Citations

Objective To examine the potential errors of a general large language model (LLM) (ie, Claude 3.5 Sonnet) on data extraction from randomised controlled trials (RCTs). Design and setting An empirical study comparing Claude 3.5 Sonnet extractions against a human-performed verification dataset. The extraction tasks for Claude 3.5 Sonnet were based solely on original RCT portable document format (PDF) files. For PDFs that could not be directly extracted by Claude 3.5 Sonnet, optical character recognition was employed to convert them into text format before extraction. Participants A random sample of 664 trials was selected from a well-established trial bank and a final data pool was established based on rigorous manual cross-checking as a reference standard. Data sources PubMed, EMBASE, Scopus, Web of Science (all databases) and the Cochrane Central Register of Controlled Trials (CENTRAL) up to February 2023. Eligibility criteria for selecting studies RCTs on children involving medication and adverse events. Main outcome measures Claude 3.5 Sonnet was applied to extract the basic information (eg, trial design, population information and source of funding) and adverse outcomes (ie, name of adverse events, number of events). Claude 3.5 Sonnet outputs were compared against the final data pool and all errors were recorded. Results are presented as error rates and with 95% CI, estimated using a generalised linear mixed model. Results For the 664 trials, a total of 23 069 data cells were extracted via Claude 3.5 Sonnet, with 10 624 for basic information and 12 445 for adverse outcomes. The overall error rate for data extraction was 6.6% (95% CI 5.4% to 8.2%), with 5.7% (95% CI 5.2% to 6.1%) in basic information and 7.6% (95% CI 4.9% to 11.8%) in adverse outcomes. When stratified the 1542 total errors by error types, misallocation (assigning data to incorrect fields; 57.1%, 881/1542) and missed or omitted data (incomplete extraction of available data; 23.2%, 357/1542) accounted for the two most frequent errors, with misallocation occurring more in basic information (53.3%, 470/881), while missed or omitted data occurred more in adverse outcomes (96.1%, 343/357). Post hoc analysis examining the association between trial reporting quality (assessed using Consolidated Standards of Reporting Trials (CONSORT) 2025 and LLM data extraction error rates indicated that higher CONSORT adherence was associated with lower extraction error rates. Conclusions The data extraction error of Claude was relatively low, but it alerts LLM applications in evidence synthesis. Detailed checking for LLM outputs should be the primary consideration for evidence synthesisers.

Abstract Background Estimating effective dose (E) for CT is often done with dose‐length product (DLP) and effective dose conversion factors, a.k.a. k ‐factors. Previous studies have primarily focused on k ‐factor values derived from certain fixed‐sized phantoms but have not extensively explored all the contributing factors and associated variations in E. Purpose In this study, we investigated the contributing factors and variations of E using k ‐factors as surrogates in adult and pediatric CT exams and provided empirical formula to estimate a comprehensive new set of k ‐factors. Methods Effective dose was estimated using NCICT (National Cancer Institute, USA), including organ and effective dose conversion coefficient library estimated by Monte‐Carlo radiation transport techniques. The k ‐factors were derived as a surrogate of DLP‐normalized E, via E/DLP (unit: mSv/mGy·cm). We examined the variations of effective dose estimates for different phantom and scan parameters represented by the calculations of k ‐factors. This included variations in patient sex, weight, height, kVp, and scan length across typical adult and pediatric CT examinations of anatomical regions such as the head, neck, chest, abdomen, and pelvis. Multivariable regression analyses with/without standardization were conducted to evaluate the significance of the dependency of k ‐factors on the studied parameters. Tube current modulation (TCM) parameters, modeled as a function of patient attenuation using variable modulation strengths, were also incorporated to reflect realistic clinical CT dose variations. Results The variations of DLP‐normalized E were substantially affected by multiple factors, exceeding 100% for adults (weight for adult pelvis) and more than 200% for pediatrics (height for pediatric head). The analysis of standardized regression coefficients revealed that patient weight is the most critical factor, except for head scans in adults, where scan length and patient height are the main determinants. In pediatric CT, both patient weight and height had critical impact. Similarly, for pediatric head CT scans, patient height and scan length are the dominant factors, as observed in adults. Consequently, the k ‐factors have a wide range, from 0.0099 to 0.0357 mSv/mGy·cm for adult chest and from 0.0085 to 0.0363 mSv/mGy·cm for adult abdomen. Empirical formulas to estimate k ‐factors were obtained based on multi‐variable regression with adjusted R square ranging from 0.71 to 0.93. Based on NCICT's TCM model, the use of TCM resulted in an additional average variation of 110% in the k ‐factor compared to non‐TCM results. Conclusions Our study demonstrated the large variations of DLP‐normalized effective dose. Standardized coefficients revealed that patient weight, height and/or scan length were the most crucial determinants. Depending on the specific TCM model used, even greater variations may be observed. This work also provided a practical framework for estimating CT effective dose using the traditional k ‐factor method with updated new conversion factors.

