Radiology Reports Research Articles

Extraction of clinical data on major pulmonary diseases from unstructured radiologic reports using a large language model.

Despite significant strides in big data technology, extracting information from unstructured clinical data remains a formidable challenge. This study investigated the utility of large language models (LLMs) for extracting clinical data from unstructured radiological reports without additional training. In this retrospective study, 1800 radiologic reports, 600 from each of the three university hospitals, were collected, with seven pulmonary outcomes defined. Three pulmonology-trained specialists discerned the presence or absence of diseases. Data extraction from the reports was executed using Google Gemini Pro 1.0, OpenAI's GPT-3.5, and GPT-4. The gold standard was predicated on agreement between at least two pulmonologists. This study evaluated the performance of the three LLMs in diagnosing seven pulmonary diseases (active tuberculosis, emphysema, interstitial lung disease, lung cancer, pleural effusion, pneumonia, and pulmonary edema) utilizing chest radiography and computed tomography scans. All models exhibited high accuracy (0.85-1.00) for most conditions. GPT-4 consistently outperformed its counterparts, demonstrating a sensitivity of 0.71-1.00; specificity of 0.89-1.00; and accuracy of 0.89 and 0.99 across both modalities, thus underscoring its superior capability in interpreting radiological reports. Notably, the accuracy of pleural effusion and emphysema on chest radiographs and pulmonary edema on chest computed tomography scans reached 0.99. The proficiency of LLMs, particularly GPT-4, in accurately classifying unstructured radiological data hints at their potential as alternatives to the traditional manual chart reviews conducted by clinicians.

Magnetic Resonance Imaging Template to Standardize Reporting of Evacuation Disorders

Magnetic resonance (MR) defecography, including both static and dynamic phases, is frequently requested by gastroenterologists and colorectal surgeons for planning the treatment of obstructive defecation syndrome and pelvic organ prolapse. However, reports often lack key information needed to guide treatment strategies, making management challenging and, at times, controversial. It has been hypothesized that using structured radiology reports could reduce missing information. In this paper, we present a structured MR defecography template report that includes nine key descriptors of rectal evacuation. The effectiveness and acceptability of this template are currently being assessed in Italy through a national interdisciplinary study.

Auto-Rad: End-to-End Report Generation from Lumber Spine MRI Using Vision–Language Model

Background: Lumbar spinal stenosis (LSS) is a major cause of chronic lower back and leg pain, and is traditionally diagnosed through labor-intensive analysis of magnetic resonance imaging (MRI) scans by radiologists. This study aims to streamline the diagnostic process by developing an automated radiology report generation (ARRG) system using a vision–language (VL) model. Methods: We utilized a Generative Image-to-Text (GIT) model, originally designed for visual question answering (VQA) and image captioning. The model was fine-tuned to generate diagnostic reports directly from lumbar spine MRI scans using a modest set of annotated data. Additionally, GPT-4 was used to convert semistructured text into coherent paragraphs for better comprehension by the GIT model. Results: The model effectively generated semantically accurate and grammatically coherent reports. The performance was evaluated using METEOR (0.37), BERTScore (0.886), and ROUGE-L (0.3), indicating its potential to produce clinically relevant content. Conclusions: This study highlights the feasibility of using vision–language models to automate report generation from medical imaging, potentially reducing the diagnostic workload for radiologists.

Prevalence and clinical impact of radiographic sclerotic lines adjacent to cementless tibial stems in revision total knee arthroplasty: a long-term follow-up study.

