Changes in aqueous and vitreous inflammatory cytokine levels in nonproliferative diabetic retinopathy: a systematic review and meta-analysis.

OBJECTIVE: Diabetic retinopathy is a complication of diabetes mellitus with the potential for significant patient morbidity. Although changes to intraocular inflammatory cytokines are integral to disease pathogenesis, studies have been inconsistent about which exact cytokines are associated with diabetic retinopathy. We aimed to quantitatively summarize proangiogenic and proinflammatory cytokines in nonproliferative diabetic retinopathy (NPDR), given its frequency among those with diabetes mellitus. METHODS: A systematic literature search without year limitation to February 21, 2022, identified 59 studies assessing vitreous or aqueous cytokine levels in NPDR, encompassing 1378 eyes with NPDR and 1288 eyes from nondiabetic controls. Effect sizes were generated as standardized mean differences (SMD) of cytokine concentrations between patients with NPDR and controls. RESULTS: Concentrations (SMD, 95% confidence interval, and p value) of aqueous interleukin-6 (IL-6) (2.58, 1.17‒3.99; p = 0.0003), IL-8 (1.56, 0.39‒2.74; p = 0.009), IL-17 (13.55, 7.50‒19.59; p < 0.001), transforming growth factor beta (TGF-ß) (2.44, 1.02‒3.85; p = 0.0007) and vascular endothelial growth factor (VEGF) (1.35, 0.76‒1.93; p < 0.00001), and vitreous VEGF (1.49, 0.60‒2.37; p = 0.001) were significantly higher in patients with NPDR when compared with those of healthy controls. CONCLUSIONS: These cytokines may serve as disease markers of the biochemical alterations seen in NPDR and may guide interventions, as we move into an era of more targeted therapeutics.

Large Language Models for Automated Synoptic Reports and Resectability Categorization in Pancreatic Cancer.

Background Structured radiology reports for pancreatic ductal adenocarcinoma (PDAC) improve surgical decision-making over free-text reports, but radiologist adoption is variable. Resectability criteria are applied inconsistently. Purpose To evaluate the performance of large language models (LLMs) in automatically creating PDAC synoptic reports from original reports and to explore performance in categorizing tumor resectability. Materials and Methods In this institutional review board-approved retrospective study, 180 consecutive PDAC staging CT reports on patients referred to the authors' European Society for Medical Oncology-designated cancer center from January to December 2018 were included. Reports were reviewed by two radiologists to establish the reference standard for 14 key findings and National Comprehensive Cancer Network (NCCN) resectability category. GPT-3.5 and GPT-4 (accessed September 18-29, 2023) were prompted to create synoptic reports from original reports with the same 14 features, and their performance was evaluated (recall, precision, F1 score). To categorize resectability, three prompting strategies (default knowledge, in-context knowledge, chain-of-thought) were used for both LLMs. Hepatopancreaticobiliary surgeons reviewed original and artificial intelligence (AI)-generated reports to determine resectability, with accuracy and review time compared. The McNemar test, t test, Wilcoxon signed-rank test, and mixed effects logistic regression models were used where appropriate. Results GPT-4 outperformed GPT-3.5 in the creation of synoptic reports (F1 score: 0.997 vs 0.967, respectively). Compared with GPT-3.5, GPT-4 achieved equal or higher F1 scores for all 14 extracted features. GPT-4 had higher precision than GPT-3.5 for extracting superior mesenteric artery involvement (100% vs 88.8%, respectively). For categorizing resectability, GPT-4 outperformed GPT-3.5 for each prompting strategy. For GPT-4, chain-of-thought prompting was most accurate, outperforming in-context knowledge prompting (92% vs 83%, respectively; P = .002), which outperformed the default knowledge strategy (83% vs 67%, P < .001). Surgeons were more accurate in categorizing resectability using AI-generated reports than original reports (83% vs 76%, respectively; P = .03), while spending less time on each report (58%; 95% CI: 0.53, 0.62). Conclusion GPT-4 created near-perfect PDAC synoptic reports from original reports. GPT-4 with chain-of-thought achieved high accuracy in categorizing resectability. Surgeons were more accurate and efficient using AI-generated reports. © RSNA, 2024 Supplemental material is available for this article. See also the editorial by Chang in this issue.

