Radiographic Reports Research Articles

Integrating VAI-Assisted Quantified CXRs and Multimodal Data to Assess the Risk of Mortality.

To address the unmet need for a widely available examination for mortality prediction, this study developed a foundation visual artificial intelligence (VAI) to enhance mortality risk stratification using chest X-rays (CXRs). The VAI employed deep learning to extract CXR features and a Cox proportional hazard model to generate a hazard score ("CXR-risk"). We retrospectively collected CXRs from patients visited outpatient department and physical examination center. Subsequently, we reviewed mortality and morbidity outcomes from electronic medical records. The dataset consisted of 41,945, 10,492, 31,707, and 4441 patients in the training, validation, internal test, and external test sets, respectively. During the median follow-up of 3.2 (IQR, 1.2-6.1) years of both internal and external test sets, the "CXR-risk" demonstrated C-indexes of 0.859 (95% confidence interval (CI), 0.851-0.867) and 0.870 (95% CI, 0.844-0.896), respectively. Patients with high "CXR-risk," above 85th percentile, had a significantly higher risk of mortality than those with low risk, below 50th percentile. The addition of clinical and laboratory data and radiographic report further improved the predictive accuracy, resulting in C-indexes of 0.888 and 0.900. The VAI can provide accurate predictions of mortality and morbidity outcomes using just a single CXR, and it can complement other risk prediction indicators to assist physicians in assessing patient risk more effectively.

Automatic structuring of radiology reports with on-premise open-source large language models.

Structured reporting enhances comparability, readability, and content detail. Large language models (LLMs) could convert free text into structured data without disrupting radiologists' reporting workflow. This study evaluated an on-premise, privacy-preserving LLM for automatically structuring free-text radiology reports. We developed an approach to controlling the LLM output, ensuring the validity and completeness of structured reports produced by a locally hosted Llama-2-70B-chat model. A dataset with de-identified narrative chest radiograph (CXR) reports was compiled retrospectively. It included 202 English reports from a publicly available MIMIC-CXR dataset and 197 German reports from our university hospital. Senior radiologist prepared a detailed, fully structured reporting template with 48 question-answer pairs. All reports were independently structured by the LLM and two human readers. Bayesian inference (Markov chain Monte Carlo sampling) was used to estimate the distributions of Matthews correlation coefficient (MCC), with [-0.05, 0.05] as the region of practical equivalence (ROPE). The LLM generated valid structured reports in all cases, achieving an average MCC of 0.75 (94% HDI: 0.70-0.80) and F1 score of 0.70 (0.70-0.80) for English, and 0.66 (0.62-0.70) and 0.68 (0.64-0.72) for German reports, respectively. The MCC differences between LLM and humans were within ROPE for both languages: 0.01 (-0.05 to 0.07), 0.01 (-0.05 to 0.07) for English, and -0.01 (-0.07 to 0.05), 0.00 (-0.06 to 0.06) for German, indicating approximately comparable performance. Locally hosted, open-source LLMs can automatically structure free-text radiology reports with approximately human accuracy. However, the understanding of semantics varied across languages and imaging findings. Question Why has structured reporting not been widely adopted in radiology despite clear benefits and how can we improve this? Findings A locally hosted large language model successfully structured narrative reports, showing variation between languages and findings. Critical relevance Structured reporting provides many benefits, but its integration into the clinical routine is limited. Automating the extraction of structured information from radiology reports enables the capture of structured data while allowing the radiologist to maintain their reporting workflow.

Evaluating the Performance and Bias of Natural Language Processing Tools in Labeling Chest Radiograph Reports.

Background Natural language processing (NLP) is commonly used to annotate radiology datasets for training deep learning (DL) models. However, the accuracy and potential biases of these NLP methods have not been thoroughly investigated, particularly across different demographic groups. Purpose To evaluate the accuracy and demographic bias of four NLP radiology report labeling tools on two chest radiograph datasets. Materials and Methods This retrospective study, performed between April 2022 and April 2024, evaluated chest radiograph report labeling using four NLP tools (CheXpert [rule-based], RadReportAnnotator [RRA; DL-based], OpenAI's GPT-4 [DL-based], cTAKES [hybrid]) on a subset of the Medical Information Mart for Intensive Care (MIMIC) chest radiograph dataset balanced for representation of age, sex, and race and ethnicity (n = 692) and the entire Indiana University (IU) chest radiograph dataset (n = 3665). Three board-certified radiologists annotated the chest radiograph reports for 14 thoracic disease labels. NLP tool performance was evaluated using several metrics, including accuracy and error rate. Bias was evaluated by comparing performance between demographic subgroups using the Pearson χ2 test. Results The IU dataset included 3665 patients (mean age, 49.7 years ± 17 [SD]; 1963 female), while the MIMIC dataset included 692 patients (mean age, 54.1 years ± 23.1; 357 female). All four NLP tools demonstrated high accuracy across findings in the IU and MIMIC datasets, as follows: CheXpert (92.6% [47 516 of 51 310], 90.2% [8742 of 9688]), RRA (82.9% [19 746 of 23 829], 92.2% [2870 of 3114]), GPT-4 (94.3% [45 586 of 48 342], 91.6% [6721 of 7336]), and cTAKES (84.7% [43 436 of 51 310], 88.7% [8597 of 9688]). RRA and cTAKES had higher accuracy (P < .001) on the MIMIC dataset, while CheXpert and GPT-4 had higher accuracy on the IU dataset. Differences (P < .001) in error rates were observed across age groups for all NLP tools except RRA on the MIMIC dataset, with the highest error rates for CheXpert, RRA, and cTAKES in patients older than 80 years (mean, 15.8% ± 5.0) and the highest error rate for GPT-4 in patients 60-80 years of age (8.3%). Conclusion Although commonly used NLP tools for chest radiograph report annotation are accurate when evaluating reports in aggregate, demographic subanalyses showed significant bias, with poorer performance in older patients. © RSNA, 2024 Supplemental material is available for this article. See also the editorial by Cai in this issue.

