RESUMEN

Tuberculosis (TB) is the leading cause of mortality among infectious diseases globally. Effectively managing TB requires early identification of individuals with TB disease. Resource-constrained settings often lack skilled professionals for interpreting chest X-rays (CXRs) used in TB diagnosis. To address this challenge, we developed "DecXpert" a novel Computer-Aided Detection (CAD) software solution based on deep neural networks for early TB diagnosis from CXRs, aiming to detect subtle abnormalities that may be overlooked by human interpretation alone. This study was conducted on the largest cohort size to date, where the performance of a CAD software (DecXpert version 1.4) was validated against the gold standard molecular diagnostic technique, GeneXpert MTB/RIF, analyzing data from 4363 individuals across 12 primary health care centers and one tertiary hospital in North India. DecXpert demonstrated 88% sensitivity (95% CI 0.85-0.93) and 85% specificity (95% CI 0.82-0.91) for active TB detection. Incorporating demographics, DecXpert achieved an area under the curve of 0.91 (95% CI 0.88-0.94), indicating robust diagnostic performance. Our findings establish DecXpert's potential as an accurate, efficient AI solution for early identification of active TB cases. Deployed as a screening tool in resource-limited settings, DecXpert could enable early identification of individuals with TB disease and facilitate effective TB management where skilled radiological interpretation is limited.

Asunto(s)

RESUMEN

BACKGROUND: Very low birthweight infants (VLBWIs) often undergo chest radiographic examinations without standardization or objectivity. This study aimed to assess the association of two radiographic scores, the Brixia and radiographic assessment of lung edema (RALE), with oxygenation index (OI) in ventilated VLBWIs and to determine the optimal cutoff values to predict hypoxic respiratory severity. METHODS: VLBWIs who received invasive respiratory support with arterial lines between January 2010 and October 2023 were enrolled in this study (n = 144). The correlation between the Brixia or RALE scores and OI was investigated. Receiver operating characteristic curve analysis was performed to determine the optimal cutoff points of the two radiographic scores for predicting OI values (OI ≥5, ≥10, and ≥15). RESULTS: The enrolled infants had a median gestational age of 27 weeks (interquartile range [IQR], 25-28 weeks) and a median birthweight of 855 g (IQR, 684-1003 g). Radiographic scoring methods correlated with the OI (Brixia score: r = 0.79, p < 0.001; RALE score: r = 0.72, p < 0.001). The optimal cutoff points for predicting OI values were as follows: Brixia score: OI ≥5, 10; OI ≥10, 13; OI ≥15, 15; RALE score: OI ≥5, 22; OI ≥10, 31; and OI ≥15, 40. CONCLUSIONS: Brixia and RALE scores are useful predictive markers of the oxygenation status in intubated VLBWIs with stable hemodynamics. These scores are easy to use and promising tools for clinicians to identify patients with a higher risk of hypoxic respiratory failure.

Asunto(s)

RESUMEN

OBJECTIVE: To explore the effects of tube voltage, radiation dose and adaptive statistical iterative reconstruction (ASiR-V) strength level on the detection and characterization of pulmonary nodules by an artificial intelligence (AI) software in ultra-low-dose chest CT (ULDCT). MATERIALS AND METHODS: An anthropomorphic thorax phantom containing 12 spherical simulated nodules (Diameter: 12 mm, 10 mm, 8 mm, 5 mm; CT value: -800HU, -630HU, 100HU) was scanned with three ULDCT protocols: Dose-1 (70kVp:0.11mSv, 100kVp:0.10mSv), Dose-2 (70kVp:0.34mSv, 100kVp:0.32mSv), Dose-3 (70kVp:0.53mSv, 100kVp:0.51mSv). All scanning protocols were repeated five times. CT images were reconstructed using four different strength levels of ASiR-V (0%=FBP, 30%, 50%, 70%ASiR-V) with a slice thickness of 1.25 mm. The characteristics of the physical nodules were used as reference standards. All images were analyzed using a commercially available AI software to identify nodules for calculating nodule detection rate (DR) and to obtain their long diameter and short diameter, which were used to calculate the deformation coefficient (DC) and size measurement deviation percentage (SP) of nodules. DR, DC and SP of different imaging groups were statistically compared. RESULTS: Image noise decreased with the increase of ASiR-V strength level, and the 70 kV images had lower noise under the same strength level (mean-value 70 kV: 40.14 ± 7.05 (dose 1), 27.55 ± 7.38 (dose 2), 23.88 ± 6.98 (dose 3); 100 kV: 42.36 ± 7.62 (dose 1); 30.78 ± 6.87 (dose 2); 26.49 ± 6.61 (dose 3)). Under the same dose level, there were no differences in DR between 70 kV and 100 kV (dose 1: 58.76% vs. 58.33%; dose 2: 73.33% vs. 70.83%; dose 3: 75.42% vs. 75.42%, all p > 0.05). The DR of GGNs increased significantly at dose 2 and higher (70 kV: 38.12% (dose 1), 60.63% (dose 2), 64.38% (dose 3); 100 kV: 37.50% (dose 1), 59.38% (dose 2), 66.25% (dose 3)). In general, the use of ASiR-V at higher strength levels (> 50%) and 100 kV provided better (lower) DC and SP. CONCLUSION: Detection rates are similar between 70 kV and 100 kV scans. The 70 kV images have better noise performance under the same ASiR-V level, while images of 100 kV and higher ASiR-V levels are better in preserving the nodule morphology (lower DC and SP); the dose levels above 0.33mSv provide high sensitivity for nodules detection, especially the simulated ground glass nodules.