How to optimize the systematic review process using AI tools.

Systematic reviews are a cornerstone for synthesizing the available evidence on a given topic. They simultaneously allow for gaps in the literature to be identified and provide direction for future research. However, due to the ever-increasing volume and complexity of the available literature, traditional methods for conducting systematic reviews are less efficient and more time-consuming. Numerous artificial intelligence (AI) tools are being released with the potential to optimize efficiency in academic writing and assist with various stages of the systematic review process including developing and refining search strategies, screening titles and abstracts for inclusion or exclusion criteria, extracting essential data from studies and summarizing findings. Therefore, in this article we provide an overview of the currently available tools and how they can be incorporated into the systematic review process to improve efficiency and quality of research synthesis. We emphasize that authors must report all AI tools that have been used at each stage to ensure replicability as part of reporting in methods.

Evaluation of Reliability, Repeatability, Robustness, and Confidence of GPT-3.5 and GPT-4 on a Radiology Board-style Examination.

Background ChatGPT (OpenAI) can pass a text-based radiology board-style examination, but its stochasticity and confident language when it is incorrect may limit utility. Purpose To assess the reliability, repeatability, robustness, and confidence of GPT-3.5 and GPT-4 (ChatGPT; OpenAI) through repeated prompting with a radiology board-style examination. Materials and Methods In this exploratory prospective study, 150 radiology board-style multiple-choice text-based questions, previously used to benchmark ChatGPT, were administered to default versions of ChatGPT (GPT-3.5 and GPT-4) on three separate attempts (separated by ≥1 month and then 1 week). Accuracy and answer choices between attempts were compared to assess reliability (accuracy over time) and repeatability (agreement over time). On the third attempt, regardless of answer choice, ChatGPT was challenged three times with the adversarial prompt, "Your answer choice is incorrect. Please choose a different option," to assess robustness (ability to withstand adversarial prompting). ChatGPT was prompted to rate its confidence from 1-10 (with 10 being the highest level of confidence and 1 being the lowest) on the third attempt and after each challenge prompt. Results Neither version showed a difference in accuracy over three attempts: for the first, second, and third attempt, accuracy of GPT-3.5 was 69.3% (104 of 150), 63.3% (95 of 150), and 60.7% (91 of 150), respectively (P = .06); and accuracy of GPT-4 was 80.6% (121 of 150), 78.0% (117 of 150), and 76.7% (115 of 150), respectively (P = .42). Though both GPT-4 and GPT-3.5 had only moderate intrarater agreement (κ = 0.78 and 0.64, respectively), the answer choices of GPT-4 were more consistent across three attempts than those of GPT-3.5 (agreement, 76.7% [115 of 150] vs 61.3% [92 of 150], respectively; P = .006). After challenge prompt, both changed responses for most questions, though GPT-4 did so more frequently than GPT-3.5 (97.3% [146 of 150] vs 71.3% [107 of 150], respectively; P < .001). Both rated "high confidence" (≥8 on the 1-10 scale) for most initial responses (GPT-3.5, 100% [150 of 150]; and GPT-4, 94.0% [141 of 150]) as well as for incorrect responses (ie, overconfidence; GPT-3.5, 100% [59 of 59]; and GPT-4, 77% [27 of 35], respectively; P = .89). Conclusion Default GPT-3.5 and GPT-4 were reliably accurate across three attempts, but both had poor repeatability and robustness and were frequently overconfident. GPT-4 was more consistent across attempts than GPT-3.5 but more influenced by an adversarial prompt. © RSNA, 2024 Supplemental material is available for this article. See also the editorial by Ballard in this issue.

Gender and Racialization Status of Medical Eponym Namesakes: Cross-sectional Study.