Comparing Commercial and Open-Source Large Language Models for Labeling Chest Radiograph Reports.

Background Rapid advances in large language models (LLMs) have led to the development of numerous commercial and open-source models. While recent publications have explored OpenAI's GPT-4 to extract information of interest from radiology reports, there has not been a real-world comparison of GPT-4 to leading open-source models. Purpose To compare different leading open-source LLMs to GPT-4 on the task of extracting relevant findings from chest radiograph reports. Materials and Methods Two independent datasets of free-text radiology reports from chest radiograph examinations were used in this retrospective study performed between February 2, 2024, and February 14, 2024. The first dataset consisted of reports from the ImaGenome dataset, providing reference standard annotations from the MIMIC-CXR database acquired between 2011 and 2016. The second dataset consisted of randomly selected reports created at the Massachusetts General Hospital between July 2019 and July 2021. In both datasets, the commercial models GPT-3.5 Turbo and GPT-4 were compared with open-source models that included Mistral-7B and Mixtral-8 × 7B (Mistral AI), Llama 2-13B and Llama 2-70B (Meta), and Qwen1.5-72B (Alibaba Group), as well as CheXbert and CheXpert-labeler (Stanford ML Group), in their ability to accurately label the presence of multiple findings in radiograph text reports using zero-shot and few-shot prompting. The McNemar test was used to compare F1 scores between models. Results On the ImaGenome dataset (n = 450), the open-source model with the highest score, Llama 2-70B, achieved micro F1 scores of 0.97 and 0.97 for zero-shot and few-shot prompting, respectively, compared with the GPT-4 F1 scores of 0.98 and 0.98 (P > .99 and < .001 for superiority of GPT-4). On the institutional dataset (n = 500), the open-source model with the highest score, an ensemble model, achieved micro F1 scores of 0.96 and 0.97 for zero-shot and few-shot prompting, respectively, compared with the GPT-4 F1 scores of 0.98 and 0.97 (P < .001 and > .99 for superiority of GPT-4). Conclusion Although GPT-4 was superior to open-source models in zero-shot report labeling, few-shot prompting with a small number of example reports closely matched the performance of GPT-4. The benefit of few-shot prompting varied across datasets and models. © RSNA, 2024 Supplemental material is available for this article.

Incidentalomas in Chest Radiograph of Apparently Healthy Individuals Presenting for Pre-Employment Medical Examination in Enugu State, Nigeria.

A chest radiograph is the most commonly requested and performed imaging for diagnosis and screening including medical examination. The study aimed to determine the prevalence and pattern of incidentalomas in radiographs of cases presenting for pre-employment medical examination. This was a retrospective review of chest radiographs conducted for patients presenting for pre-employment medical examination from January to December 2022. Relevant data including sociodemographic characteristics, clinical details and chest radiograph reports were retrieved using study proforma from all patients with complete medical data were analyzed using statistical package for social sciences (SPSS) version 23 with the level of statistical significance set as P <0.05 taken as statistically significant. Three hundred and eleven chest radiographs of subjects aged 20-49 years were evaluated, of which 22 (7.1%) incidentalomas were found. Most were females (55%). Of these incidentalomas, 12 (54.6%) were in the cardiovascular system, while the skeletal and pulmonary systems accounted for eight (36.4%) and two (9.1%), respectively. In all participants, clinically significant findings were eight (2, 6%), comprising of cardiomegaly five (1.6%), leash of vessels in the upper zone one (0.3%), blunt costophrenic angle one (0.3%), and right-sided aortic arch one (0.3%). Age and gender have no statistical significance in the incidental findings. Although the incidence of clinically significant incidentalomas appears low in our environment where beliefs and lack of adequate healthcare financing prevent the majority from seeking timely medical attention, chest radiograph remains an invaluable tool for pre-employment medical examination. Some underlying medical conditions could be picked up, further investigation sought to save life, and it serves as a baseline with which future findings may be compared.

