Radiology Reports Research Articles

Imaging findings of gastric glomus tumors.

Sparse literature describes the imaging findings of gastric glomus tumors (GGT), focusing on benign tumors. We are not aware of prior radiology reports on malignant GGT. The aim of the study was to determine whether it is possible to differentiate between benign and malignant GGT on CT or MR. Institutional radiology and pathology databases were queried for the diagnosis of GGT between January 2010 to December 2023. Of 22 identified subjects, five were excluded due to non-availability of preoperative CT or MR images and three due to lack of pathological confirmation in our institution. The study cohort comprised of 14 patients (males = 6) with median age of 65 years (range 31 to 79 years). Two abdominal radiologists in consensus reviewed all relevant CT and MR images. There were 10 benign and 4 malignant GGT. Benign tumors were smaller than malignant GGT (median size of 2.0cm vs. 5.3cm, p = 0.03), more likely to exhibit homogeneous hyperenhancement (9/10 vs. 0/4, p < 0.01), and demonstrated intramural rather than exophytic growth. There was no substantial difference in T2 signal or diffusion restriction between benign and malignant GGT. On follow up, benign GGT were essentially stable in size, while malignant GGT grew. A biopsy proven GGT larger than 5cm or showing necrosis is likely to be malignant. This is important since preoperative endoscopic ultrasound-guided fine needle aspiration may be indeterminate for malignancy.

Deep Learning for Automated Classification of Hip Hardware on Radiographs.

To develop a deep learning model for automated classification of orthopedic hardware on pelvic and hip radiographs, which can be clinically implemented to decrease radiologist workload and improve consistency among radiology reports. Pelvic and hip radiographs from 4279 studies in 1073 patients were retrospectively obtained and reviewed by musculoskeletal radiologists. Two convolutional neural networks, EfficientNet-B4 and NFNet-F3, were trained to perform the image classification task into the following most represented categories: no hardware, total hip arthroplasty(THA), hemiarthroplasty, intramedullary nail, femoral neck cannulated screws, dynamic hip screw, lateral blade/plate, THA with additional femoral fixation, and post-infectious hip. Model performance was assessed on an independent test set of 851 studies from 262 patients and compared to individual performance of five subspecialty-trained radiologists using leave-one-out analysis against an aggregate gold standard label. For multiclass classification, the area under the receiver operating characteristic curve (AUC) for NFNet-F3 was 0.99 or greater for all classes, and EfficientNet-B4 0.99 or greater for all classes except post-infectious hip, with an AUC of 0.97. When compared with human observers, models achieved an accuracy of 97%, which is non-inferior to four out of five radiologists and outperformed one radiologist. Cohen's kappa coefficient for both models ranged from 0.96 to 0.97, indicating excellent inter-reader agreement. A deep learning model can be used to classify a range of orthopedic hip hardware with high accuracy and comparable performance to subspecialty-trained radiologists.

AI implementation: Radiologists' perspectives on AI-enabled opportunistic CT screening

ObjectiveAI adoption requires perceived value by end-users. AI-enabled opportunistic CT screening (OS) detects incidental clinically meaningful imaging risk markers on CT for potential preventative health benefit. This investigation assesses radiologists' perspectives on AI and OS. MethodsAn online survey was distributed to 7500 practicing radiologists among ACR membership of which 4619 opened the emails. Familiarity with and views of AI applications were queried and tabulated, as well as knowledge of OS and inducements and impediments to use. ResultsRespondent (n = 211) demographics: mean age 55 years, 73 % male, 91 % diagnostic radiologists, 46 % in private practice. 68 % reported using AI in practice, while 52 % were only somewhat familiar with AI. 70 % viewed AI positively though only 46 % reported AI's overall accuracy met expectations. 57 % were unfamiliar with OS, with 52 % of those familiar having a positive opinion. Patient perceptions were the most commonly reported (25 %) inducement for OS use. Provider (44 %) and patient (40 %) costs were the most common impediments. Respondents reported that osteoporosis/osteopenia (81 %), fatty liver (78 %), and atherosclerotic cardiovascular disease risk (76 %) could be well assessed by OS. Most indicated OS output requires radiologist oversight/signoff and should be included in a separate “screening” section in the Radiology report. 28 % indicated willingness to spend 1–3 min reviewing AI-generated output while 18 % would not spend any time. Society guidelines/recommendations were most likely to impact OS implementation. DiscussionRadiologists' perspectives on AI and OS provide practical insights on AI implementation. Increasing end-user familiarity with AI-enabled applications and development of society guidelines/recommendations are likely essential prerequisites for Radiology AI adoption.