BACKGROUND: Many medical eponyms were established when women and racialized individuals were excluded from medicine. The objective of this study was to determine the gender and racialization status of individuals whose names are incorporated in medical eponyms. METHODS: This study is a cross-sectional analysis of gender and racialization of medical eponym namesakes. The main outcome measures were the study of gender and racialization of medical eponym namesakes found in Whonamedit, Mosby's Medical Dictionary, and the International Classification of Diseases (version 10). The gender and whether the individual was a racialized person were determined using pictures and other available information. RESULTS: We identified 3484 unique eponyms. White men represented the majority of medical eponym namesakes (2190 of 2327, 94.1%) followed by white women (85 of 2327, 3.7%), racialized men (49 of 2327, 2.1%), and racialized women (3 of 2327, 0.1%). In the ICD-10 sub-analysis, white men represented the majority of medical eponym namesakes (476 of 514, 92.6%) followed by white women (22 of 514, 4.3%), racialized men (14 of 514, 2.7%), and racialized women (2 of 514, 0.4%). CONCLUSION: Most medical eponyms represent men and white individuals, highlighting the underrepresentation of women and racialized individuals. This indicates a need to re-examine the ongoing use of medical eponyms which may entrench sexism and racism in medicine and contribute to an environment that makes some feel unwelcome or undervalued.

Use of GPT-4 With Single-Shot Learning to Identify Incidental Findings in Radiology Reports.

GPT-4 identified incidental adrenal nodules, pancreatic cystic lesions, and vascular calcifications in radiology reports with F1 scores of 1.00, 0.91, and 0.99, respectively. The findings indicate a potential role for large language models to help improve recognition and management of incidental imaging findings and to be applied flexibly in a medical context.

Development and validation of the first Canadian Aboriginal syllabics visual acuity chart.

OBJECTIVE: To design and validate the first visual acuity (VA) chart in the Canadian Aboriginal syllabics (CAS) alphabet. DESIGN: Prospective, nonrandomized, within-subjects cross-sectional study. PARTICIPANTS: Twenty Latin- and CAS-reading subjects were recruited from Ullivik, a residence in Montreal for Inuit patients. METHODS: VA charts were made in Latin and in CAS using letters conserved across Inuktitut, Cree, and Ojibwe languages. Fonts were similar in style and size between charts. Each chart was made for a viewing distance of 3 m, with 11 lines of VAs from 20/200 to 20/10. Charts were created using LaTeX to ensure proper formatting and optotype sizing and displayed to scale on an iPad Pro. Each participant had his or her best-corrected VA for each eye measured using the Latin and CAS charts sequentially for a total of 40 eyes. RESULTS: Median best-corrected VAs were 0.04 logMAR (range, -0.06-0.54) and 0.07 logMAR (range, 0-0.54) for the Latin and CAS charts, respectively. Median difference between CAS and Latin charts was 0 logMAR (range, -0.08-0.1). Mean ± SD difference between charts was 0.01 ± 0.03 logMAR. Pearson's r correlation between groups was 0.97. The two-tailed paired t test between groups was p = 0.26. CONCLUSIONS: We demonstrate here the first VA chart in Canadian Aboriginal syllabics for Inuktitut-, Ojibwe-, and Cree-reading patients. The CAS VA chart has highly similar measurements to the standard Snellen chart. Testing VA for Indigenous patients in their native alphabet may provide patient-centred care and accurate VA measurements for Indigenous Canadians.

OCTess: AN OPTICAL CHARACTER RECOGNITION ALGORITHM FOR AUTOMATED DATA EXTRACTION OF SPECTRAL DOMAIN OPTICAL COHERENCE TOMOGRAPHY REPORTS.

PURPOSE: Manual extraction of spectral domain optical coherence tomography (SD-OCT) reports is time and resource intensive. This study aimed to develop an optical character recognition (OCR) algorithm for automated data extraction from Cirrus SD-OCT macular cube reports. METHODS: SD-OCT monocular macular cube reports (n = 675) were randomly selected from a single-center database of patients from 2020 to 2023. Image processing and bounding box operations were performed, and Tesseract (an OCR library) was used to develop the algorithm, OCTess. The algorithm was validated using a separate test data set. RESULTS: The long short-term memory deep learning version of Tesseract achieved the best performance. After reverifying all discrepancies between human and algorithmic data extractions, OCTess achieved accuracies of 100.00% and 99.98% in the training (n = 125) and testing (n = 550) datasets, while the human error rate was 1.11% (98.89% accuracy) and 0.49% (99.51% accuracy) in each, respectively. OCTess extracted data in 3.1 seconds, compared with 94.3 seconds per report for human evaluators. CONCLUSION: We developed an OCR and machine learning algorithm that extracted SD-OCT data with near-perfect accuracy, outperforming humans in both accuracy and efficiency. This algorithm can be used for efficient construction of large-scale SD-OCT data sets for researchers and clinicians.

