Journal of the American College of Radiology : JACR最新文献

Background: Annual low-dose CT (LDCT) is recommended after a negative lung cancer screening (LCS) CT. Patient follow-up is inconsistent, varying in both timing and imaging type (LDCT or routine chest CT). This study evaluates how different definitions of adherence to follow-up affect LCS adherence rates.

Study design and methods: This retrospective study (2016-2023) evaluated LCS follow-up within a New York health care system. Binary adherence was defined as follow-up chest CT acquisition and evaluated using six definitions of adherence. Variables included (1) three follow-up time frames (15, 24, and unlimited months) and (2) two follow-up CT types (LDCT only versus any chest CT). A generalized linear model assessed the impact of time frame and CT type on adherence. A novel four-category adherence schema was developed.

Results: In 13,773 patients, LDCT-only binary adherence (n = 10,237) was 35.6%, 47.8%, and 55.9% at 15, 24, and unlimited months, respectively. Broadening the evaluation to any type of follow-up chest CT (n = 10,436), adherence rose to 45.8%, 59.2%, and 68.4%. Adherence was significantly higher with (1) the 24-month and unlimited time frames compared with 15 months (P < .001) and (2) considering any chest CT versus LDCT only (P < .001). With the four-category schema (n = 13,773), LDCT-only rates were: on time (26.4%), late (15.1%), never followed up (32.7%), and not overdue (25.7%). Including any chest CT, the respective values were 34.7%, 17.1%, 24.0%, and 24.2%.

Interpretation: This study demonstrates the impact of varying adherence definitions on LCS follow-up rates. Standardized definitions would facilitate program comparison. The proposed four-category schema delineates the screening status of a LCS program's entire cohort.

Objective: To assess the feasibility, acceptability, and efficacy of a patient-radiologist video image review.

Methods: In this prospective observational study, radiologist-led patient video visits were piloted from December 2021 to December 2023. Visits were offered to patients in our lung screening program and patients with incidental pulmonary nodules. All visits were conducted using Zoom virtual meeting platform (Zoom Communications, Inc. San Jose, California). At each visit, the radiologist reviewed the imaging findings and answered all questions. Our patient navigator scheduled the recommended follow-up by the radiologist and documented the visit in our institution's electronic medical record. Visits concluded with a patient survey.

Results: In all, 156 video visits were offered to patients with a median age of 61 years (interquartile range, 54-67); 92 were women. Sixty-three (40%) of the video visits offered were declined by patients, 73 (78%) visits were completed by 71 patients, 10 (11%) patients did not attend the scheduled visit, and 10 (11%) patients were scheduled outside of our study time frame (after December 30, 2023). The mean video visit duration was 12 min. Of the 156 patients, 127 had follow-up imaging scheduled. Patients who completed a video visit were significantly more likely to return for follow-up imaging compared with those who declined video visits (62 of 66, 94%, versus 49 of 61, 80%) (P = .019). Of the 58 patients who completed the postvisit questionnaire, 95% (57 of 58) reported that the visit made them more likely to return for follow-up imaging, 59% (34 of 58) reported a decrease in their imaging-related anxiety, and 98% (57 of 58) wanted to use imaging review for future scans.

Conclusion: Radiologist-patient virtual image reviews are feasible when performed in conjunction with patient navigators. Virtual visits provide an opportunity to engage patients, address communication gaps, and impact imaging follow-up rates.

Purpose: The aim of this study was to evaluate the feasibility of using large language models (LLMs) to automate ground truth label extraction from radiology reports, enabling scalable assessment and monitoring of radiologic artificial intelligence (AI) tools. The framework was tested by validating AI model performance on a newly installed photon-counting CT (PCCT) scanner.

Methods: Four FDA-cleared deep learning-based computer-aided detection and triage tools targeting pulmonary embolism, intracranial hemorrhage, cervical spinal fractures, and vertebral compression fractures were retrospectively analyzed. Radiology reports from examinations acquired using the new PCCT scanner and conventional scanners were processed using an LLM (Llama 3.3) to extract binary ground truth labels. AI outputs were compared with these labels to estimate performance metrics. Discrepant cases were adjudicated by three human annotators, with interrater reliability measured using Fleiss's κ test. Performance metrics were recalculated after partial human correction of LLM errors.

Results: LLM-extracted labels enabled rapid performance assessment across all four diagnostic tasks. There were no statistically significant differences in performance between the PCCT and non-PCCT cohorts. In discrepant cases, the agreement between LLM labels and final human annotations (κ = 0.731) was comparable with interreader agreement (κ = 0.720), supporting the reliability of LLM labeling.