FP3.13 - Acute Complicated Diverticulitis: Colorectal Cancer in Disguise? Five Years of Follow-up Practice at a 600-bed Hospital

Abstract Aims Complicated diverticular disease is associated with colonic neoplasia rates of 7.8% to 10.9%. ACPGBI guidelines recommend optical or virtual colonoscopy six weeks following an attack of acute complicated diverticulitis unless performed within the preceding two years. It has been posited that flexible sigmoidoscopy may be sufficient to rule out malignancy given that diverticulitis predominantly affects the sigmoid colon. Five years of practice at a 600-bed hospital was audited to the ACPGBI standard and malignancy rates and locations investigated. Methods All emergency admissions coded for acute complicated diverticulitis from January 2017 to December 2022 were included. Patient electronic records were reviewed retrospectively for demographic data, operation notes, as well as radiology, endoscopy and histology reports. Results A total of 197 patients were included in the study period. 97 cases underwent emergency or semi-elective resection within one year. 8 patients had colonic visualisation in the two years prior to their attack, leaving 92 cases in whom colonic visualisation was indicated. 44 patients had colonoscopy, 22 had flexible sigmoidoscopy, 2 had CT colonography and 24 had no further imaging. 7 malignancies were identified, 5 from resection specimens and 2 from endoscopy. All tumours were in sigmoid colon. Conclusions The ACPGBI standard was met in 50% (46/92) of cases. If flexible sigmoidoscopy were deemed acceptable this would increase to 74% (68/92). Malignancy was identified in 4% (7/197) of cases. All malignancies were in the range of flexible sigmoidoscopy, suggesting full colonoscopy may be superfluous. Less extensive endoscopy may reduce procedure duration, bowel preparation and patient discomfort.

Prospective Evaluation of Artificial Intelligence Triage of Intracranial Hemorrhage on Noncontrast Head CT Examinations.

Background: Retrospective studies evaluating artificial intelligence (AI) algorithms for intracranial hemorrhage (ICH) detection on noncontrast CT (NCCT) have shown promising results but lack prospective validation. Objective: To evaluate the impact on radiologists' real-world aggregate performance for ICH detection and report turnaround times for ICH-positive examinations of a radiology department's implementation of an AI triage and notification system for ICH detection on head NCCT examinations. Methods: This prospective single-center study included adult patients who underwent head NCCT examinations from May 12, 2021 to June 30, 2021 (phase 1) or September 30, 2021 to December 4, 2021 (phase 2). Before phase 1, the radiology department implemented a commercial AI triage system for ICH detection that processed head NCCT examinations and notified radiologists of positive results through a widget with a floating pop-up display. Examinations were interpreted by neuroradiologists or emergency radiologists, who evaluated examinations without and with AI assistance in phase 1 and phase 2, respectively. A panel of radiologists conducted a review process for all examinations with discordance between the radiology report and AI and a subset of remaining examinations, to establish the reference standard. Diagnostic performance and report turnaround times were compared using Pearson chi-square test and Wilcoxon rank-sum test, respectively. Bonferroni correction was used to account for five diagnostic performance metrics (adjusted significance threshold, .01 [α=.05/5]). Results: A total of 9954 examinations from 7371 patients (mean age, 54.8±19.8 years; 3773 female, 3598 male) were included. In phases 1 and 2, 19.8% (735/3716) and 21.9% (1368/6238) of examinations, respectively, were positive for ICH (P=.01). Radiologists without versus with AI showed no significant difference in accuracy (99.5% vs 99.2%), sensitivity (98.6% vs 98.9%), PPV (99.0% vs 99.7%), or NPV (99.7% vs 99.7%) (all P>.01); specificity was higher for radiologists without than with AI (99.8% vs 99.3%, respectively, P=.004). Mean report turnaround time for ICH-positive examinations was 147.1 minutes without AI versus 149.9 minutes with AI (P=.11). Conclusion: An AI triage system for ICH detection did not improve radiologists' diagnostic performance or report turnaround times. Clinical Impact: This large prospective real-world study does not support use of AI assistance for ICH detection.