Decoding Radiology Reports: Artificial Intelligence-Large Language Models Can Improve the Readability of Hand and Wrist Orthopedic Radiology Reports.

The purpose of this study was to assess the effectiveness of an Artificial Intelligence-Large Language Model (AI-LLM) at improving the readability of hand and wrist radiology reports. The radiology reports of 100 hand and/or wrist radiographs, 100 hand and/or wrist computed tomography (CT) scans, and 100 hand and/or wrist magnetic resonance imaging (MRI) scans were extracted. The following prompt command was inserted into the AI-LLM: "Explain this radiology report to a patient in layman's terms in the second person: [Report Text]." The report length, Flesch reading ease score (FRES), and Flesch-Kincaid reading level (FKRL) were calculated for the original radiology report and the AI-LLM-generated report. The accuracy of the AI-LLM report was assessed via a 5-point Likert scale. Any "hallucination" produced by the AI-LLM-generated report was recorded. There was a statistically significant improvement in mean FRES scores and FKRL scores in the AI-LLM-generated radiograph report, CT report, and MRI report. For all AI-LLM-generated reports, the mean reading level improved to below an eighth-grade reading level. The mean Likert score for the AI-LLM-generated radiograph report, CT report, and MRI report was 4.1 ± 0.6, 3.9 ± 0.6, and 3.9 ± 0.7, respectively. The hallucination rate in the AI-LLM-generated radiograph report, CT report, and MRI report was 3%, 6%, and 6%, respectively. This study demonstrates that AI-LLM effectively improves the readability of hand and wrist radiology reports, underscoring the potential application of AI-LLM as a promising and innovative patient-centric strategy to improve patient comprehension of their imaging reports.Level of Evidence: IV.

The Epidemiologic and Clinical Features of Radiographic-Confirmed Community-Acquired Pneumonia Among Chinese Children: A Retrospective Hospital-Based Study.

To help understand the disease burden of vaccine-preventable bacterial disease, we delineated the epidemiologic and clinical characteristics of radiographic-confirmed community-acquired pneumonia (CXR-CAP) among Chinese children. We retrospectively screened the electronic database of the hospital information system to identify all pediatric CAP cases admitted to the Children's Hospital of Soochow University between 2010 and 2014. Radiographic findings and clinical data were extracted from the medical charts through individual chart reviews. CXR-CAP cases were defined as the presence of consolidation or pleural effusion noted on chest radiograph reports. We employed a multivariate logistic regression model to identify the potential risk factors associated with CXR-CAP. Among the 27,485 hospitalized CAP cases with radiologic data, 6322 (23.00%) were identified as CXR-CAP cases, while 21,163 (77.00%) were categorized as non-CXR-CAP cases. Children with CXR-CAP were notably older than those without CXR-CAP (non-CXR-CAP; χ2 = 1313.22; P < 0.01). CXR-CAP cases exhibited a higher rate of intensive care unit admission (3.55% vs. 1.94%; P < 0.01), extended hospital stays (73.87% vs. 63.79%; P < 0.01) and increased mortality rates (0.19% vs. 0.04%; P < 0.01). The factors associated with CXR-CAP included age (>12 months), season (summer and autumn), fever, abnormal breath sounds, C-reactive protein (>8 mg/L) and alanine transaminase (>40 U/L). CXR-CAP cases consisted of a substantial proportion of hospitalized patients with CAP and had more severe clinical manifestations than in-patients without CXR-CAP among Chinese children.

Optimal Large Language Model Characteristics to Balance Accuracy and Energy Use for Sustainable Medical Applications.

Background Large language models (LLMs) for medical applications use unknown amounts of energy, which contribute to the overall carbon footprint of the health care system. Purpose To investigate the tradeoffs between accuracy and energy use when using different LLM types and sizes for medical applications. Materials and Methods This retrospective study evaluated five different billion (B)-parameter sizes of two open-source LLMs (Meta's Llama 2, a general-purpose model, and LMSYS Org's Vicuna 1.5, a specialized fine-tuned model) using chest radiograph reports from the National Library of Medicine's Indiana University Chest X-ray Collection. Reports with missing demographic information and missing or blank files were excluded. Models were run on local compute clusters with visual computing graphic processing units. A single-task prompt explained clinical terminology and instructed each model to confirm the presence or absence of each of the 13 CheXpert disease labels. Energy use (in kilowatt-hours) was measured using an open-source tool. Accuracy was assessed with 13 CheXpert reference standard labels for diagnostic findings on chest radiographs, where overall accuracy was the mean of individual accuracies of all 13 labels. Efficiency ratios (accuracy per kilowatt-hour) were calculated for each model type and size. Results A total of 3665 chest radiograph reports were evaluated. The Vicuna 1.5 7B and 13B models had higher efficiency ratios (737.28 and 331.40, respectively) and higher overall labeling accuracy (93.83% [3438.69 of 3665 reports] and 93.65% [3432.38 of 3665 reports], respectively) than that of the Llama 2 models (7B: efficiency ratio of 13.39, accuracy of 7.91% [289.76 of 3665 reports]; 13B: efficiency ratio of 40.90, accuracy of 74.08% [2715.15 of 3665 reports]; 70B: efficiency ratio of 22.30, accuracy of 92.70% [3397.38 of 3665 reports]). Vicuna 1.5 7B had the highest efficiency ratio (737.28 vs 13.39 for Llama 2 7B). The larger Llama 2 70B model used more than seven times the energy of its 7B counterpart (4.16 kWh vs 0.59 kWh) with low overall accuracy, resulting in an efficiency ratio of only 22.30. Conclusion Smaller fine-tuned LLMs were more sustainable than larger general-purpose LLMs, using less energy without compromising accuracy, highlighting the importance of LLM selection for medical applications. © RSNA, 2024 Supplemental material is available for this article.