Asunto(s)

RESUMEN

Recently, multi-modal vision-language foundation models have gained significant attention in the medical field. While these models offer great opportunities, they still face crucial challenges, such as the requirement for fine-grained knowledge understanding in computer-aided diagnosis and the capability of utilizing very limited or even no task-specific labeled data in real-world clinical applications. In this study, we present MaCo, a masked contrastive chest X-ray foundation model that tackles these challenges. MaCo explores masked contrastive learning to simultaneously achieve fine-grained image understanding and zero-shot learning for a variety of medical imaging tasks. It designs a correlation weighting mechanism to adjust the correlation between masked chest X-ray image patches and their corresponding reports, thereby enhancing the model's representation learning capabilities. To evaluate the performance of MaCo, we conducted extensive experiments using 6 well-known open-source X-ray datasets. The experimental results demonstrate the superiority of MaCo over 10 state-of-the-art approaches across tasks such as classification, segmentation, detection, and phrase grounding. These findings highlight the significant potential of MaCo in advancing a wide range of medical image analysis tasks.

Asunto(s)

RESUMEN

Background: Lung cancer is one of the most common types of cancer in the United Kingdom. It is often diagnosed late. The 5-year survival rate for lung cancer is below 10%. Early diagnosis may improve survival. Software that has an artificial intelligence-developed algorithm might be useful in assisting with the identification of suspected lung cancer. Objectives: This review sought to identify evidence on adjunct artificial intelligence software for analysing chest X-rays for suspected lung cancer, and to develop a conceptual cost-effectiveness model to inform discussion of what would be required to develop a fully executable cost-effectiveness model for future economic evaluation. Data sources: The data sources were MEDLINE All, EMBASE, Cochrane Database of Systematic Reviews, Cochrane CENTRAL, Epistemonikos, ACM Digital Library, World Health Organization International Clinical Trials Registry Platform, clinical experts, Tufts Cost-Effectiveness Analysis Registry, company submissions and clinical experts. Searches were conducted from 25 November 2022 to 18 January 2023. Methods: Rapid evidence synthesis methods were employed. Data from companies were scrutinised. The eligibility criteria were (1) primary care populations referred for chest X-ray due to symptoms suggestive of lung cancer or reasons unrelated to lung cancer; (2) study designs that compared radiology specialist assessing chest X-ray with adjunct artificial intelligence software versus radiology specialists alone and (3) outcomes relating to test accuracy, practical implications of using artificial intelligence software and patient-related outcomes. A conceptual decision-analytic model was developed to inform a potential full cost-effectiveness evaluation of adjunct artificial intelligence software for analysing chest X-ray images to identify suspected lung cancer. Results: None of the studies identified in the searches or submitted by the companies met the inclusion criteria of the review. Contextual information from six studies that did not meet the inclusion criteria provided some evidence that sensitivity for lung cancer detection (but not nodule detection) might be higher when chest X-rays are interpreted by radiology specialists in combination with artificial intelligence software than when they are interpreted by radiology specialists alone. No significant differences were observed for specificity, positive predictive value or number of cancers detected. None of the six studies provided evidence on the clinical effectiveness of adjunct artificial intelligence software. The conceptual model highlighted a paucity of input data along the course of the diagnostic pathway and identified key assumptions required for evidence linkage. Limitations: This review employed rapid evidence synthesis methods. This included only one reviewer conducting all elements of the review, and targeted searches that were conducted in English only. No eligible studies were identified. Conclusions: There is currently no evidence applicable to this review on the use of adjunct artificial intelligence software for the detection of suspected lung cancer on chest X-ray in either people referred from primary care with symptoms of lung cancer or people referred from primary care for other reasons. Future work: Future research is required to understand the accuracy of adjunct artificial intelligence software to detect lung nodules and cancers, as well as its impact on clinical decision-making and patient outcomes. Research generating key input parameters for the conceptual model will enable refinement of the model structure, and conversion to a full working model, to analyse the cost-effectiveness of artificial intelligence software for this indication. Study registration: This study is registered as PROSPERO CRD42023384164. Funding: This award was funded by the National Institute for Health and Care Research (NIHR) Evidence Synthesis programme (NIHR award ref: NIHR135755) and is published in full in Health Technology Assessment; Vol. 28, No. 50. See the NIHR Funding and Awards website for further award information.