A retrospective chart review of a consecutive series of patients who underwent revision TKA at a single tertiary care institution between January 2001 and December 2009 was performed. Data were collected regarding patient demographics, stem size, types of implants, and complications. Radiological reports and radiographs were evaluated for sclerotic lines, propagation, and their location around the tibial stem. Aseptic loosening rates were compared between patients with and without sclerotic lines. One hundred and fifty-three patients with at least 15 years of follow-up were identified. The mean follow-up was 16.5 ± 4.12 years (range 15-23). Overall, 19 patients required re-revision surgery. Survivorship free of failure due to aseptic loosening was 94.8% (95% CI 76.4-98) at 15 years, with three patients requiring re-revision surgery for aseptic loosening. Twelve (7.8%) patients were found to have sclerotic lines surrounding the tibial stem, all under 2 mm in width. Two (1%) patients with sclerotic lines required re-revision surgery, one patient due to infection and one due to aseptic loosening. There was no significant difference in rates of aseptic loosening in patients with or without sclerotic lines. Radiographic sclerotic lines around revision TKA stems-measuring less than 2mm wide, nonprogressive, and located at the lateral-distal aspect of the stem-were identified in fewer than 8% of cases and did not correlate with re-revision surgery.

Centralized Investigator Review of Radiological and Functional Imaging Reports in Real-World Oncology Studies: The SACHA-France Experience.

SACHA-France (NCT04477681) is a prospective real-world study that collects clinical safety and efficacy data of novel anticancer therapies prescribed off-label or on compassionate use to patients <25 years. From March 2020 until February 2024, 640 patients with solid tumors or lymphomas were included, with 176 (28%) reported objective tumor responses. Centralized medical monitoring of local radiological/functional imaging reports by the SACHA coordinating investigator led to response modification in 45 out of 176 cases (26%), highlighting the relevance of the medical review of study data. We suggest this pragmatic approach for improving clinical trial data when centralized radiological review is not performed.

The impact of different radiology report formats on patient information processing: a systematic review.

Since radiology reports are primarily written for health professionals, patients may experience difficulties understanding jargon and terminology used, leading to anxiety and confusion. This review evaluates the impact of different radiology report formats on outcomes related to patient information processing, including perception, decision (behavioral intention), action (actual health behavior), and memory (recall of information). PubMed, Web of Science, EMBASE, and PsycInfo were searched for relevant qualitative and quantitative articles describing or comparing ways of presenting diagnostic radiology reports to patients. Two reviewers independently screened for relevant articles and extracted data from those included. The quality of articles was assessed using the Mixed Methods Appraisal Tool. Eighteen studies, two qualitative and sixteen quantitative, were included. Sixteen studies compared multiple presentation formats, most frequently traditional unmodified reports (n = 15), or reports with anatomic illustrations (n = 8), lay summaries (n = 6) or glossaries (n = 6). Glossaries, illustrations, lay summaries, lay reports or lay conclusions all significantly improved participants' cognitive perception and perception of communication of radiology reports, compared to traditional reports. Furthermore, these formats increased affective perception (e.g., reduced anxiety and worry), although only significant for lay reports and conclusions. Modifying traditional radiology reports with glossaries, illustrations or lay language enhances patient information processing. Question Identifying the impact of different radiology report formats on outcomes related to patient information processing to enhance patient engagement through online access to radiology reports. Findings Lay language summaries, glossaries with patient-oriented definitions, and anatomic illustrations increase patients' satisfaction with and understanding of their radiology reports. Clinical relevance To increase patients' satisfaction, perceived usefulness and understanding with radiology reports, the use of lay language summaries, glossaries with patient-oriented definitions, and anatomic illustrations is recommended. These modifications decrease patients' unnecessary insecurity, confusion, anxiety and physician consultations after viewing reports.

EDI&amp;S SO04 Accuracy of Outsourced Radiology Reports in Emergency Surgical Care; Do they provide a high-quality cost-effective service?