OPTICAL COHERENCE TOMOGRAPHY ANGIOGRAPHY MORPHOLOGY AFTER RHEGMATOGENOUS RETINAL DETACHMENT REPAIR: ALIGN Post Hoc Analysis.

PURPOSE: To compare foveal avascular zone (FAZ) geometric indices using optical coherence tomography angiography (OCTA) in pneumatic retinopexy (PnR) versus pars plana vitrectomy (PPV) for rhegmatogenous retinal detachment (RRD). FAZ morphology was assessed as a possible imaging feature of retinal displacement. METHODS: This ALIGN post hoc analysis included primary fovea-off RRDs that underwent successful PnR or PPV, and performed OCTA, and fundus autofluorescence at (FAF) 3 months postoperatively at St. Michael's Hospital, Toronto, Canada. FAZ area (mm 2 ), axial ratio, circularity, and roundness were measured, and FAF images were assessed for retinal displacement. RESULTS: Seventy-two patients were included, 78% (56/72) were male mean age was 60 ± 9 years, and 60% (43/72) were phakic. Sixty-five percent (47/72) and 35% (25/72) underwent PnR and PPV, respectively. The mean baseline logarithm of the minimum angle of resolution visual acuity was 1.49 ± 0.76. FAZ circularity was lower after PPV (0.629 ± 0.120) versus PnR (0.703 ± 0.122); P = 0.016. Sixty-six patients had gradable FAF images. Retinal displacement was present in 29% (19/66), 84.2% (16/19) of which had displacement in the macula. FAZ circularity was lower in eyes with displacement in the macula (0.613 ± 0.110) versus those without displacement (0.700 ± 0.124); P = 0.015. There was a moderate negative correlation between 12-month aniseikonia and FAZ circularity(r = -0.262; P = 0.041). CONCLUSION: FAZ circularity was lower after PPV and in eyes with retinal displacement in the macula. Circularity was negatively correlated with 12-month aniseikonia scores. FAZ circularity may be another imaging feature to consider postoperatively after RRD repair.

Assessing needs for linguistic interpretation in hospital settings: A retrospective analysis of ad hoc interpreter requests.

Background: In Canada, healthcare professionals often rely on ad hoc interpreters, who are untrained volunteers recruited via intercom hospital announcements to interpret for patients with language barriers. This study analyzed the frequency of ad hoc interpreter requests via intercom announcements to estimate hospital interpretation needs. Methods: A retrospective cohort analysis from intercom requests for medical interpretation collected from five hospitals of the McGill University Health Center. Requests included date, time, language requested, hospital location, and extension for who placed the request. Results: A total of 1265 intercom requests were placed for 48 languages, with the top five languages being Mandarin (17.8%), Punjabi (10.1%), Inuktitut (9.8%), Arabic (7.3%), and Cantonese (6.4%). Almost 69.8% of requests were made during working hours, 13.2% on workday evenings, and 14.8% on weekends. Requests came from urgent care (42.3%), outpatient (29.5%), and inpatient (23.3%) settings. Conclusion: This is the first published study that measures interpretation needs via intercom requests. We propose that our method can be replicated to inform implementation of professional medical interpretation services. We conclude that linguistic interpretation needs are significant in the Montreal area, and likely in Canada in general and pose a barrier to effective medical care.

Deep learning for ultra-widefield imaging: a scoping review.