Lung Cancer Staging Using Chest CT and FDG PET/CT Free-Text Reports: Comparison Among Three ChatGPT Large-Language Models and Six Human Readers of Varying Experience.

Background: Although radiology reports are commonly used for lung cancer staging, this task can be challenging given radiologists' variable reporting styles as well as reports' potentially ambiguous and/or incomplete staging-related information. Objective: To compare performance of ChatGPT large-language models (LLMs) and human readers of varying experience in lung cancer staging using chest CT and FDG PET/CT free-text reports. Methods: This retrospective study included 700 patients (mean age, 73.8±29.5 years; 509 male, 191 female) from four institutions in Korea who underwent chest CT or FDG PET/CT for non-small cell lung cancer initial staging from January, 2020 to December, 2023. Examinations' reports used a free-text format, written exclusively in English or in mixed English and Korean. Two thoracic radiologists in consensus determined the overall stage group (IA, IB, IIA, IIB, IIIA, IIIB, IIIC, IVA, IVB) for each report using the AJCC 8th-edition staging system, establishing the reference standard. Three ChatGPT models (GPT-4o, GPT-4, GPT-3.5) determined an overall stage group for each report using a script-based application programming interface, zero-shot learning, and prompt incorporating a staging system summary. Six human readers (two fellowship-trained radiologists with lesser experience than the radiologists who determined the reference standard, two fellows, two residents) also independently determined overall stage groups. GPT-4o's overall accuracy for determining the correct stage among the nine groups was compared with that of the other LLMs and human readers using McNemar tests. Results: GPT-4o had an overall staging accuracy of 74.1%, significantly better than the accuracy of GPT-4 (70.1%, p=.02), GPT-3.5 (57.4%, p<.001), and resident 2 (65.7%, p<.001); significantly worse than the accuracy of fellowship-trained radiologist 1 (82.3%, p<.001) and fellowship-trained radiologist 2 (85.4%, p<.001); and not significantly different from the accuracy of fellow 1 (77.7%, p=.09), fellow 2 (75.6%, p=.53), and resident 1 (72.3%, p=.42). Conclusions: The best-performing model, GPT-4o, showed no significant difference in staging accuracy versus fellows, but significantly worse performance versus fellowship-trained radiologists. The findings do not support use of LLMs for lung cancer staging in place of expert healthcare professionals. Clinical Impact: The findings indicate the importance of domain expertise for performing complex specialized tasks such as cancer staging.

Evaluation of guided reporting: quality and reading time of automatically generated radiology report in breast magnetic resonance imaging using a dedicated software solution.