Abstract Background Computed Tomography (CT) scans play a crucial role in emergency surgical care enabling proper management and prompt decision making. In our study we aimed to compare abdominal CT scan reports created by inhouse radiologists to those created by outsourced radiology reporting service for patients admitted with acute surgical pathologies. The study also looked at the impact of the reporting style (free text or structured format) on the quality of scan reports. Method The study was conducted in a busy general surgery unit located in the North West of England. Patients who were admitted acutely under the surgical specialty and had abdominal CT scans over a period of three months from August 2023 to October 2023 were included. Patients who were admitted following trauma or had non abdominal CT scans have been excluded. Statistical analysis was conducted to compare inhouse radiology reports to outsourced ones. Scan reports were thoroughly scrutinised to identify whether they included all relevant information needed and whether they were written in structured or free text format. Results A total of 192 patients were identified with a median age of 59 years. Outsourced radiology reports were found to be significantly underreporting comments on the biliary system (OR=0.369, P=0.046), spleen (OR=0.414, P=0.044), kidneys (OR=0.354, P=0.027), lymph nodes (OR=0.412, P=0.011), mesentery and peritoneum (OR=0.364, P=0.004), vasculature (OR=0.381, P=0.002), bones and soft tissues (OR=0.273, P=0.003), as well as lung bases (OR=0.300, P=0.009). Structured format was found to outperform free text style on comments on spleen (OR=8.064, P=0.044), adrenals (OR=8.468, P=0.039), mesentery and peritoneum (OR=4.618, P=0.000), as well as bone and soft tissues (OR=9.295, P=0.031). Conclusion CT scans reported by outsourced radiology service were found to be significantly underreporting certain key elements. Additionally, free style reporting format was found to miss some crucial aspects of abdominal scan reports. That adds further challenges to clinicians assessing and managing acute surgical patients. Further studies auditing different outsourced companies are recommended to ensure patients are provided with high-quality cost-effective emergency service.

Documentation of incidentally noted hepatic steatosis to emergency department patients: A retrospective study.

Hepatic steatosis is a common incidental finding on emergency department (ED) imaging studies, occurring in up to 10% of studies, and carries significant long-term morbidity. Frequently considered an unimportant finding, it is unknown how often ED patients are informed of hepatic steatosis. Our objective was to examine hepatic steatosis inclusion in ED discharge materials. Data from discharged patients at an urban academic ED (>90,000 visits) with abdominal imaging (computed tomography or ultrasound) from 2019 to 2022 were screened. Patients with radiology reports documenting hepatic steatosis were included. Two trained data abstractors analyzed discharge materials and coded disclosure of steatosis (present/absent). Data abstraction also noted how the finding was included (e.g., follow-up instructions, provision of radiology report). Factors associated with patient disclosure were examined through regression models including age, race, ethnicity, insurance, number of imaging studies during encounter, type of imaging study, and inclusion of hepatic steatosis in the impression section of radiology report. Of 10,677 radiology reports, 1209 (11.3%) had documented hepatic steatosis. The mean (±SD) age was 47.1 (±14.1) years; 56.4% were female, 53.5% were White, and 30.7% Hispanic. Only 173 of 1209 patients (14.3%) received any discharge documentation of hepatic steatosis. In 65% of cases where disclosure occurred, the imaging report was pasted verbatim into discharge material. There were no significant differences in discharge documentation by demographic groups. Mention within the radiology report impression (rather than the report body alone) occurred in 73.8% of records and was the strongest predictor of disclosure to patients (adjusted odds ratio 2.18, 95% confidence interval 1.39-3.54). Despite the high prevalence of hepatic steatosis in radiology reports, rates of documentation in patient-facing discharge materials are notably low, exposing a possible communication failure with consequences for diagnosis.

Implementation of structured radiology reporting and its associated accuracy in comparison to pancreas multi-disciplinary clinic expert radiology review.