Conclusions: LLMs can be used to automate ground truth label extraction from radiology reports, offering a scalable and efficient alternative to manual annotation. This method supports rapid local validation of AI tools, even in response to input drift from new imaging hardware.

Objective: To evaluate the effectiveness of artificial intelligence (AI)-assisted examination for cancer detection in medical imaging.

Methods: We searched seven databases from January 1, 2017, until June 30, 2024, to identify randomized controlled trials (RCTs). The primary outcomes were detection rates and patient-centered outcomes. Pooled relative risks (RRs) with 95% confidence intervals (CIs) were calculated.

Results: We included 49 RCTs covering seven cancer types, with 79.6% (n = 39) being colorectal cancer. AI-assisted examination showed varying effects on detection rates across different cancer types. Specifically, regarding colorectal cancer, AI increased detection rates for both adenoma (pooled RR = 1.22, 95% CI: 1.17-1.28, 36 RCTs) and polyp (pooled RR = 1.20, 95% CI: 1.14-1.26, 28 RCTs). For esophageal cancer, positive effects were also observed on the detection rates of high-risk esophageal lesions (RR = 2.01, 95% CI: 1.06-3.80, 1 RCT) as well as superficial esophageal squamous cell carcinoma and precancerous lesions (RR = 1.38, 95% CI: 1.03-1.86, 1 RCT). Moreover, statistically significant improvement in detection rates were observed in prostate cancer (pooled RR = 1.40, 95% CI: 1.10-1.77, 1 RCT with 3 arms), actionable lung nodules (RR = 2.38, 95% CI: 1.25-4.55, 1 RCT) for lung cancer, and breast cancer (RR = 1.20, 95% CI: 1.00-1.45, 1 RCT). However, no significant effect was observed on the detection rates of gastric or liver cancer.

Conclusions: AI-assisted examinations may improve certain detection rates but not all among seven cancer types. There is a notable lack of patient-centered outcomes, crucial for evaluating the ultimate benefits to patients. Future research should give priority to assessing the impact of AI on patient-centered outcomes beyond diagnostic accuracy.

Background: Lung cancer screening (LCS) uptake remains low nationally. We evaluated the feasibility and preliminary efficacy of a multilevel intervention to improve LCS uptake in a pilot trial.

Methods: Eligible patients were 50 to 80 and met 2021 United States Preventive Services Task Force LCS criteria. The Empower LCS intervention included (1) a decision aid; (2) a text reminder to encourage LCS discussion with primary care providers (PCPs); (3) PCP notifications on eligibility and barriers, and (4) financial hardship and health-related social needs support. Screening outcomes (LCS discussions, orders, and completion) at 6 months were assessed using medical records and surveys. Changes in LCS knowledge and health beliefs were assessed with surveys.

Results: In all, 70 patients enrolled (mean age: 62.5 ± 6.3; 70% male; 1.4% Black, 18.6% Asian, 44.3% White, 35.7% other); 45.7% were Hispanic, and 41% were current smokers. Common LCS barriers included cost concerns (40%, 28 of 70) and fear of finding something wrong (34.3%, 24 of 70). All received the decision aid, text reminder, and PCP alert. Of the patients, 72.9% reported financial hardship or health-related social needs and received support. At 6 months, 71.4% (50 of 70) discussed LCS with their PCP, 51.4% (36 of 70) received low-dose CT orders, and 27.1% (19 of 70) completed screening (52.8% of those with order). Completion exceeded the national average of 16% (P = 0.01). Knowledge and perceived severity changed significantly (knowledge: from 1.91 to 2.67, P = .01; severity: from 16.3 to 18.1, P = .0003). No significant changes were observed in perceived barriers or self-efficacy.

Conclusion: The Empower LCS intervention was feasible and improved LCS uptake. However, only half of those with LCS order, completed screening, suggesting the need for enhanced navigation.

Medicare Part C, better known as Medicare Advantage, is a federal program that allows Americans eligible for Medicare to obtain health care coverage through private insurers. For beneficiaries, there may be a number of potential benefits associated with Medicare Advantage over traditional Medicare, but in recent years the program has come under scrutiny due to concerns about cost, quality, and transparency. In this article, we will review a brief history of Medicare and Medicare Advantage as well as address the advantages and concerns regarding Medicare Advantage and why this is important for radiology.