Unstructured, free-text dictation (FT), the current standard in breast magnetic resonance imaging (MRI) reporting, is considered time-consuming and prone to error. The purpose of this study is to assess the usability and performance of a novel, software-based guided reporting (GR) strategy in breast MRI. Eighty examinations previously evaluated for a clinical indication (e.g., mass and focus/non-mass enhancement) with FT were reevaluated by three specialized radiologists using GR. Each radiologist had a different number of cases (R1, n = 24; R2, n = 20; R3, n = 36). Usability was assessed by subjective feedback, and quality was assessed by comparing the completeness of automatically generated GR reports with that of their FT counterparts. Errors in GR were categorized and analyzed for debugging with a final software version. Combined reading and reporting times and learning curves were analyzed. Usability was rated high by all readers. No non-sense, omission/commission, or translational errors were detected with the GR method. Spelling and grammar errors were observed in 3/80 patient reports (3.8%) with GR (exclusively in the discussion section) and in 36/80 patient reports (45%) with FT. Between FT and GR, 41 patient reports revealed no content differences, 33 revealed minor differences, and 6 revealed major differences that resulted in changes in treatment. The errors in all patient reports with major content differences were categorized as content omission errors caused by improper software operation (n = 2) or by missing content in software v. 0.8 displayable with v. 1.7 (n = 4). The mean combined reading and reporting time was 576 s (standard deviation: 327 s; min: 155 s; max: 1,517 s). The mean times for each reader were 485, 557, and 754 s, and the respective learning curves evaluated by regression models revealed statistically significant slopes (P = 0.002; P = 0.0002; P < 0.0001). Overall times were shorter compared with external references that used FT. The mean combined reading and reporting time of MRI examinations using FT was 1,043 s and decreased by 44.8% with GR. GR allows for complete reporting with minimized error rates and reduced combined reading and reporting times. The streamlining of the process (evidenced by lower reading times) for the readers in this study proves that GR can be learned quickly. Reducing reporting errors leads to fewer therapeutic faults and lawsuits against radiologists. It is known that delays in radiology reporting hinder early treatment and lead to poorer patient outcomes. While the number of scans and images per examination is continuously rising, staff shortages create a bottleneck in radiology departments. The IT-based GR method can be a major boon, improving radiologist efficiency, report quality, and the quality of simultaneously generated data.

Multimodal Disease Classification and Severity analysis Approaches using Machine and Deep Learning

Chest X-rays are a common diagnostic tool for pulmonary and cardiac conditions in hospitals because they provide a clear picture of the patient's thorax. With the use of image-to-text radiology report production, medical imaging results may be automatically described in radiology reports. There are a lot of different pieces of patient data that radiologists may access, but most current systems only use the picture data. the objective of developing AI systems with a focus on humans, with the ability to learn radiologists' search habits via their eye movements, with the hope of enhancing DL system categorisation. The goal of this research is to evaluate several multimodal DL architectures in collaboration with trained radiologists to see which ones work best. In particular, this study aims to build strong DL models for medical picture analysis by investigating the integration of several data modalities, such as eye tracking data and patients' clinical data. A multimodal DL model integrating clinical data and chest X-rays (CXRs) was suggested by us. Findings demonstrated that baseline performance was unaffected by directly supplying fixation masks of radiologists' gaze patterns as input. Confine Areas Using R-CNN (Recurrent Neural Networks).

Standardized reporting for Head CT Scans in patients suspected of traumatic brain injury (TBI): An international expert endeavor.

Traumatic brain injury (TBI) is a major source of health loss and disability worldwide. Accurate and timely diagnosis of TBI is critical for appropriate treatment and management of the condition. Neuroimaging plays a crucial role in the diagnosis and characterization of TBI. Computed tomography (CT) is the first-line diagnostic imaging modality typically utilized in patients with suspected acute mild, moderate and severe TBI. Radiology reports play a crucial role in the diagnostic process, providing critical information about the location and extent of brain injury, as well as factors that could prevent secondary injury. However, the complexity and variability of radiology reports can make it challenging for healthcare providers to extract the necessary information for diagnosis and treatment planning. In this article, we report the efforts of an international group of TBI imaging experts to develop a clinical radiology report template for CT scans obtained in patients suspected of TBI and consisting of fourteen different subdivisions (CT technique, mechanism of injury or clinical history, presence of scalp injuries, fractures, potential vascular injuries, potential injuries involving the extra-axial spaces, brain parenchymal injuries, potential injuries involving the cerebrospinal fluid spaces and the ventricular system, mass effect, secondary injuries, prior or coexisting pathology).

Assessing the accuracy of artificial intelligence in the diagnosis and management of orbital fractures: Is this the future of surgical decision-making?