To evaluate the feasibility of implementation of structured reporting in the setting of a high-volume pancreatic multidisciplinary clinic (PMDC) and to assess its value by comparing the accuracy of structured reports with expert imaging reviews. A single institutional prospective cohort study was conducted during March 2022 to May 2024 to understand the feasibility of implementation of structured reporting (SR) for all patients who were seen in our weekly PMDC. Descriptive and regression analyses were performed to find an association between SR and difference in vascular involvement of the primary pancreatic tumor between the radiology report and expert radiologist review (gold standard) during PMDC. Among 466 patients seen in the PMDC, 426 (91.4%) had reports generated prior to PMDC. Of this, 294 reports met the inclusion criteria. The usage of SR increased from 58.3% in Mar 2022 to 87.8% in June 2024. Majority of the reports that used SR (n = 226, 76.9%), were performed for initial staging (n = 197, 67.0%) of PC. The median years of experience of reading radiologists that used non-SR was 14 (IQR: 8-27) years, while it was 9 (IQR -9-15) years for those who used SR (p = 0.030). Of note, as compared to the radiology report, increased vascular involvement was noted in PMDC review 62.5% (20 out of 32) of the time with non-SRs, whereas increased vascular involvement during PMDC review was noted in only 36.6% (48 out of 132) of the time with SRs. On multivariable analysis, using SR lowered the odds of increase in vascular involvement during PMDC review by 0.29 times (95CIs 0.11-0.79; p = 0.015). SR is feasible and superior to the free-text reporting with respect to the accuracy of peri-pancreatic vascular involvement. While its use cannot replace the PMDC radiology review, it can nonetheless be an indispensable tool in clinical management, particularly in a non-PMDC setting.

Abstract 4119869: Automatic assignment of CAD-RADS categories in coronary CTA reports using large language model

Background: The effectiveness of different large language models (LLM) in extracting important information from CT angiography (CTA) radiology reports is unknown. Purpose: The Coronary Artery Disease-Reporting and Data System (CAD-RADS) is a standardized reporting system for CTA that categorizes the severity of coronary artery disease to aid in clinical decision-making. To evaluate the agreement between human readers and LLMs in assigning CAD-RADS categories based on CTA reports. Materials and Methods: This retrospective study analyzed reports from patients who underwent coronary CTA for screening or diagnosis at Johns Hopkins Hospital. Reports with CAD-RADS 0–5 findings were collected from Jan. 2020 to Oct. 2023. Initially, 6,212 CTA reports were retrieved. After excluding non-organized reports and retaining only those with CAD-RADS scores, 590 reports remained. Board-certified radiologists and three large language models (GPT-3.5, GPT-4, and Llama3) assigned CAD-RADS categories based on the original radiologists' findings. Two radiologists reviewed the CAD scores, with only one discrepancy in the 590 reports, indicating high inter-agreement. Agreement between human readers and LLMs for CAD-RADS categories was assessed using the Gwet agreement coefficient (AC1 value). Frequencies of changes in CAD-RADS category assignments that could potentially affect clinical management (CAD-RADS 0-2 vs. 3-5) were calculated and compared. Typically, CAD-RADS 0-2 indicates mild disease leading to conservative management, while CAD-RADS 3-5 denotes further testing or disease requiring aggressive intervention to prevent adverse cardiovascular events. Results: Among 590 reports, agreement between original and reviewing radiologists was almost perfect (AC1 = 0.985). Good agreement was also found between original radiologists and Llama3, GPT-3.5, and GPT-4 (AC1 = 0.861, 0.907, 0.941, respectively). Differences in CAD-RADS category upgrades/downgrades potentially affecting clinical management were observed: 0 of 590 (0%) for human readers, 56 of 590 (9.4%) for Llama3, 33 of 590 (5.5%) for GPT-3.5, and 23 of 590 (3.9%) for GPT-4 upgrading CAD-RADs; 8 of 590 (1.4%) for human readers, 19 of 590 (3.2%) for Llama3, 22 of 590 (3.7%) for GPT-3.5, and 12 of 590 (2.0%) for GPT-4 downgrading RAD-CADs. Conclusion: LLMs showed good agreement with human readers in assigning CAD-RADS categories on CTA reports. However, they were more likely to upgrade CAD-RADS category when compared to radiologists.

Discrepancy Rates in Acute Abdominal CT: An Audit of In-House and Remote Reporting Compared to Intraoperative Laparoscopic and Laparotomy Findings.