Lung cancer is one of the most common types of cancer in the United Kingdom. Early diagnosis may improve survival, as lung cancer is often diagnosed late. Chest X-rays can be used to identify features of lung cancer. There can be delays in getting X-rays, and sometimes features of lung cancer are not seen on them. Artificial intelligence software may help by finding features of cancer on chest X-rays and highlighting them. A radiologist will look at the X-rays and information from the software. There is a lack of information about how lung cancer diagnosis could change if artificial intelligence software is used and what the costs may be to the National Health Service. This project looked at the use of artificial intelligence software in the detection of lung cancer in people referred from primary care. Software companies were invited to provide evidence. There were no studies that looked at this topic among people from primary care. We summarised the closest evidence we could find instead. All of this had flaws, so we could not tell if the results were accurate or helpful to this review. It was not clear if artificial intelligence helped to find cancers or improve people's health. We made a theoretical model to discuss the best way to assess if artificial intelligence software might be cost-effective in detecting lung cancer and what evidence would be needed to do this in a fully working model. Costs and alternative pricing models provided by five companies were used to calculate the cost of adding artificial intelligence software to review chest X-rays in people referred from their general practitioner, for the first 5 years, based on one National Health Service trust. Future studies are needed to identify the impact of adjunct artificial intelligence on test accuracy, clinical decision-making and patient outcomes (e.g. mortality and morbidity).

Asunto(s)

RESUMEN

BACKGROUND The incidence of lung diseases in premature newborns is significantly higher than in full-term newborns due to their underdeveloped lungs. Ultrasound and X-ray are commonly-used bedside examinations in neonatology. This study primarily compares the efficacy of chest X-ray (CXR) and lung ultrasound (LUS) images in evaluating lung consolidation and edema in premature newborns at Neonatal Intensive Care Units (NICU). MATERIAL AND METHODS A retrospective analysis was conducted on LUS and CXR examination results, along with clinical records of premature newborns admitted to our hospital's NICU from November 1, 2019, to December 31, 2021. CXR and LUS scans were performed on the same newborn within a day. We evaluated the consolidations and edema by interpreting the CXR and LUS images, then compared the findings. RESULTS Out of 75 cases, 34 showed lung consolidations on LUS (45%), while only 14 exhibited consolidations on CXR (19%). The detection rate of consolidations by LUS was significantly higher compared to CXR (34/75 vs 14/75, P<0.001). Differences were observed between the 2 bedside examinations in identifying consolidations, with some cases seen only on LUS. CXR struggled to accurately assess the severity of lung edema visible on LUS, showing significant disparity in detecting interstitial edema (53/75 vs 21/75, P<0.001). CONCLUSIONS LUS outperforms chest CXR for bedside assessment of lung consolidation and edema in premature newborns.