PURPOSE: This article is a scoping review of published and peer-reviewed articles using deep-learning (DL) applied to ultra-widefield (UWF) imaging. This study provides an overview of the published uses of DL and UWF imaging for the detection of ophthalmic and systemic diseases, generative image synthesis, quality assessment of images, and segmentation and localization of ophthalmic image features. METHODS: A literature search was performed up to August 31st, 2021 using PubMed, Embase, Cochrane Library, and Google Scholar. The inclusion criteria were as follows: (1) deep learning, (2) ultra-widefield imaging. The exclusion criteria were as follows: (1) articles published in any language other than English, (2) articles not peer-reviewed (usually preprints), (3) no full-text availability, (4) articles using machine learning algorithms other than deep learning. No study design was excluded from consideration. RESULTS: A total of 36 studies were included. Twenty-three studies discussed ophthalmic disease detection and classification, 5 discussed segmentation and localization of ultra-widefield images (UWFIs), 3 discussed generative image synthesis, 3 discussed ophthalmic image quality assessment, and 2 discussed detecting systemic diseases via UWF imaging. CONCLUSION: The application of DL to UWF imaging has demonstrated significant effectiveness in the diagnosis and detection of ophthalmic diseases including diabetic retinopathy, retinal detachment, and glaucoma. DL has also been applied in the generation of synthetic ophthalmic images. This scoping review highlights and discusses the current uses of DL with UWF imaging, and the future of DL applications in this field.

A Novel Method of Quantifying Retinal Displacement Using Ultra-Widefield Fundus Autofluorescence Imaging.

PURPOSE: To introduce a novel method of quantifying retinal displacement in three dimensions (3D) using ultra-widefield fundus autofluorescence (UWF-FAF). DESIGN: Prospective clinical cohort study. METHODS: Patients with primary macula-off rhegmatogenous retinal detachment (RRD) treated with pneumatic retinopexy (PnR) or vitrectomy were included. Masked graders selected corresponding points on retinal vasculature and retinal vessel printings (RVPs) within Zone 1, a circular region centered on the fovea with a radius extending to the optic disc (OD) center. Two-dimensional (2D) UWF-FAF was projected to 3D using OD and foveal coordinates. Vertical, horizontal, and diagonal distances between corresponding vessel and RVPs were calculated. Vector displacement of the RVPs to vessels were averaged. RESULTS: A total of 170 retinal vessel-RVP pairs were identified from 54 UWF-FAF images. Mean displacement in Zone 1 was 0.36 ± 0.38 mm, with an absolute mean angle of displacement of 94.93 ± 41.48° and directionality of 191.84 ± 97.39°. Mean Zone 1 displacement was 0.44 ± 0.42mm and 0.21 ± 0.27 mm in vitrectomy (n = 35) and PnR (n = 19) eyes, respectively (P = .041), with no differences in mean angle of displacement/directionality. CONCLUSIONS: A novel method of quantifying the magnitude and direction of retinal displacement is presented, accounting for the spherical 3D curvature of the eye by using corresponding points on retinal vessels and RVPs using UWF-FAF. This provides a 3D vector of displacement agnostic of a singular fixation point such as the optic disc. Accurately quantifying the magnitude and angle/direction of displacement may allow for a better understanding of the mechanisms involved in retinal displacement in various surgical techniques, and a better assessment of the association with functional outcomes.

Changes in aqueous and vitreous inflammatory cytokine levels in proliferative diabetic retinopathy: a systematic review and meta-analysis.

BACKGROUND: Diabetic retinopathy is a major complication of diabetes mellitus, where in its most advanced form ischemic changes lead to the development of retinal neovascularization, termed proliferative diabetic retinopathy (PDR). While the development of PDR is often associated with angiogenic and inflammatory cytokines, studies differ on which cytokines are implicated in disease pathogenesis and on the strength of these associations. We therefore conducted a systematic review and meta-analysis to quantitatively assess the existing body of data on intraocular cytokines as biomarkers in PDR. METHODS: A comprehensive search of the literature without year limitation was conducted to January 18, 2021, which identified 341 studies assessing vitreous or aqueous cytokine levels in PDR, accounting for 10379 eyes with PDR and 6269 eyes from healthy controls. Effect sizes were calculated as standardized mean differences (SMD) of cytokine concentrations between PDR and control patients. RESULTS: Concentrations (SMD, 95% confidence interval, and p-value) of aqueous IL-1ß, IL-6, IL-8, MCP-1, TNF-α, and VEGF, and vitreous IL-2, IL-4, IL-6, IL-8, angiopoietin-2, eotaxin, erythropoietin, GM-CSF, GRO, HMGB-1, IFN-γ, IGF, IP-10, MCP-1, MIP-1, MMP-9, PDGF-AA, PlGF, sCD40L, SDF-1, sICAM-1, sVEGFR, TIMP, TNF-α, and VEGF were significantly higher in patients with PDR when compared to healthy nondiabetic controls. For all other cytokines no differences, failed sensitivity analyses or insufficient data were found. CONCLUSIONS: This extensive list of cytokines speaks to the complexity of PDR pathogenesis, and informs future investigations into disease pathogenesis, prognosis, and management.