The challenges of installing Picture Archiving and Communication Systems (PACS) in a low- or poor-income resource nation: Our experience in a teaching hospital in Sub-Saharan Africa

The practice of radiology as a subspecialty of medicine has undergone remarkable advancement as a result of picture archiving and communication Systems(PACS) technology. This technology allows for storage and retrieval of radiographic images and reports at any time, and possible transfer to other locations for secondary viewing and opinions. PACS enables healthcare professionals to access and review medical images quickly, allowing for faster and more accurate diagnosis and treatment planning. However, there are challenges involved in its installation and operations in our health facility, a teaching hospital in Sub-Saharan Africa. These challenges range from lack of huge financial resources required to acquire PACS equipment and accessories to lack of steady and regular electric power supply. Other challenges include inadequate dedicated space for its installation, inadequate functional cooling systems for its sensitive components, inadequate internet services for end users and paucity of qualified and trained personnel. There is, therefore, need for increased government and private sector funding in order to surmount these challenges, thereby enhancing health promotion in Sub-Saharan Africa.

Nonsteroidal anti-inflammatory drug-induced small bowel strictures (diaphragm disease) – an under-recognized cause of small bowel obstruction

BackgroundSmall bowel obstruction is a major source of morbidity and mortality that carries a significant economic burden. Recurrent small bowel obstruction may be secondary to circumferential strictures (small bowel diaphragm disease), an under-recognized entity secondary to long-term NSAID use. We aim to describe the sensitivity of preoperative CT enterography in patients with surgically treated small bowel diaphragm disease. MethodsWe retrospectively reviewed adult patients who underwent elective small bowel resection for small bowel obstruction performed by a single minimally-invasive surgeon between 2010-2023. Patient history, radiographic, endoscopic, operative, and pathology reports were reviewed for reference to NSAID use, small bowel strictures, diaphragms, and enteropathy. Exclusion criteria were prior radiation, inflammatory bowel disease, malignancy, adhesive disease, and anastomotic strictures. ResultsA total of 225 patients were identified; of those, 22 patients (10%) met inclusion criteria. The mean age was 60.7 years (range 29-78), with 15 women (68%). All patients underwent minimally invasive small bowel resection for obstruction with histopathologic evidence of stricture without evidence of transmural inflammation, granuloma, or dysplasia and confirmed NSAID use (n=22; 100%). Anemia was present in 36% (n=8). Preoperative CT or MR enterography was performed in 18 patients (82%) and stricturing was reported in 13 patients (72%). Intraoperatively, palpation identified strictures in all patients. ConclusionNSAID-induced small bowel injury is an under-recognized condition that, in severe cases, can present as small bowel obstruction. Surgeons should consider diaphragm disease in patients with obstruction and NSAID use, in which preoperative CT or MR enterography may be useful but cannot rule out disease.

Abstract PO4-12-06: Interstitial Lung Disease (ILD) in Sequential Antibody Drug Conjugate (ADC) use in patients with Metastatic Breast Cancer (MBC): A Multi-Institutional experience