Non-traumatic abdominal pain is a common emergency presentation frequently managed by general surgeons. Abdominopelvic computed tomography (CT) scans are the most popular imaging modality in this context. In many hospitals, the rising demand for urgent and emergent scans out-of-hours has necessitated the outsourcing of this service to teleradiology companies, whereby reports are generated at sites remote from the image acquisition. The primary aim of this study was to determine the discrepancy rates of preoperative CT imaging by source (teleradiology compared to in-house). This was a retrospective monocentric study conducted at a busy district general hospital over a seven-month period. Patient demographic data, operative notes, and radiology reports (by source) were collated for all patients aged ≥16 years presenting with atraumatic abdominal pain who underwent abdominopelvic CT with subsequent surgical intervention (laparoscopy and/or laparotomy). Seventy-one patients were identified by initial screening, and 10 patients (11 scans) met the criteria for a "major" discrepancy. Overall discrepancy rates were calculated at 5.6% for scans reported off-site compared to 9.9% for reports generated by in-house radiologists. This study demonstrated lower discrepancy rates in scans reported remotely and can be used as the catalyst for improving aspects of in-house CT reporting.

Abstract 4124675: Deep Learning Screening of Cardiac MRIs Uncovers Undiagnosed Hypertrophic Cardiomyopathy in the UK BioBank

Introduction: The prevalence of hypertrophic cardiomyopathy (HCM) in the UK Biobank based on ICD-10 codes (.07%) is lower than global estimates of disease prevalence (0.2 - 0.5%). Prior studies using this data have remarked on the limitations of findings given likely underdiagnosis. The availability of cardiac MRI scans on a fraction of the participants offers an opportunity to identify missed diagnoses. Aims: This study seeks to utilize a generalizable deep learning model to detect likely cases of undiagnosed hypertrophic cardiomyopathy from cardiac MRIs in the UK Biobank. Methods: The foundational model was trained on a multi-institutional dataset of 14,073 cardiac MRIs via a self-supervised contrastive learning approach that sought to minimize the divergence between scans and their associated radiology reports. The pre-trained model was fine-tuned to diagnose hypertrophic cardiomyopathy on a distinct cohort of 4,870 MRIs with 368 cases of HCM, achieving an AUC of 0.94. The fine-tuned model was applied to the UK Biobank cardiac MRI dataset to ascertain predicted probabilities of HCM. Cases exceeding a threshold of 95% – correlating to the top 0.5% of cases (expected specificity of 97% and sensitivity of 60%) – were screened in for manual reading. In a blinded fashion, a board-certified radiologist was tasked with diagnosing HCM on a sample of cases composed of high and low predicted probabilities. Results: Of the 43,017 patients with cardiac MRIs, only 9 (.02%) had an ICD diagnosis of HCM. 266 cardiac MRIs were manually reviewed: 216 had greater than 95% predicted probability of HCM; 50 negative controls were randomly selected amongst cases with predicted probability less than 10%. The radiologist concurred with an HCM diagnosis for 115 cases (sensitivity 53%, specificity 98%), 112 of which were previously undiagnosed. The prevalence of hypertension and aortic stenosis did not significantly differ between the cohort of true positives (69.2%) and false positives (76.6%). The corrected prevalence of HCM in the UK BioBank MRI cohort is estimated at 0.28%. Conclusions: The findings of this study illustrate the remarkable ability of a generalizable deep learning model to detect undiagnosed cases of a rare disease process from cardiac MRIs. This is an important milestone that may allow for widespread screening of hypertrophic cardiomyopathy while minimizing demand for radiologist labor, and thereby allow patients to reap the substantial benefits of earlier treatment.