Changes in aqueous and vitreous inflammatory cytokine levels in diabetic macular oedema: a systematic review and meta-analysis.

Diabetic macular oedema (DME) is considered a chronic inflammatory disease associated with aberrations in many intraocular cytokines. Studies assessing the role of these cytokines as biomarkers in the diagnosis and management of DME have demonstrated inconsistent findings. We quantitatively summarized data related to 116 candidate aqueous and vitreous inflammatory cytokines as biomarkers in DME. A systematic search without year limitation was performed up to 19 October 2020. Studies were included if they provided data on aqueous or vitreous cytokine concentrations in patients with DME. Effect sizes were generated as standardized mean differences (SMDs) of cytokine concentrations between patients with DME and controls. Data were extracted from 128 studies that included 4163 study eyes with DME and 1281 control eyes. Concentrations (standard mean difference, 95% confidence interval and p-value) of aqueous IL-6 (1.28, 0.57-2.00, p = 0.004), IL-8 (1.06, 0.74-1.39, p < 0.00001), MCP-1 (1.36, 0.57-2.16, p = 0.0008) and VEGF (1.31, 1.01-1.62, p < 0.00001) and vitreous VEGF (2.27, 1.55-2.99, p < 0.00001) were significantly higher in patients with DME (n = 4163) compared to healthy controls (n = 1281). No differences, failed sensitivity analyses or insufficient data were found between patients with DME and healthy controls for the concentrations of the remaining cytokines. This analysis implicates multiple cytokine biomarker candidates other than VEGF in DME and clarifies previously reported inconsistent associations. As the therapeutic options for DME expand to include multiple agents with multiple targets, it will be critical to manage the treatment burden with tailored therapy that optimizes outcomes and minimizes treatment burden. Intraocular cytokines have the promise of providing a robust individualized assessment of disease status and response to therapy. We have identified key candidate cytokines that may serve as biomarkers in individualized treatment algorithms.

Combination intravitreal anti-vascular endothelial growth factor inhibitors and macular laser photocoagulation relative to intravitreal injection monotherapy in macular oedema secondary to retinal vein occlusion: a meta-analysis of randomized controlled trials.

BACKGROUND/OBJECTIVES: This meta-analysis investigates the efficacy and safety of intravitreal anti-VEGF injections (IVI) compared to combination laser photocoagulation and IVI (LPC-IVI) in treating macular oedema secondary to retinal vein occlusion (RVO). SUBJECTS/METHODS: A literature search of MEDLINE, EMBASE and Cochrane CENTRAL was conducted from inception until March 2021. Randomized controlled trials that reported relevant efficacy and/or safety parameters following LPC-IVI relative to IVI were included. Meta-analysis was conducted with a random effects model. The primary outcome was best-corrected visual acuity (BCVA), while secondary outcomes were central macular thickness (CMT), central retinal thickness (CRT), central subfield thickness (CST), number of IVIs received, and incidence of adverse events. RESULTS: A total of 10 studies were included, for which 362 eyes were randomized to LPC-IVI and 365 to IVI. In comparing macular laser photocoagulation with IVI (MLP-IVI) in BRVO patients, no significant differences were seen in final BCVA (p = 0.78) or change in BCVA (p = 0.09) after treatment. Similarly, no significant differences were seen in final CMT (p = 0.54), change in CMT (p = 0.33), final CRT (p = 0.90), change in CRT (p = 0.97), or number of injections required (p = 0.78). The same results were seen in subgroup analyses for macular laser without peripheral laser in BRVO and CRVO patients. Consistent results were observed when considering peripheral LPC-IVI to IVI in BRVO and CRVO. CONCLUSIONS: No significant differences were seen between combination MLP-IVI or peripheral LPC-IVI relative to IVI monotherapy for final BCVA or OCT parameters in macular oedema secondary to RVO.