Abstract Introduction: Interstitial Lung Disease (ILD) is a known adverse event associated with Antibody Drug Conjugates (ADCs), causing fibrosis of the lung. In the DESTINY-Breast 04, TROPICS-02, and ASCENT trials the rate of ILD/Pneumonitis was 12.1%, 0.4%, and 0% respectively. In these studies, the time to first cross-sectional imaging after initiation of ADC was 6 weeks. However, prior treatment with a topoisomerase-1 inhibitor ADC was not permitted in these studies, so the influence of sequential ADC treatments remains unknown. Here, we aimed to characterize the frequency and management of ILD in sequential use of ADCs in a multi-institution retrospective analysis. Methods: We identified patients (pts) with HR+/HER2-low and HR-/HER2-low MBC who had received both Trastuzumab-deruxtecan (T-DXd) and Sacituzumab govitecan (SG) monotherapy (in either order, with or without intervening therapies) who were treated at five academic centers between 2020-2023. Patients had received treatment with the ADC in either routine clinical treatment or as a prior participant on a clinical trial with ADC monotherapy. Information regarding ILD diagnosis, imaging, timing of onset, peak and resolution along with management information was collected and analyzed. Retrospective assessment of ILD toxicity was graded using CTCAE v5.0, based on available clinical and radiographic reports. Results: 10/60 pts who received ADC therapy in this multi-institutional cohort were diagnosed with ILD (all grades). No differences in age or BMI were observed in those who developed ILD versus those who did not, and 9/10 pts were White, 1 was Asian, with a median age of 59 years (47-79 years). Pertinent history in these pts included 3/10 had a prior smoking history, 8/10 had a prior cardiac or pulmonary comorbidity (including sleep apnea, lung cancer, hypertension, hyperlipidemia, thrombosis, or stroke). Prior to receiving their first ADC, 7/10 patients had metastatic disease to the lung. All 10 patients were diagnosed with ILD during treatment with T-DXd. 4 pts initially received T-DXd, then developed ILD and then upon recovery, subsequently received SG. The remaining 6 pts initially received SG (with no reports of ILD), and then received T-DXd and subsequently developed ILD. In the entire cohort, the median time from ADC initiation to first CT scan was 8 weeks (range 1-22 weeks). Within the cohort that developed ILD, the median time from initiation of T-DXd to first CT scan was also 8 weeks (range 6-13 weeks). The median time to diagnoses of ILD was 3 months after initiation of T-DXd (range 0.4-14.5 months). Dyspnea was the most common presenting symptom in 9/10 pts and at diagnosis of ILD, dyspnea was graded as III in 6/10 pts, II in 3/10 pts, and I in 1/10 pts. Diagnostic evaluation included CT chest (9/10), pulmonary consultation (8/10), pulmonary function testing (2/10), and bronchoscopy (1/10). 4/10 patients presented at their highest-grade toxicity at diagnosis. All patients stopped T-DXd treatment, 7 were hospitalized, 7 required supplemental oxygen, and 8 were prescribed steroids for treatment. Of those receiving steroids, the most common starting dose was prednisone 1 mg/kg and doses ranged from 0.5 mg/kg to 9mg/kg prednisone equivalents. No patient received a secondary immunosuppressive agent. The steroid treatment duration ranged from 1-4 months. None of the 10 pts underwent re-challenge with T-DXd and 3/10 developed grade V ILD (death). Conclusion: We observed a 16.6% incidence of all grade ILD (5% grade V ILD) with T-DXd and 0% with SG. In this small cohort study, we did not observe a clear increased risk for ILD with sequential treatment of ADCs. Practice patterns were consistent with recommended diagnostic evaluation and steroid management. Additional strategies are needed to characterize pts most at risk for developing ILD and to identify situations where re-challenge may be safe. Citation Format: Sarah Premji, Laura Huppert, Reshma Mahtani, Samantha Fisch, Naomi Dempsey, Angelina Raimonde-Taylor, Saya Jacob, Laura Quintal, Jo Chien, Michelle Melisko, Ana Sandoval-Leon, Lauren Carcas, Manmeet Ahluwalia, Natasha Harpalani, Jenna Hoppenworth, Dame Idossa, Ruta Rao, Hope Rugo, Karthik Giridhar. Interstitial Lung Disease (ILD) in Sequential Antibody Drug Conjugate (ADC) use in patients with Metastatic Breast Cancer (MBC): A Multi-Institutional experience [abstract]. In: Proceedings of the 2023 San Antonio Breast Cancer Symposium; 2023 Dec 5-9; San Antonio, TX. Philadelphia (PA): AACR; Cancer Res 2024;84(9 Suppl):Abstract nr PO4-12-06.

Diagnostic correlation between plain radiographic and histopathology reports among patients with long bones tumours treated at a Tertiary Hospital in Tanzania

Background: Bone tumours are significant global health concern, with a rising number of new cases being diagnosed each year. The incidence of bone tumours has shown a significant increase in recent years, as evidenced by a 50% rise in new cases in 2020 alone. Early and accurate diagnosis is crucial for effective treatment and improved patient outcomes. The diagnosis of primary bone tumours is based on clinical, radiological and histopathological findings. Objective: This study aimed to determine the diagnostic correlation between plain radiographic and histopathological reports in diagnosing long bone tumours. Methods: A cross-sectional study was conducted between 2018 and 2021. Participants who underwent a histological examination for long bone tumours and met the inclusion criteria were enrolled. Relevant information was obtained from patient records then sensitivity, specificity, and accuracy of the plain radiograph were calculated with histopathology serving as the gold standard. Results: The study included 83 participants, the majority were aged 30 years old or younger. Forty six percent of the cases were malignant tumours while benign were 54%. Osteosarcoma was found to be the most common tumour type both by radiological and histological diagnosis and femur was the most commonly affected bone. Plain radiograph was found to have a sensitivity of 92.1%, specificity of 73.3% and accuracy of 82%. Conclusion: This study found a strong diagnostic correlation between plain radiographic and histopathology reports in patients with long bone tumours. This suggests that plain radiography can serve as a useful screening tool for long bone tumours in resource-limited settings, allowing for early and accurate preliminary diagnoses.

