rejection episodes. The modus of transplantation (preemptive or postdialysis) did not influence the results. In conclusion, patient and graft survival in transplanted children significantly improved with time, thus reflecting greater medical and surgical experience, new immunosuppressive drugs, and better compliance.

Left ventricular non-compaction cardiomyopathy is a rare congenital cardiomyopathy that affects both children and adults. Since the clinical manifestations are not sufficient to establish diagnosis, echocardiography is the diagnostic tool that makes it possible to document ventricular non-compaction and establish prognostic factors. We report a 47-year-old woman with a history of dilated cardiomyopathy with unknown etiology. Echocardiography showed mild left ventricular enlargement with severe systolic dysfunction (EF = 20–25%). According to cardiac magnetic resonance imaging findings non-compaction left ventricle with hypertrophic cardiomyopathy was considered, and right ventricular septal biopsy was recommended. Right ventricular endomyocardial biopsy showed moderate hypertrophy of cardiac myocytes with foci of myocytolysis and moderate interstitial fibrosis. No evidence of infiltrative deposition was seen.

We show that recent (mid-to-late 2024) commercial large language models (LLMs) are capable of good quality metadata extraction and annotation with very little work on the part of investigators for several exemplar real-world annotation tasks in the neuroimaging literature. We investigated the GPT-4o LLM from OpenAI which performed comparably with several groups of specially trained and supervised human annotators. The LLM achieves similar performance to humans, between 0.91 and 0.97 on zero-shot prompts without feedback to the LLM. Reviewing the disagreements between LLM and gold standard human annotations we note that actual LLM errors are comparable to human errors in most cases, and in many cases these disagreements are not errors. Based on the specific types of annotations we tested, with exceptionally reviewed gold-standard correct values, the LLM performance is usable for metadata annotation at scale. We encourage other research groups to develop and make available more specialized \"micro-benchmarks,\" like the ones we provide here, for testing both LLMs, and more complex agent systems annotation performance in real-world metadata annotation tasks.

Information technology systems within health care, such as picture archiving and communication system (PACS) in radiology, can have a positive impact on production but can also risk compromising quality. The widespread use of PACS has removed the previous feedback loop between radiologists and technologists. Instead of direct communication of quality discrepancies found for an examination, the radiologist submitted a paper-based quality-control report. A web-based issue-reporting tool can help restore some of the feedback loop and also provide possibilities for more detailed analysis of submitted errors. The purpose of this study was to evaluate the hypothesis that data from use of an online error reporting software for quality control can focus our efforts within our department. For the 372,258 radiologic examinations conducted during the 6-month period study, 930 errors (390 exam protocol, 390 exam validation, and 150 exam technique) were submitted, corresponding to an error rate of 0.25 %. Within the category exam protocol, technologist documentation had the highest number of submitted errors in ultrasonography (77 errors [44 %]), while imaging protocol errors were the highest subtype error for computed tomography modality (35 errors [18 %]). Positioning and incorrect accession had the highest errors in the exam technique and exam validation error category, respectively, for nearly all of the modalities. An error rate less than 1 % could signify a system with a very high quality; however, a more likely explanation is that not all errors were detected or reported. Furthermore, staff reception of the error reporting system could also affect the reporting rate.