Asunto(s)

RESUMEN

BACKGROUND AND OBJECTIVE: Registration of pulmonary computed tomography (CT) images with radiation-induced lung diseases (RILD) was essential to investigate the voxel-wise relationship between the formation of RILD and the radiation dose received by different tissues. Although various approaches had been developed for the registration of lung CTs, their performances remained clinically unsatisfactory for registration of lung CT images with RILD. The main difficulties arose from the longitudinal change in lung parenchyma, including RILD and volumetric change of lung cancers, after radiation therapy, leading to inaccurate registration and artifacts caused by erroneous matching of the RILD tissues. METHODS: To overcome the influence of the parenchymal changes, a divide-and-conquer approach rooted in the coherent point drift (CPD) paradigm was proposed. The proposed method was based on two kernel ideas. One was the idea of component structure wise registration. Specifically, the proposed method relaxed the intrinsic assumption of equal isotropic covariances in CPD by decomposing a lung and its surrounding tissues into component structures and independently registering the component structures pairwise by CPD. The other was the idea of defining a vascular subtree centered at a matched branch point as a component structure. This idea could not only provide a sufficient number of matched feature points within a parenchyma, but avoid being corrupted by the false feature points resided in the RILD tissues due to globally and indiscriminately sampling using mathematical operators. The overall deformation model was built by using the Thin Plate Spline based on all matched points. RESULTS: This study recruited 30 pairs of lung CT images with RILD, 15 of which were used for internal validation (leave-one-out cross-validation) and the other 15 for external validation. The experimental results showed that the proposed algorithm achieved a mean and a mean of maximum 1 % of average surface distances <2 and 8 mm, respectively, and a mean and a maximum target registration error <2 mm and 5 mm on both internal and external validation datasets. The paired two-sample t-tests corroborated that the proposed algorithm outperformed a recent method, the Stavropoulou's method, on the external validation dataset (p < 0.05). CONCLUSIONS: The proposed algorithm effectively reduced the influence of parenchymal changes, resulting in a reasonably accurate and artifact-free registration.

Asunto(s)

RESUMEN

BACKGROUND: Thoracic computed tomography scans (CT) are used by several study groups to investigate the circulatory structures (heart and vessels) located behind the pressure point for chest compressions. Yet, it remains unclear how the positioning of these structures is influenced by factors such as intubation, the respiratory cycle and arm positioning. METHODS: We retrospectively analyzed data of adult patients with in- or out-of-hospital cardiac arrest who underwent thoracic CT imaging within one year before or up to six months after arrest. A region of interest (ROI) behind the pressure point was defined. The largest structure within this region was defined as "leading circulatory structure", which was the primary outcome. Airway status (intubated versus spontaneous breathing), respiratory cycle (inspiration, expiration, resting expiratory position), and arm position (up over the head versus down beside the trunk) served as covariates in an ordinal regression model. RESULTS: Among 500 initially screened patients, 411 (82.2 %) were included in the analysis. There was a significant association between the arm position and the leading circulatory structure behind the pressure point. However, no association was found with airway status or respiratory cycle. The most frequently identified leading circulatory structure was the left atrium (arms up: 41.8 %, down: 50.7 %), followed by the ascending aorta (up: 23.8 % vs. down: 16.7 %). The left ventricle was the leading structure in only one case (0.2 %, arms down). CONCLUSION: This study shows that arm position is significantly associated with the leading circulatory structure behind the pressure point for chest compressions in cardiac arrest.

Asunto(s)

RESUMEN

OBJECTIVE: To assess the effect of a new lung enhancement filter combined with deep learning image reconstruction (DLIR) algorithm on image quality and ground-glass nodule (GGN) sharpness compared to hybrid iterative reconstruction or DLIR alone. MATERIALS AND METHODS: Five artificial spherical GGNs with various densities (-250, -350, -450, -550, and -630 Hounsfield units) and 10 mm in diameter were placed in a thorax anthropomorphic phantom. Four scans at four different radiation dose levels were performed using a 256-slice CT (Revolution Apex CT, GE Healthcare). Each scan was reconstructed using three different reconstruction algorithms: adaptive statistical iterative reconstruction-V at a level of 50% (AR50), Truefidelity (TF), which is a DLIR method, and TF with a lung enhancement filter (TF + Lu). Thus, 12 sets of reconstructed images were obtained and analyzed. Image noise, signal-to-noise ratio, and contrast-to-noise ratio were compared among the three reconstruction algorithms. Nodule sharpness was compared among the three reconstruction algorithms using the full-width at half-maximum value. Furthermore, subjective image quality analysis was performed. RESULTS: AR50 demonstrated the highest level of noise, which was decreased by using TF + Lu and TF alone (P = 0.001). TF + Lu significantly improved nodule sharpness at all radiation doses compared to TF alone (P = 0.001). The nodule sharpness of TF + Lu was similar to that of AR50. Using TF alone resulted in the lowest nodule sharpness. CONCLUSION: Adding a lung enhancement filter to DLIR (TF + Lu) significantly improved the nodule sharpness compared to DLIR alone (TF). TF + Lu can be an effective reconstruction technique to enhance image quality and GGN evaluation in ultralow-dose chest CT scans.