Abstract 4139198: A Systematic Approach to Prompting Large Language Models for Automated Feature Extraction from Cardiovascular Imaging Reports

Introduction: Cardiovascular radiology reports contain valuable diagnostic information linked to images, but the unstructured text format makes feature extraction difficult on a large scale. Large language models (LLMs) allow for feature extraction where string parsing alone is insufficient, but require careful prompting for accurate results. Hypothesis: We hypothesize that a systematic prompting approach using LLMs can expedite the extraction of features from unstructured text in transesophageal echocardiography (TEE) reports. Methods: The data consisted of 7106 intraoperative TEE reports, 600 of which were manually reviewed to obtain pre- and post-intervention ground truth values for left ventricular ejection fraction (LVEF), right ventricular systolic function (RVSF), and tricuspid regurgitation (TR). Reports are paired with an imaging study consisting of 50-200 clips. For each feature considered, 100 of the 600 labeled reports were used to engineer a prompt in Llama-2 that maximized feature extraction accuracy. Results: We found that using multiple, shorter prompts yielded higher accuracy than did fewer, longer prompts. Additionally, when imposing semantic information onto a numerical scale, prompt engineering in combination with string parsing (Figure 1) gave the best results. When evaluated on the 500 labeled reports withheld for testing, the finalized prompts had accuracies of 94.1%, 94.8%, and 91.3% for LVEF, RVSF, and TR, respectively. Using this strategy, 5000 intraoperative TEE reports were analyzed and used to train and evaluate a regression model for LVEF estimation from TEE clips (Figure 2). Conclusion: We have shown that performing prompt engineering on Llama-2 can be used to extract features from unstructured TEE reports in an accurate manner. As an extension of these methods, automated feature prediction from echocardiograms can be used to create rapid, low-cost, and accessible cardiac assessments.

ChatGPT vs Gemini: Comparative Accuracy and Efficiency in CAD-RADS Score Assignment from Radiology Reports.

This study aimed to evaluate the accuracy and efficiency of ChatGPT-3.5, ChatGPT-4o, Google Gemini, and Google Gemini Advanced in generating CAD-RADS scores based on radiology reports. This retrospective study analyzed 100 consecutive coronary computed tomography angiography reports performed between March 15, 2024, and April 1, 2024, at a single tertiary center. Each report containing a radiologist-assigned CAD-RADS score was processed using four large language models (LLMs) without fine-tuning. The findings section of each report was input into the LLMs, and the models were tasked with generating CAD-RADS scores. The accuracy of LLM-generated scores was compared to the radiologist's score. Additionally, the time taken by each model to complete the task was recorded. Statistical analyses included Mann-Whitney U test and interobserver agreement using unweighted Cohen's Kappa and Krippendorff's Alpha. ChatGPT-4o demonstrated the highest accuracy, correctly assigning CAD-RADS scores in 87% of cases (κ = 0.838, α = 0.886), followed by Gemini Advanced with 82.6% accuracy (κ = 0.784, α = 0.897). ChatGPT-3.5, although the fastest (median time = 5s), was the least accurate (50.5% accuracy, κ = 0.401, α = 0.787). Gemini exhibited a higher failure rate (12%) compared to the other models, with Gemini Advanced slightly improving upon its predecessor. ChatGPT-4o outperformed other LLMs in both accuracy and agreement with radiologist-assigned CAD-RADS scores, though ChatGPT-3.5 was significantly faster. Despite their potential, current publicly available LLMs require further refinement before being deployed for clinical decision-making in CAD-RADS scoring.