Orbital fractures are common, but their management remains controversial. The aim of the present study was to assess the accuracy of an advanced artificial intelligence (AI) model, ChatGPT-4, in surgical decision-making, with a focus on orbital fracture diagnosis and management.A retrospective observational analysis was conducted by involving a sample of 30 orbital fracture cases diagnosed and managed at the Geneva University Hospital, Switzerland. The process involved creating patient vignettes from anonymised medical records and presenting them to ChatGPT-4 in three stages: initial diagnosis, refinement with radiological reports and surgical intervention decisions. The performance of ChatGPT-4 in providing the appropriate surgical strategy was evaluated through measures of sensitivity, specificity, positive predictive value and negative predictive value, with the actual management used as the benchmark for accuracy.The AI model could correctly diagnose the fracture in 100 % of the cases. It demonstrated a specificity of 100 % and sensitivity of 57 % for treatment recommendation, indicating its effectiveness in recognising patients who truly required an intervention; however, it demonstrated a moderate performance in correctly identifying cases that were better suited for conservative treatment. Cohen's Kappa statistic for interrater reliability of the choice of treatment was 0.44, indicating a weak level of agreement between ChatGPT and the physician's choice of treatment.The study demonstrates that AI tools such as ChatGPT-4 can offer a high degree of accuracy in diagnosing orbital fractures and recognising patients requiring surgical intervention; however, it performed less satisfactorily in correctly identifying patients who were better suited for non-surgical treatment.

Indications du scanner post-mortem à l’Institut médico-légal de Paris en 2021–2022

Objective: Evaluate recent developments of large-scale post-mortem CT (PMCT) in the forensic practice at the Paris Forensic Institute.Methods: All decedents who underwent a PMCT between 2021 and 2022 were included. Variables, extracted from the PMCT information form and report metadata, included: demographic characteristics, context, degree of urgency, identity knowledge, use of CT for comparative identification, time from PMCT to radiology report.Results: 1775 PMCT were performed on request of 5 different courts. Decedents were mostly male (n=1256, 71%) and median age was 58 years (IQR 43-72). The 3 most frequent contexts were: traumatic death (n=618, 35%), unknown cause of death in preserved (n=593, 33%) or putrefied decedents (n=434, 25%). Identity was unknown in 164 (9%) decedents and supposed in 504 (29%) cases, among which a comparative identification was successfully performed using CT in 41/504 (8%) patients. The median delay from PMCT to radiologic report was 6h (IQR 1.8-18.6hrs) and was significantly shorter for urgent as compared to non-urgent reports (2.3 hours [IQR 1.2-6.1] vs. 8.4 hours [IQR 2.4-21.4], p<0.001).Conclusion: Large-scale PMCT usage is possible in forensic setting, in varied indications and contexts. Fast edition of radiologic reports allows its integration into time constraints of the judiciary system.

Constructing a Large Language Model to Generate Impressions from Findings in Radiology Reports.

Background The specialization and complexity of radiology makes the automatic generation of radiologic impressions (ie, a diagnosis with differential diagnosis and management recommendations) challenging. Purpose To develop a large language model (LLM) that generates impressions based on imaging findings and to evaluate its performance in professional and linguistic dimensions. Materials and Methods Six radiologists recorded imaging examination findings from August 2 to 31, 2023, at Shanghai General Hospital and used the developed LLM before routinely writing report impressions for multiple radiologic modalities (CT, MRI, radiography, mammography) and anatomic sites (cranium and face, neck, chest, upper abdomen, lower abdomen, vessels, bone and joint, spine, breast), making necessary corrections and completing the radiologic impression. A subset was defined to investigate cases where the LLM-generated impressions differed from the final radiologist impressions by excluding identical and highly similar cases. An expert panel scored the LLM-generated impressions on a five-point Likert scale (5 = strongly agree) based on scientific terminology, coherence, specific diagnosis, differential diagnosis, management recommendations, correctness, comprehensiveness, harmlessness, and lack of bias. Results In this retrospective study, an LLM was pretrained using 20 GB of medical and general-purpose text data. The fine-tuning data set comprised 1.5 GB of data, including 800 radiology reports with paired instructions (describing the output task in natural language) and outputs. Test set 2 included data from 3988 patients (median age, 56 years [IQR, 40-68 years]; 2159 male). The median recall, precision, and F1 score of LLM-generated impressions were 0.775 (IQR, 0.56-1), 0.84 (IQR, 0.611-1), and 0.772 (IQR, 0.578-0.957), respectively, using the final impressions as the reference standard. In a subset of 1014 patients (median age, 57 years [IQR, 42-69 years]; 528 male), the overall median expert panel score for LLM-generated impressions was 5 (IQR, 5-5), ranging from 4 (IQR, 3-5) to 5 (IQR, 5-5). Conclusion The developed LLM generated radiologic impressions that were professionally and linguistically appropriate for a full spectrum of radiology examinations. © RSNA, 2024 Supplemental material is available for this article.