Asunto(s)

RESUMEN

BACKGROUND: Lung ultrasonography is an emerging tool in diagnosing community-acquired pneumonia (CAP) - a major cause of mortality worldwide. The objective of the study was to determine the diagnostic performance of point-of-care ultrasound (POCUS) of the lung compared to the chest radiograph in the diagnosis of CAP in adults. METHODS: Adults ≥ 18 years presenting at the general and medical outpatient clinics, medical and emergency wards with symptoms of suspected CAP were evaluated using a portable ultrasound device and single posteroanterior chest radiograph. Sensitivity, specificity, positive and negative predictive values (PPV and NPV), positive and negative likelihood ratios (LR+ and LR-) with corresponding 95% confidence intervals were computed for the lung ultrasound (LUS) against the chest radiograph as the criterion standard. RESULTS: Out of the 65 patients eventually studied, 50 (76.9%) were diagnosed with pneumonia by chest radiograph. The sensitivity, specificity, PPV, NPV, LR+, LR- and DOR for the LUS against the chest radiograph, respectively, were 96% (95%CI, 86.3% - 99.5%), 93.3% (95%CI, 68.1% - 99.8%), 98.0% (95%CI, 87.8% - 99.7%), 87.5% (64.1% - 96.5%), 14.4 (95%CI, 2.2 - 95.7), 0.04 (95%CI, 0.01 - 0.17) and 336 (28.3 - 3985.0). The overall accuracy was 95.4% (95%CI, 87.1 - 99.0%). The median time to completion of the LUS was 13 minutes. CONCLUSION: Lung ultrasound at the point of care is a reasonably accurate tool for the diagnosis of CAP in adults presenting with typical features.

CONTEXTE: L'échographie pulmonaire est un outil émergent dans le diagnostic de la pneumonie communautaire (CAP) ­ une cause majeure de mortalité dans le monde entier. L'objectif de l'étude était de déterminer la performance diagnostique de l'échographie pulmonaire au point de soins (POCUS) par rapport à la radiographie thoracique dans le diagnostic de la CAP chez les adultes. MÉTHODES: Les adultes ≥ 18 ans se présentant aux cliniques générales et médicales, aux services médicaux et d'urgence avec des symptômes de CAP suspectée ont été évalués à l'aide d'un appareil d'échographie portable et d'une radiographie thoracique postéroantérieure unique. La sensibilité, la spécificité, les valeurs prédictives positive et négative (PPV et NPV), les rapports de vraisemblance positifs et négatifs (LR+ et LR-) avec les intervalles de confiance correspondants à 95 % ont été calculés pour l'échographie pulmonaire (LUS) par rapport à la radiographie thoracique comme norme de référence. RÉSULTATS: Sur les 65 patients étudiés, 50 (76,9 %) ont été diagnostiqués avec une pneumonie par adiographie thoracique. La sensibilité, la spécificité, la PPV, la NPV, les LR+, LR- et DOR pour la LUS par rapport à la radiographie thoracique étaient respectivement de 96 % (IC à 95%, 86,3% ­ 99,5%), 93,3% (IC à 95%, 68,1% ­ 99,8%), 98,0% (IC à 95%, 87,8% - 99,7%), 87,5% (64,1% - 96,5%), 14,4 (IC à 95%, 2,2 ­ 95,7), 0,04 (IC à 95 %, 0,01 ­ 0,17) et 336 (28,3 ­ 3985,0). La précision globale était de 95,4 % (IC à 95%, 87,1 ­ 99,0%). Le temps médian pour l'achèvement de la LUS était de 13 minutes. CONCLUSION: L'échographie pulmonaire au point de soins est un outil raisonnablement précis pour le diagnostic de la CAP chez les adultes présentant des caractéristiques typiques. MOTS-CLÉS: Échographie pulmonaire, Radiographie thoracique, Pneumonie communautaire, Précision diagnostique, Ressources limitées.