NIMG-63. LEVERAGING LLMS FOR ACCURATE DIFFERENTIATION OF RADIATION NECROSIS AND TUMOR PROGRESSION IN BRAIN MRI REPORTS: A STUDY ON AUTOMATED SCORING AND CLINICAL IMPLICATIONS

Abstract BACKGROUND Accurate differentiation between radiation necrosis and tumor progression or recurrence in patients treated with stereotactic radiosurgery for brain metastases is critical for guiding clinical management. This study leverages the advanced natural language processing capabilities of Meta Llama3, an artificial intelligence (AI) large language model (LLM), combined with prompt engineering, to rapidly categorize brain magnetic resonance imaging (MRI) radiology reports. Our objective was to develop an automated scoring system to classify concern for radiation necrosis, tumor progression, equivocal findings, or stable exams. METHODS Using a comprehensive dataset of reports annotated by expert radiologists, we ran inference on a 70-billion parameter Llama3 model (temperature 0.2, top_p 0.9) with specific prompts designed to capture the nuanced language and diagnostic criteria related to radiation necrosis or tumor progression. RESULTS The first pass was performed on a training dataset of 107 reports and did not predefine the clinical conditions. This demonstrated 43.4% accuracy in scoring when compared to a human user’s classification of each report. Agreement between the human reader and Llama3 was assessed using the Gwet agreement coefficient, AC1=0.411 (99%CI 0.396-0.426). Multiple iterations of targeted prompt engineering were then employed to narrow the definition of radiation necrosis and tumor progression, with specific examples and nuanced language used to achieve a higher degree of accuracy at 72.0%, AC1=0.719 (99%CI 0.717-0.722). DISCUSSION/FUTURE DIRECTIONS This surpasses recent demonstration of lower human-LLM agreement in radiographic score assignment, with further room for calibration on a dataset of several thousand reports. Next, we will correlate automated interpretations with actual clinical management decisions for radiation necrosis (e.g., initiation of steroids, bevacizumab, Laser Interstitial Thermal Therapy, and/or repeat imaging). This automated scoring system holds significant potential for LLMs in clinical applications. Future work will focus on integrating this model into clinical workflows and expanding its capabilities to include longitudinal monitoring of patient outcomes.

ChatGPT and radiology report: potential applications and limitations.

Large language models like ChatGPT, with their growing accessibility, are attracting increasing interest within the artificial intelligence medical field, particularly in the analysis of radiology reports. These present a valuable opportunity to explore the potential clinical applications of large language models, given their huge capabilities in processing and understanding written language. Early research indicates that ChatGPT could offer benefits in radiology reporting. ChatGPT can assist but not replace radiologists in achieving diagnoses, generating structured reports, extracting data, identifying errors or incidental findings, and can also serve as a support in creating patient-friendly reports. However, ChatGPT also has intrinsic limitations, such as hallucinations, stochasticity, biases, deficiencies in complex clinical scenarios, data privacy and legal concerns. To fully utilize the potential of ChatGPT in radiology reporting, careful integration planning and rigorous validation of their outputs are crucial, especially for tasks requiring abstract reasoning or nuanced medical context. Radiologists' expertise in medical imaging and data analysis positions them exceptionally well to lead the responsible integration and utilization of ChatGPT within the field of radiology. This article offers a topical overview of the potential strengths and limitations of ChatGPT in radiological reporting.

Collaboration between clinicians and vision-language models in radiology report generation.

Automated radiology report generation has the potential to improve patient care and reduce the workload of radiologists. However, the path toward real-world adoption has been stymied by the challenge of evaluating the clinical quality of artificial intelligence (AI)-generated reports. We build a state-of-the-art report generation system for chest radiographs, called Flamingo-CXR, and perform an expert evaluation of AI-generated reports by engaging a panel of board-certified radiologists. We observe a wide distribution of preferences across the panel and across clinical settings, with 56.1% of Flamingo-CXR intensive care reports evaluated to be preferable or equivalent to clinician reports, by half or more of the panel, rising to 77.7% for in/outpatient X-rays overall and to 94% for the subset of cases with no pertinent abnormal findings. Errors were observed in human-written reports and Flamingo-CXR reports, with 24.8% of in/outpatient cases containing clinically significant errors in both report types, 22.8% in Flamingo-CXR reports only and 14.0% in human reports only. For reports that contain errors we develop an assistive setting, a demonstration of clinician-AI collaboration for radiology report composition, indicating new possibilities for potential clinical utility.