Nonvisualized Ovaries on Ultrasound: Correlation With Surgical Pathology.

The risk of malignancy in nonvisualized ovaries on pelvic ultrasound is presumed to be close to zero per imaging correlation; the goal of this manuscript is to define the risk of malignancy in nonvisualized ovaries on pelvic ultrasound as defined by surgical pathology. Records for patients with pelvic ultrasound and surgical pathology containing the word "ovary" or "ovaries" performed at our institution between 10/1/2015 and 9/30/2021 were reviewed. Data for ovarian visualization were extracted from the radiology report and correlated with surgical pathology results within each ovary. Eighty-seven ovaries in 71 patients out of 422 ovaries (20.6%) in 215 eligible patients were not visualized on ultrasound. Twenty ovaries were excluded because imaging showed large pelvic mass, and 19 ovaries were excluded because surgical pathology for the ovary of interest was not available. A total of 48 ovaries in 37 patients were nonvisualized and had available surgical pathology. Out of 48 nonvisualized ovaries, 31 were normal on surgical pathology and 17 had abnormalities, with 15 benign lesions (12 of which were ≤1 cm in size). Two ovaries in 1 patient contained malignant lesions; although the ovaries were not visualized on ultrasound, the scan demonstrated peritoneal carcinomatosis. In conclusion, a high proportion of ovaries (20.6%, 87/422) are not visualized on pelvic ultrasound, and surgical pathology reveals ovarian lesions in 35.4% (17/48) of nonvisualized ovaries on pelvic ultrasound, with the majority being subcentimeter benign lesions. In the absence of peritoneal carcinomatosis, nonvisualized ovaries had no malignant lesions.

Helping patients navigate the immediate release of medical records: MedEd, a novel patient engagement technology

BackgroundThrough online health portals, patients receive complex medical reports without interpretation from their healthcare provider. This study evaluated the usability of MedEd, a patient engagement tool providing definitions of medical terminology in breast pathology and radiology reports. MethodsIndividuals who underwent a normal screening mammogram were invited to complete semi-structured interviews where they downloaded MedEd and discussed their download experience. Acceptability, appropriateness, and feasibility of MedEd were evaluated. Results143 individuals were invited to participate, and 14 semi-structured interviews were completed. Participants reported ease of downloading and navigating MedEd with concerns about privacy and others’ abilities to download. Participants demonstrated high acceptability (mean 4.48/5, SD 0.95), appropriateness (mean 4.66/5, SD 0.83), and feasibility (mean 4.48/5, SD 1.04) scores. ConclusionParticipants expressed excitement for future use of MedEd and provided suggestions for improvements. Next steps include evaluating comprehension of real breast reports while using MedEd and expanding patient access.

Clinical value of MR defecography: What additional knowledge is provided by the radiologist to the surgeon?