Asunto(s)

Asunto(s)

RESUMEN

COVID-19 has resulted in a significant global impact on health, the economy, education, and daily life. The disease can range from mild to severe, with individuals over 65 or those with underlying medical conditions being more susceptible to severe illness. Early testing and isolation are vital due to the virus's variable incubation period. Chest radiographs (CXR) have gained importance as a diagnostic tool due to their efficiency and reduced radiation exposure compared to CT scans. However, the sensitivity of CXR in detecting COVID-19 may be lower. This paper introduces a deep learning framework for accurate COVID-19 classification and severity prediction using CXR images. U-Net is used for lung segmentation, achieving a precision of 0.9924. Classification is performed using a Convulation-capsule network, with high true positive rates of 86% for COVID-19, 93% for pneumonia, and 85% for normal cases. Severity assessment employs ResNet50, VGG-16, and DenseNet201, with DenseNet201 showing superior accuracy. Empirical results, validated with 95% confidence intervals, confirm the framework's reliability and robustness. This integration of advanced deep learning techniques with radiological imaging enhances early detection and severity assessment, improving patient management and resource allocation in clinical settings.

Asunto(s)

RESUMEN

BACKGROUND: Calculating size-specific dose estimates (SSDEs) requires measurement of the patient's anteroposterior (AP) and lateral thickness based on computed tomography (CT) images. However, these measurements can be subject to variation due to inter-observer and intra-observer differences. This study aimed to investigate the impact of these variations on the accuracy of the calculated SSDE. METHODS: Four radiographers with 1-10 years of experience were invited to measure the AP and lateral thickness on 30 chest, abdomen, and pelvic CT images. The images were sourced from an internet-based database and anonymized for analysis. The observers were trained to perform the measurements using MicroDicom software and asked to repeat the measurements 1 week later. The study was approved by the institutional review board at Taibah University, and written informed consent was obtained from the observers. Statistical analyses were performed using Python libraries Pingouin (version 0.5.3), Seaborn (version 0.12.2), and Matplotlib (version 3.7.1). RESULTS: The study revealed excellent inter-observer agreement for the calculated effective diameter and AP thickness measurements, with Intraclass correlation coefficients (ICC) values of 0.95 and 0.96, respectively. The agreement for lateral thickness measurements was lower, with an ICC value of 0.89. The second round of measurements yielded nearly the same levels of inter-observer agreement, with ICC values of 0.97 for the effective diameter, 1.0 for AP thickness, and 0.88 for lateral thickness. When the consistency of the observer was examined, excellent consistency was found for the calculated effective diameter, with ICC values ranging from 0.91 to 1.0 for all observers. This was observed despite the lower consistency in the lateral thickness measurements, which had ICC values ranging from 0.78 to 1.0. CONCLUSIONS: The study's findings suggest that the measurements required for calculating SSDEs are robust to inter-observer and intra-observer differences. This is important for the clinical use of SSDEs to set diagnostic reference levels for CT scans.

Asunto(s)