Serious bacterial infection risk in recently immunized febrile infants in the emergency department

Study objectivesFever following immunizations is a common presenting chiefcomplaint among infants. The 2021 American Academy of Pediatrics (AAP) febrile infant clinical practice guidelines exclude recently immunized (RI) infants. This is a challenge for clinicians in the management of the febrile RI young infant. The objective of this study was to assess the prevalence of SBI in RI febrile young infants between 6 and 12 weeks of age. MethodsThis was a retrospective chart review of infants 6–12 weeks who presented with a fever ≥38 °C to two U.S. military academic Emergency Departments over a four-year period. Infants were considered recently immunized (RI) if they had received immunizations in the preceding 72 h prior to evaluation and not recently immunized (NRI) if they had not received immunizations during this time period. The primary outcome was prevalence of serious bacterial infection (SBI) further delineated into invasive-bacterial infection (IBI) and non-invasive bacterial infection (non-IBI) based on culture and/or radiograph reports. ResultsOf the 508 febrile infants identified, 114 had received recent immunizations in the preceding 72 h. The overall prevalence of SBI was 11.4% (95% CI = 8.9–14.6) in our study population. The prevalence of SBI in NRI infants was 13.7% (95% CI = 10.6–17.6) compared to 3.5% (95% CI = 1.1–9.3) in RI infants. The relative risk of SBI in the setting of recent immunizations was 0.3 (95% CI = 0.1–0.7). There were no cases of invasive-bacterial infections (IBI) in the RI group with all but one of the SBI being urinary tract infections (UTI). The single non-UTI was a case of pneumonia in an infant who presented with respiratory symptoms within 24 h of immunizations. ConclusionThe risk of IBI (meningitis or bacteremia) in RI infants aged 6 to 12 weeks is low. Non-IBI within the first 24 h following immunization was significantly lower than in febrile NRI infants. UTIs remain a risk in the RI population and investigation with urinalysis and urine culture should be encouraged. Shared decision making with families guide a less invasive approach to the care of these children. Future research utilizing a large prospective multi-center data registry would aid in further defining the risk of both IBI and non-IBI among RI infants.

The Gartland classification for expediting supracondylar humerus fracture triage: a collaborative approach to structured reporting between pediatric radiologists and orthopedists

BackgroundThe modified Gartland classification is an important tool for evaluation of pediatric supracondylar humerus fractures (SCHF) because it can direct treatment decisions. Gartland type I can be managed outpatient, while emergent surgical consult occurs with type II and III. This study assesses the interobserver reliability of the Gartland classification between pediatric radiologists and orthopedic providers. MethodsA retrospective review of 320 children diagnosed with a SCHF at a single tertiary children's hospital during 2022 was conducted. The Gartland classification documented in the radiographic report by a pediatric radiologist and the classification documented in the first encounter with an orthopedic provider was collected. Kappa value was used to assess interobserver reliability of classifications between radiologists and orthopedic providers. A second group of 76 Gartland type I SCHF from 2015, prior to our institution's implementation of structured reporting, was reviewed for comparison of unnecessary orthopedic consults at initial presentation. ResultsThe Gartland classification has excellent interobserver reliability between radiologists and orthopedic providers with 90 % (289/320) agreement and kappa of 0.854 (confidence interval: 0.805–0.903). The most frequent disagreement that occurred was fractures classified as type II by radiology and type III by orthopedics. There were similar rates of consults for the 2015 and 2022 cohorts (p = 0.26). ConclusionThe Gartland classification system is a reliable and effective tool for communication between radiologists and orthopedic providers. Implementing a structured reporting system has the potential to improve triage efficiency for SCHF.

The prognosis for return to athletic function for Thoroughbred racehorses in Hong Kong with injuries to the palmaroproximal aspect of the metacarpus diagnosed using low-field magnetic resonance imaging.

To evaluate and compare the prognosis for Thoroughbred racehorses to return to galloping and racing with injuries to the palmaroproximal metacarpus diagnosed with MRI. 29 flat racing Thoroughbreds at the Hong Kong Jockey Club that underwent MRI between 2014 and 2022. Clinical, radiographic, ultrasonographic, and MRI reports were collected from veterinary clinical records, and these were combined with training and racing data. Horses were categorized on the basis of MRI diagnosis: (1) proximal suspensory ligament (PSL) involvement only, (2) PSL and concurrent proximal third metacarpal (MC3) bone involvement, and (3) proximal MC3 bone involvement only. The following were compared for prognosis for return to athletic function: return to galloping or racing, and reinjury. Overall, the prognosis for return to athletic function was fair, with 92% (22/24; P = .53) and 67% (16/24; P = .73) of horses returning to galloping and racing, respectively. There was a relatively low reinjury rate, with 18% (4/22) of horses reinjuring. Horses with concurrent injury to both the PSL and proximal MC3 bone (Category 2) took longer to return to gallop (median, 116; IQR, 100.5 to 160), when compared with horses having only PSL injury (median, 69; IQR, 43 to 80; P = .04). Of the 4 horses that reinjured, 3 (75%) were horses in Category 2. The findings from this study suggest that injuries involving both PSL and proximal MC3 bone concurrently require a longer rehabilitation period than those with PSL involvement alone.