Accuracy of Outsourced Radiology Reports in Emergency Surgical Care: Do They Provide a High-Quality, Cost-Effective Service?

Accuracy of Outsourced Radiology Reports in Emergency Surgical Care: Do They Provide a High-Quality, Cost-Effective Service?

Clinical Characteristics and Outcomes of Hospitalized AECOPDs Secondary to SARS-CoV-2 versus Other Respiratory Viruses.

To compare clinical characteristics and outcomes of hospitalized acute exacerbations of COPD (AECOPD)s secondary to SARS-CoV-2 versus other respiratory viruses amongst a highly vaccinated population in the Omicron era. Retrospective cohort study; analysis of hospital medical records and linked pathology and radiology reports. Tertiary health network in Victoria, Australia; January 2022-August 2022. Key clinical information including comorbidities, vaccination status, treatments administered and outcomes such as hospital length of stay, ICU admission, non-invasive ventilation usage and inpatient mortality. One hundred ninety-nine viral AECOPDs - 125 SARS-CoV-2 and 74 other viruses were identified. Of the SARS-CoV-2 group. 13.6% were unvaccinated, 17.6% partially and 68.0% fully vaccinated. The SARS-CoV-2 group were older (77.2 vs 68.9, p < 0.00001) with more comorbidities (1[1-2] vs 1[0-2], p = 0.008) and lower candidacy for full resuscitation (25.6% vs 56.8%, p < 0.0001). Mortality tended to be higher among SARS-CoV2 admission (9.6% v 2.7%, p = 0.066) but rates of ICU admission (10.4% v 13.5%, p = 0.507), length of hospitalisation (5[3-8] vs 5[3-9], p = 0.9) and readmission within 30 days (25% vs 33.3%, p = 0.184) were similar. In a highly vaccinated population in the Omicron era, COPD patients requiring hospitalisation with SARS-CoV-2 are older with more comorbidities than those admitted with other respiratory viruses. Length of hospitalisation and ICU utilisation was similar. Inpatient mortality may be higher.

Precise Image-level Localization of Intracranial Hemorrhage on Head CT Scans with Deep Learning Models Trained on Study-level Labels.

Purpose To develop a highly generalizable weakly supervised model to automatically detect and localize image-level intracranial hemorrhage (ICH) by using study-level labels. Materials and Methods In this retrospective study, the proposed model was pretrained on the image-level Radiological Society of North America dataset and fine-tuned on a local dataset by using attention-based bidirectional long short-term memory networks. This local training dataset included 10 699 noncontrast head CT scans in 7469 patients, with ICH study-level labels extracted from radiology reports. Model performance was compared with that of two senior neuroradiologists on 100 random test scans using the McNemar test, and its generalizability was evaluated on an external independent dataset. Results The model achieved a positive predictive value (PPV) of 85.7% (95% CI: 84.0, 87.4) and an area under the receiver operating characteristic curve of 0.96 (95% CI: 0.96, 0.97) on the held-out local test set (n = 7243, 3721 female) and 89.3% (95% CI: 87.8, 90.7) and 0.96 (95% CI: 0.96, 0.97), respectively, on the external test set (n = 491, 178 female). For 100 randomly selected samples, the model achieved performance on par with two neuroradiologists, but with a significantly faster (P < .05) diagnostic time of 5.04 seconds per scan (vs 86 seconds and 22.2 seconds for the two neuroradiologists, respectively). The model's attention weights and heatmaps visually aligned with neuroradiologists' interpretations. Conclusion The proposed model demonstrated high generalizability and high PPVs, offering a valuable tool for expedited ICH detection and prioritization while reducing false-positive interruptions in radiologists' workflows. Keywords: Computer-Aided Diagnosis (CAD), Brain/Brain Stem, Hemorrhage, Convolutional Neural Network (CNN), Transfer Learning Supplemental material is available for this article. © RSNA, 2024 See also the commentary by Akinci D'Antonoli and Rudie in this issue.