RESUMEN

Background Radiology practices have a high volume of unremarkable chest radiographs and artificial intelligence (AI) could possibly improve workflow by providing an automatic report. Purpose To estimate the proportion of unremarkable chest radiographs, where AI can correctly exclude pathology (ie, specificity) without increasing diagnostic errors. Materials and Methods In this retrospective study, consecutive chest radiographs in unique adult patients (≥18 years of age) were obtained January 1-12, 2020, at four Danish hospitals. Exclusion criteria included insufficient radiology reports or AI output error. Two thoracic radiologists, who were blinded to AI output, labeled chest radiographs as "remarkable" or "unremarkable" based on predefined unremarkable findings (reference standard). Radiology reports were classified similarly. A commercial AI tool was adapted to output a chest radiograph "remarkableness" probability, which was used to calculate specificity at different AI sensitivities. Chest radiographs with missed findings by AI and/or the radiology report were graded by one thoracic radiologist as critical, clinically significant, or clinically insignificant. Paired proportions were compared using the McNemar test. Results A total of 1961 patients were included (median age, 72 years [IQR, 58-81 years]; 993 female), with one chest radiograph per patient. The reference standard labeled 1231 of 1961 chest radiographs (62.8%) as remarkable and 730 of 1961 (37.2%) as unremarkable. At 99.9%, 99.0%, and 98.0% sensitivity, the AI had a specificity of 24.5% (179 of 730 radiographs [95% CI: 21, 28]), 47.1% (344 of 730 radiographs [95% CI: 43, 51]), and 52.7% (385 of 730 radiographs [95% CI: 49, 56]), respectively. With the AI fixed to have a similar sensitivity as radiology reports (87.2%), the missed findings of AI and reports had 2.2% (27 of 1231 radiographs) and 1.1% (14 of 1231 radiographs) classified as critical (P = .01), 4.1% (51 of 1231 radiographs) and 3.6% (44 of 1231 radiographs) classified as clinically significant (P = .46), and 6.5% (80 of 1231) and 8.1% (100 of 1231) classified as clinically insignificant (P = .11), respectively. At sensitivities greater than or equal to 95.4%, the AI tool exhibited less than or equal to 1.1% critical misses. Conclusion A commercial AI tool used off-label could correctly exclude pathology in 24.5%-52.7% of all unremarkable chest radiographs at greater than or equal to 98% sensitivity. The AI had equal or lower rates of critical misses than radiology reports at sensitivities greater than or equal to 95.4%. These results should be confirmed in a prospective study. © RSNA, 2024 Supplemental material is available for this article. See also the editorial by Yoon and Hwang in this issue.

Asunto(s)

RESUMEN

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.

Asunto(s)

RESUMEN

BACKGROUND: To assess changes in bone density and vertebral body height of patients undergoing lung transplant surgery using computed tomography (CT). METHODS: This institutional review board (IRB) approved retrospective observational study enrolled patients with a history of lung transplant who had at least two chest CT scans. Vertebral body bone density (superior, middle, and inferior sections) and height (anterior, middle, and posterior sections) were measured at T1-T12 at baseline and follow up CT scans. Changes in the mean bone density, mean vertebral height, vertebral compression ratio (VBCR), percentage of anterior height compression (PAHC), and percentage of middle height compression (PMHC) were calculated and analyzed. RESULTS: A total of 93 participants with mean age of 58 ± 12.3 years were enrolled. The most common underlying disease that led to lung transplants was interstitial lung diseases (57 %). The inter-scan interval was 34.06 ± 24.8 months. There were significant changes (p-value < 0.05) in bone density at all levels from T3 to T12, with the greatest decline at the T10 level from 163.06 HU to 141.84 HU (p-value < 0.05). The average VBCR decreased from 96.91 to 96.15 (p-value < 0.05). CONCLUSION: Routine chest CT scans demonstrate a gradual decrease in vertebral body bone density over time in lung transplant recipients, along with evident anatomic changes such as vertebral body bone compression. This study shows that utilizing routine chest CT for lung transplant recipients can be regarded as a cost-free tool for assessing the vertebral body bone changes in these patients and potentially aiding in the prevention of complications related to osteoporosis.

Asunto(s)