The prevalence of bilateral and ipsilateral radiographic osteoarthritis is high in White, Asian and Native Hawaiian/Pacific Islanders presenting for unilateral knee or hip arthroplasty.

It is estimated that one-third of patients presenting with unilateral joint pain have contralateral osteoarthritis (OA) at first presentation. Most studies have primarily examined White patient cohorts. The purpose of this study was to determine the prevalence of contralateral joint OA for patients presenting for unilateral total knee (TKA), unicompartmental knee (UKA) or total hip arthroplasty (THA) among Asian, Native Hawaiian/Pacific Islander and White patients. Bilateral radiographic reports at initial presentation of 2,312 subjects who underwent unilateral arthroplasties (332 UKAs, 933 TKAs and 1,047 THAs) were reviewed. The presence of contralateral OA was recorded and compared by racial group and type of arthroplasty performed. Parametric statistical analyses were performed to determine differences between groups. Multivariable analyses were completed for each arthroplasty group to determine the influence on the presence of contralateral OA, presented as odds ratios and 95% confidence intervals. Contralateral joint OA was present in 86.7%, 90.4% and 70.4% of UKA, TKA and THA patients, respectively. Concurrent hip OA was present in 41.6% and 59.5% of UKA and TKA patients. No racial differences in the prevalence of contralateral knee OA were found for knee arthroplasty patients. White patients (74.6%) had a greater prevalence of contralateral hip OA compared to Asians (66.5%, p = 0.037) amongst THA recipients. Increased age and body mass index were significantly associated with the presence of contralateral knee OA. Increased age, being male and being White were significant contributors for the presence of contralateral hip OA. The prevalence of contralateral joint OA and concurrent hip OA is high in all three racial groups. Due to the extensive prevalence of contralateral and concurrent knee and hip OA, bilateral radiographic evaluation should be considered for all patients presenting with unilateral hip or knee pain due to OA.

Positive Effect of a Financial Incentive on Radiologist Compliance With Quality Metric Placement in Knee Radiography Reports

PurposeOngoing quality improvement (QI) processes in the authors’ department include the insertion of a Kellgren-Lawrence (KL) osteoarthritis grading template in knee radiography reports to decrease unnecessary MRI. However, uniform adoption of this grading system is lacking. Department-wide financial incentives were instituted to improve compliance with QI metrics. The purpose of this study was to evaluate the effect of a financial incentive on KL grading system use and to compare compliance rates of musculoskeletal (MSK) radiologists with those of general radiologists who were not financially incentivized to use KL grading. MethodsPercentages of all knee radiography reports containing KL grading with standardized follow-up recommendations were determined by querying the departmental radiology database before and after the introduction of the new quality-based financial incentive. Preincentive compliance rates for MSK and general radiologists were compared with an adoption period and two separate 6-month postincentive periods. ResultsIn total, 52,673 reports were retrospectively analyzed for KL grading use (41,670 reports interpreted by MSK radiologists and 11,003 interpreted by general radiologists). Increase in compliance was greatest among MSK radiologists’ reports during the incentivized adoption period (from 36.1% to 53.2%). This improvement was sustained among MSK radiologists and averaged 62.7% during the most recently studied postimplementation period. A lesser degree of improvement in compliance was observed in nonincentivized general radiologists’ reports (from 19.3% to 27.5%); during the postimplementation follow-up period, their compliance decreased to 26.5%. ConclusionsThe introduction of a financial incentive resulted in significantly increased adoption of QI practices with sustained improvement among incentivized MSK radiologists compared with nonincentivized general radiologists.

Fully automatic summarization of radiology reports using natural language processing with large language models