PurposeTo evaluate the clinical value of MR defecography in supporting surgical decision-making and management strategies in a consecutive series of patients. Material and methodsA consecutive series of patients with clinical suspicion of pelvic disorders who underwent MR defecography at a single university hospital from January 2021 to June 2024 were included. MR defecography was performed at rest (axial/sagittal and coronal T2-weighted sequences), during squeezing, straining, and evacuation (dynamic sagittal T2-weighted sequences). An expert surgeon assessed his satisfaction about the diagnostic information given by the prospective radiological report as: 1: insufficient, 2: sufficient, 3: good, 4: excellent. For patients who underwent surgery, the same surgeon assessed the usefulness of the MR defecography, using a score from 1 to 4 (1: discordant findings, useless exam; 2: concordant findings without new information; 3: concordant findings, useful new information; 4: concordant findings, extremely useful new information). ResultsA total of 218 patients (age 27 − 79, mean 60 ± 11; 193 females and 25 males) entered the analysis. For 211 of them (97 %), the radiologist was able to evaluate the images and determine an MR defecography-based diagnosis; in 7 patients (3 %), a diagnosis was not provided due to exam interruption for low patient’s compliance. In 195/211 patients (92 %), the radiological diagnosis matched the clinical suspect that prompted the examination. The satisfaction about the diagnostic information given by the radiological report of the 211 exams was assessed by the surgeon as follows: excellent (n = 99, 47 %), good (n = 48, 23 %), sufficient (n = 35, 16 %), and insufficient (n = 29, 14 %). Out of 211 patients, 73 (34.6 %) underwent surgery. MR defecography provided concordant findings with relevant additional information in 63 (86.3 %), concordant findings with additional information in 3 (4.1 %), concordant findings without additional information in 4 (5.5 %), and discordant findings in 3 (4.1 %). ConclusionsThe diagnostic information prospectively provided by MR defecography resulted to be good or excellent in 70 % of the patients, adding clinical value in the preoperative surgical setting in 86 % of those patients who underwent surgery.

Age prediction from COVID-19 blood test for ensuring robust artificial intelligence

With the advancement of artificial intelligence (AI) nowadays, the world is experiencing conveniences in automating some complex and tedious tasks, such as analysing large data and predicting the future by mimicking human expertise. AI has also shown promise for mitigating future crisis, such as pandemic. Since the beginning of the COVID-19, several AI models have been published by the researchers to help the healthcare to fight in this situation. However, before deploying the model, one needs to ensure that the model is robust and safe to learn from the real environment, especially in medical domain, where the uncertainty and incomplete information are not unusual. In the effort of providing robust AI, we proposed to use patient age as one of the feasible feature for ensuring vigorous AI models from electronic health record. We conducted several experiment with 28 blood test items and radiologist report from 1,000 COVID-19 patients. Our result shows that with the predicted age as an additional feature in mortality classification task, the model is significantly improved when compared to adding the actual age. We also reported our findings regarding the predicted age in the dataset.

Quantifying Radiology Residents' Learning Curves in Report Writing Performance Through Report Comparison and Jaccard Similarity.

Background Report writing skills are a core competency to be acquired during residency, yet objective tools for tracking performance are lacking. Purpose To investigate whether the Jaccard index, derived from report comparison, can objectively illustrate learning curves in report writing performance throughout radiology residency. Materials and Methods Retrospective data from 246 984 radiology reports written from September 2017 to November 2022 in a tertiary care radiology department were included. Reports were scored using the Jaccard similarity coefficient (ie, a quantitative expression of the amount of edits performed; range, 0-1) of residents' draft (unsupervised initial attempt at a complete report) or preliminary reports (following joint readout with attending physicians) and faculty-reviewed final reports. Weighted mean Jaccard similarity was compared between years of experience using Welch analysis of variance with post hoc testing overall, per imaging division, and per modality. Relationships with years and quarters of resident experience were assessed using Spearman correlation. Results This study included 53 residents (mean report count, 4660 ± 3546; 1-5 years of experience). Mean Jaccard similarity of preliminary reports increased by 6% from 1st-year to 5th-year residents (0.86 ± 0.22 to 0.92 ± 0.15; P < .001). Spearman correlation demonstrated a strong relationship between residents' experience and higher report similarity when aggregated for years (rs = 0.99 [95% CI: 0.85, 1.00]; P < .001) or quarters of experience (rs = 0.90 [95% CI: 0.73, 0.96]; P < .001). For residents' draft reports, Jaccard similarity increased by 14% over the course of the 5-year residency program (0.68 ± 0.27 to 0.82 ± 0.23; P < .001). Subgroup analysis confirmed similar trends for all imaging divisions and modalities (eg, in musculoskeletal imaging, from 0.77 ± 0.31 to 0.91 ± 0.16 [P < .001]; rs = 0.98 [95% CI: 0.72, 1.00] [P < .001]). Conclusion Residents' report writing performance increases with experience. Trends can be quantified with the Jaccard index, with a 6% improvement from 1st- to 5th-year residents, indicating its effectiveness as a tool for evaluating training progress and guiding education over the course of residency. © RSNA, 2024 Supplemental material is available for this article. See also the editorial by Bruno in this issue.