Abstract Photon-counting detector CT (PCD-CT) directly detects individual photons and measures their energies, enabling the acquisition of multi-energy data from a single x-ray source with higher spatial resolution, thereby allowing visualization of smaller arteries and veins and other important imaging features not seen at conventional CT. It also substantially reduces calcium and stent blooming artifacts. Appropriate patient selection and task-based optimization of acquisition, reconstruction, and interpretation are critical to realize these benefits. Vascular imaging examinations for which early evidence shows that PCD-CT demonstrates higher diagnostic performance compared to conventional CT include coronary CT angiography (CTA) for evaluation of partially calcified plaques or stents, peripheral runoff CTA, detection of cerebrospinal fluid-venous fistulas, and head CTA for evaluation of small intracranial vessels and aneurysms. Current limitations include excessive image noise with the sharpest reconstruction kernels and challenges when adapting conventional CT protocols (eg, requiring sharper kernels, thinner slices, higher matrices, tailored scan modes) to achieve time-efficient image reconstruction and hanging protocols. New solutions and directions will include deep-learning-based image noise reduction, improved spectral separation, strategic triage of specific patient populations to PCD-CT systems, and subspecialty clinical practice recommendations for this new technology.

Introduction Increasing publication of fraudulent clinical trials poses a serious threat to evidence-based medicine. In the VITALITY Study I, we demonstrated that contamination by retracted trials significantly distorts evidence synthesis. The upcoming VITALITY Study II will take a step further and investigate how such problematic evidence undermines the credibility of guideline recommendations. Methods and analyses The VITALITY Study II will adopt a retrospective cohort design and will be structured as three work packages (WPs). To start with, a cohort of clinical practice guidelines (CPGs) that were contaminated by retracted trials and/or meta-analyses that included retracted trials will be established based on forward citation searching (WP1). Then, recommendations from these CPGs that used evidence from retracted trials and/or meta-analyses that synthesised these retracted trials will be re-evaluated after removing such problematic evidence, in terms of both the direction and strength of effect sizes (WP2). Finally, the subsequent impact on patient outcomes attributable to distorted recommendations will be estimated on a hypothetical population, measured by the number of expected benefit loss and the number of expected harm increment per 100 000 patients (WP3). Ethics and dissemination Formal ethical approval is not necessary for this study as it does not involve human or animal participants, nor does it involve the collection of primary data. We will disseminate the findings through publication in peer-reviewed journals and, whenever possible, presentations at academic conferences.