PurposeNatural language processing using language models has yielded promising results in various fields. Language models can help improve the workflow of radiologists. This retrospective study aimed to construct and evaluate language models for automatic summarization of radiology reports. MethodsTwo radiology report datasets from the MIMIC Chest X-ray (MIMIC-CXR) database and the Japan Medical Image Database (JMID) were included in this study. The MIMIC-CXR is an open database comprising chest radiograph reports. The JMID is a large database comprising computed tomography and magnetic resonance imaging reports from 10 academic medical centers in Japan. A total of 128,032 and 1,101,271 reports were included in this study from the MIMIC-CXR database and JMID, respectively. Four Text-to-Text Transfer Transformer (T5) models were constructed. Recall-Oriented Understudy for Gisting Evaluation (ROUGE), a quantitative metric, was used to evaluate the quality of the text summarized from 19,205 and 58,043 test sets from the MIMIC-CXR and JMID, respectively. The Wilcoxon signed-rank test was used to evaluate the differences among the ROUGE values of the four T5 models. Moreover, the subsets of automatically summarized text in the test sets were manually evaluated by two radiologists. The best T5 models were selected for automatic summarization using the Wilcoxon signed-rank test. ResultsThe quantitative metrics of the best T5 models were as follows: ROUGE-1 = 57.75 ± 30.99, ROUGE-2 = 49.96 ± 35.36, and ROUGE-L = 54.07 ± 32.48 in the MIMIC-CXR; and ROUGE-1 = 50.00 ± 29.24, ROUGE-2 = 39.66 ± 30.21, and ROUGE-L = 47.87 ± 29.44 in the JMID. The radiologists’ evaluations revealed 86% and 85% of the texts automatically summarized from the MIMIC-CXR and JMID, respectively, to be clinically useful. ConclusionThe T5 models constructed in this study were able to perform automatic summarization of the radiology reports. The radiologists’ evaluations demonstrated most of the automatically summarized texts to be clinically valuable.

MIMIC-CXR Radyoloji Raporlarının DDİ Yöntemleriyle Anlamsal ve Yapısal Analizi

Artificial intelligence that aims to imitate human decision-making processes, using human knowledge as a foundation, is a critical research area with various practical applications in different disciplines. In the health domain, machine learning and image processing techniques are increasingly being used to assist in diagnosing diseases using laboratory results, findings, MRI, tomography, or radiology images, and etc. However, many healthcare reports, such as epicrisis summaries prepared by clinical experts, contain crucial and valuable information. In addition to information extraction from healthcare reports, applications such as automatic healthcare report generation are among the natural language processing research areas based on this knowledge and experience. The primary goals are to reduce the workload of clinical experts, minimize the likelihood of errors, and save time to speed up the diagnosis process. The MIMIC-CXR dataset is a huge dataset consisting of chest radiographs and reports prepared by radiology experts related to these images. This study focuses on the structural and semantic analysis of MIMIC-CXR radiography reports. Before developing a natural language processing-based model, preprocessing steps were applied to the dataset, and the results of syntactic and semantic analyses performed on unstructured report datasets are presented. This study is expected to provide insights for developing language models, particularly for developing a natural language processing model on the MIMIC-CXR dataset.

Turning radiology reports into epidemiological data to track seasonal pulmonary infections and the COVID-19 pandemic

ObjectivesTo automatically label chest radiographs and chest CTs regarding the detection of pulmonary infection in the report text, to calculate the number needed to image (NNI) and to investigate if these labels correlate with regional epidemiological infection data.Materials and methodsAll chest imaging reports performed in the emergency room between 01/2012 and 06/2022 were included (64,046 radiographs; 27,705 CTs). Using a regular expression-based text search algorithm, reports were labeled positive/negative for pulmonary infection if described.Data for regional weekly influenza-like illness (ILI) consultations (10/2013–3/2022), COVID-19 cases, and hospitalization (2/2020–6/2022) were matched with report labels based on calendar date. Positive rate for pulmonary infection detection, NNI, and the correlation with influenza/COVID-19 data were calculated.ResultsBetween 1/2012 and 2/2020, a 10.8–16.8% per year positive rate for detecting pulmonary infections on chest radiographs was found (NNI 6.0–9.3). A clear and significant seasonal change in mean monthly detection counts (102.3 winter; 61.5 summer; p < .001) correlated moderately with regional ILI consultations (weekly data r = 0.45; p < .001).For 2020–2021, monthly pulmonary infection counts detected by chest CT increased to 64–234 (23.0–26.7% per year positive rate, NNI 3.7–4.3) compared with 14–94 (22.4–26.7% positive rate, NNI 3.7–4.4) for 2012–2019. Regional COVID-19 epidemic waves correlated moderately with the positive pulmonary infection CT curve for 2020–2022 (weekly new cases: r = 0.53; hospitalizations: r = 0.65; p < .001).ConclusionText mining of radiology reports allows to automatically extract diagnoses. It provides a metric to calculate the number needed to image and to track the trend of diagnoses in real time, i.e., seasonality and epidemic course of pulmonary infections.Clinical relevanceDigitally labeling radiology reports represent previously neglected data and may assist in automated disease tracking, in the assessment of physicians’ clinical reasoning for ordering radiology examinations and serve as actionable data for hospital workflow optimization.Key Points• Radiology reports, commonly not machine readable, can be automatically labeled with the contained diagnoses using a regular-expression based text search algorithm.• Chest radiograph reports positive for pulmonary infection moderately correlated with regional influenza-like illness consultations (weekly data; r = 0.45; p < .001) and chest CT reports with the course of the regional COVID-19 pandemic (new cases: r = 0.53; hospitalizations: r = 0.65; p < 0.001).• Rendering radiology reports into data labels provides a metric for automated disease tracking, the assessment of ordering physicians clinical reasoning and can serve as actionable data for workflow optimization.