The prognostic relevance of pleural effusion in patients with COVID-19 - A German multicenter study

PurposeThis study evaluates the prognostic significance of pleural effusion (PE) in COVID-19 patients across thirteen centers in Germany, aiming to clarify its role in predicting clinical outcomes. MethodsIn this retrospective analysis within the RACOON project (Radiological Cooperative Network of the COVID-19 pandemic), 1183 patients (29.3 % women, 70.7 % men) underwent chest CT to assess PE. We investigated PE's association with 30-day mortality, ICU admission, and the need for mechanical ventilation. ResultsPE was detected in 31.5 % of patients, showing a significant correlation with 30-day mortality (47.5 % in non-survivors vs. 27.3 % in survivors, p < 0.001), with a hazard ratio of 2.22 (95 % CI 1.65–2.99, p < 0.001). No significant association was found between PE volume or density and mortality. ICU admissions were noted in 46.8 % of patients, while mechanical ventilation was required for 26.7 %. ConclusionPleural effusion is present in a significant portion of COVID-19 patients and independently predicts increased 30-day mortality, underscoring its value as a prognostic marker. Its identification, irrespective of volume or density, should be a priority in radiological reports to guide clinical decision-making.

A Novel Approach for Stratifying Pulmonary Edema Severity on Chest X-ray via Dual-Mechanic Self-Learning and Bidirectional Multi-Modal Cross-Attention Algorithms

Abstract Accurate assessment of pulmonary edema severity in acute decompensated congestive heart failure (CHF) patients is vital for treatment decisions. Traditional methods face challenges due to the complexity of chest X-ray (CXR) and unstructured radiology reports. We proposed a method combining self-supervised learning and multimodal cross-attention to address these challenges. Dual-mechanic self-supervised pre-training enhances feature extraction using contrastive learning between text and image features, and generative learning between images. A bidirectional multi-modal cross-attention model integrates image and text information for fine-tuning, improving model performance. Four CXR datasets consisting of 519, 437 images were used for pre-training; 1200 randomly selected image-text pairs were used for fine-tuning and partitioned into train, validation, and test sets at 3: 1: 1. Ablation studies for pre-training and fine-tuning approaches indicated their practicality as evidenced by the optimal macro F1 score of 0.667 and optimal macro-AUC of 0.904. It also outperformed other state-of-the-art multi-modality methods. The novel approach could accurately assess pulmonary edema severity, offering crucial support for CHF patient management.

Artificial Intelligence in Multilingual Interpretation and Radiology Assessment for Clinical Language Evaluation (AI-MIRACLE).

The AI-MIRACLE Study investigates the efficacy of using ChatGPT 4.0, a large language model (LLM), for translating and simplifying radiology reports into multiple languages, aimed at enhancing patient comprehension. The study assesses the model's performance across the most spoken languages in the U.S., emphasizing the accuracy and clarity of translated and simplified radiology reports for non-medical readers. This study employed ChatGPT 4.0 to translate and simplify selected radiology reports into Vietnamese, Tagalog, Spanish, Mandarin, and Arabic. Hindi was used as a preliminary test language for validation of the questionnaire. Performance was assessed via Google form surveys distributed to bilingual physicians, which assessed the translation accuracy and clarity of simplified texts provided by ChatGPT 4. Responses from 24 participants showed mixed results. The study underscores the model's varying success across different languages, emphasizing both potential applications and limitations. ChatGPT 4.0 shows promise in breaking down language barriers in healthcare settings, enhancing patient comprehension of complex medical information. However, the performance is inconsistent across languages, indicating a need for further refinement and more inclusive training of AI models to handle diverse medical contexts and languages. The study highlights the role of LLMs in improving healthcare communication and patient comprehension, while indicating the need for continued advancements in AI technology, particularly in the translation of low-resource languages.