Abstract Background Photon‐counting‐detector (PCD) CT systems offer ultra‐high spatial resolution, yet the visual spatial resolution on clinical images often constrained by large pixel size, yielding resolutions below system capabilities. While reducing pixel size and using sharp kernels enhance visual spatial resolution, it increases noise, compromising image quality. Purpose To investigate the combined effects of pixel size and reconstruction kernels on visual spatial resolution using phantom and clinical images and to develop a dedicated high‐resolution deep convolutional neural network (HR‐CNN) to better utilize the intrinsic high spatial resolution of PCD‐CT in clinical imaging. Methods The relationship between spatial resolution, reconstruction kernel, and pixel size was investigated to identify strategies for utilizing the full spatial resolution potential of PCD‐CT. To overcome the increased noise associated with high‐resolution settings, a dedicated HR‐CNN was developed to push the limit of spatial resolution in routine PCD‐CT exams. The HR‐CNN was trained using patient exams acquired with ultra‐high‐resolution (UHR) mode and reconstructed with a 150‐mm field of view (FOV), matrix size of 1024×1024 (0.15‐mm pixel size) and sharpest quantitative kernel (Qr89). The impact of FOV, kernel, and denoising on spatial resolution was studied using bar‐pattern phantoms and a pilot clinical evaluation including 5 patients with interstitial lung diseases. Two thoracic radiologists evaluated 4 different FOV/reconstruction conditions: (1) FOV‐410/Qr56‐Iterative reconstruction (IR), (2) FOV‐410/Qr89‐IR, (3) FOV‐150/Qr89‐IR, and (4) FOV‐150/Qr89‐HR‐CNN in terms of overall image quality, noise, visual spatial resolution, and overall preference. Results With a FOV of 410 mm, the Qr89 sharp kernel displayed bar‐patterns up to 14 lp/cm, not much higher than the routine lung kernel Qr56. When the FOV was reduced to 150 mm, Qr89‐IR allowed for the visualization of line pair patterns ranging from 18 to 20 lp/cm, with 20 lp/cm being moderately discernible. The application of Qr89‐HR‐CNN yielded further improvement, enabling the display of line pair patterns as high as 20–22 lp/cm. In patient cases, both radiologists consistently ranked the FOV‐150 images processed with HR‐CNN as superior across metrics including overall image quality, noise reduction, visual spatial resolution, and overall preference. The HR‐CNN reduced the noise in patients’ images by 93.0 ± 0.6% and 44.9 ± 5.3% in comparison with the original FBP and IR images, respectively. Conclusions The spatial resolution of PCD‐CT is not maximized in routine practice due to the large FOV and high noise levels at sharp kernels. The proposed HR‐CNN denoising method, along with small pixel size, may allow the high spatial resolution toward the system limit to be implemented in practice, which is beneficial in the diagnosis of many diseases, including interstitial lung disease.

Abstract Background Dual‐contrast imaging of iodine and bismuth has shown promise in previous phantom and animal studies utilizing spectral CT. However, these studies often employed Pepto‐Bismol containing bismuth subsalicylate with dosages exceeding the recommended limits for human consumption, limiting their clinical applicability. Purpose The purpose of this study was to improve iodine–bismuth dual‐contrast imaging on a clinical photon‐counting CT (PCCT) system using a noise‐optimized virtual monoenergetic imaging (VMI) approach. Methods Four bismuth samples (0.6, 1.3, 2.5, and 5.1 mg/mL) and three iodine rods (1.0, 2.0, and 5.0 mg/mL) were inserted into a multi‐energy CT phantom and scanned on a PCCT system at 140 kV. VMIs were generated from 40 to 190 keV in 10 keV increments. Iodine and bismuth maps were computed for all possible keV pairs using a generic image‐based three‐material decomposition method and compared with maps generated from the original low‐/high‐energy PCCT images. Mass concentrations, noise levels, and individual as well as joint root‐mean‐square error (RMSE) for iodine and bismuth were measured to determine the optimal keV pair. The contrast‐to‐noise ratios (CNRs) of low concentration samples (bismuth: 0.6 and 1.3 mg/mL; iodine: 1.0 mg/mL) were calculated at the optimal keV pair and compared to the values determined with the original PCCT images. In an IRB‐approved feasibility study, a research patient was administered two doses of Pepto‐Bismol prior to a routine abdomen‐pelvis CT scan. Material decomposition was performed using both the original low‐/high‐energy images and the optimal VMI pair identified in the phantom study. Bismuth signals in the stomach and proximal small bowel were assessed qualitatively and quantitatively. Results The VMI pair of 110 and 190 keV was determined to be optimal, achieving the lowest joint RMSE of 0.49 mg/mL, compared to 1.42 mg/mL for the original low‐/high‐energy pair. It also yielded the highest CNR, representing a 190–318% improvement over the original low/high‐energy decomposition. Qualitatively, both phantom and patient studies demonstrated improved iodine–bismuth separation and reduced noise using the optimized VMI pair. Conclusions These findings suggest that noise‐optimized VMI selection can enhance dual‐contrast material decomposition performance on clinical PCCT, supporting its potential for translation into gastrointestinal imaging applications.