RESUMEN

BACKGROUND: Chest X-ray (CXR) is one of the most commonly performed imaging tests worldwide. Due to its wide usage, there is a growing need for automated and generalizable methods to accurately diagnose these images. Traditional methods for chest X-ray analysis often struggle with generalization across diverse datasets due to variations in imaging protocols, patient demographics, and the presence of overlapping anatomical structures. Therefore, there is a significant demand for advanced diagnostic tools that can consistently identify abnormalities across different patient populations and imaging settings. We propose a method that can provide a generalizable diagnosis of chest X-ray. METHOD: Our method utilizes an attention-guided decomposer network (ADSC) to extract disease maps from chest X-ray images. The ADSC employs one encoder and multiple decoders, incorporating a novel self-consistency loss to ensure consistent functionality across its modules. The attention-guided encoder captures salient features of abnormalities, while three distinct decoders generate a normal synthesized image, a disease map, and a reconstructed input image, respectively. A discriminator differentiates the real and the synthesized normal chest X-rays, enhancing the quality of generated images. The disease map along with the original chest X-ray image are fed to a DenseNet-121 classifier modified for multi-class classification of the input X-ray. RESULTS: Experimental results on multiple publicly available datasets demonstrate the effectiveness of our approach. For multi-class classification, we achieve up to a 3% improvement in AUROC score for certain abnormalities compared to the existing methods. For binary classification (normal versus abnormal), our method surpasses existing approaches across various datasets. In terms of generalizability, we train our model on one dataset and tested it on multiple datasets. The standard deviation of AUROC scores for different test datasets is calculated to measure the variability of performance across datasets. Our model exhibits superior generalization across datasets from diverse sources. CONCLUSIONS: Our model shows promising results for the generalizable diagnosis of chest X-rays. The impacts of using the attention mechanism and the self-consistency loss in our method are evident from the results. In the future, we plan to incorporate Explainable AI techniques to provide explanations for model decisions. Additionally, we aim to design data augmentation techniques to reduce class imbalance in our model.

Asunto(s)

RESUMEN

PURPOSE: To determine the performance of volumetric dual energy low kV and iodine radiomic features for the differentiation of intrathoracic lymph node histopathology, and influence of contrast protocol. MATERIALS AND METHODS: Intrathoracic lymph nodes with histopathologic correlation (neoplastic, granulomatous sarcoid, benign) within 90 days of DECT chest imaging were volumetrically segmented. 1691 volumetric radiomic features were extracted from iodine maps and low-kV images, totaling 3382 features. Univariate analysis was performed using 2-sample t-test and filtered for false discoveries. Multivariable analysis was used to compute AUCs for lymph node classification tasks. RESULTS: 129 lymph nodes from 72 individuals (mean age 61 ± 15 years) were included, 52 neoplastic, 51 benign, and 26 granulomatous-sarcoid. Among all contrast enhanced DECT protocol exams (routine, PE and CTA), univariable analysis demonstrated no significant differences in iodine and low kV features between neoplastic and non-neoplastic lymph nodes; in the subset of neoplastic versus benign lymph nodes with routine DECT protocol, 199 features differed (p = .01- < 0.05). Multivariable analysis using both iodine and low kV features yielded AUCs >0.8 for differentiating neoplastic from non-neoplastic lymph nodes (AUC 0.86), including subsets of neoplastic from granulomatous (AUC 0.86) and neoplastic from benign (AUC 0.9) lymph nodes, among all contrast protocols. CONCLUSIONS: Volumetric DECT radiomic features demonstrate strong collective performance in differentiation of neoplastic from non-neoplastic intrathoracic lymph nodes, and are influenced by contrast protocol.

Asunto(s)

RESUMEN

Background and Objectives: This study's objective was to investigate the influence of increased scan speed and pitch on image quality and nodule volumetry in patients who underwent ultra-low-dose chest computed tomography (CT). Material and Methods: One hundred and two patients who had lung nodules were included in this study. Standard-speed, standard-pitch (SSSP) ultra-low-dose CT and high-speed, high-pitch (HSHP) ultra-low-dose CT were obtained for all patients. Image noise was measured as the standard deviation of attenuation. One hundred and sixty-three nodules were identified and classified according to location, volume, and nodule type. Volume measurement of detected pulmonary nodules was compared according to nodule location, volume, and nodule type. Motion artifacts at the right middle lobe, the lingular segment, and both lower lobes near the lung bases were evaluated. Subjective image quality analysis was also performed. Results: The HSHP CT scan demonstrated decreased motion artifacts at the left upper lobe lingular segment and left lower lobe compared to the SSSP CT scan (p < 0.001). The image noise was higher and the radiation dose was lower in the HSHP scan (p < 0.001). According to the nodule type, the absolute relative volume difference was significantly higher in ground glass opacity nodules compared with those of part-solid and solid nodules (p < 0.001). Conclusion: Our study results suggest that HSHP ultra-low-dose chest CT scans provide decreased motion artifacts and lower radiation doses compared to SSSP ultra-low-dose chest CT. However, lung nodule volumetry should be performed with caution for ground glass opacity nodules.

Asunto(s)

RESUMEN

Background and Objectives: This study aims to analyze survival in peritoneal and hemodialysis patients using chest radiography and biochemical parameters, determine common dialysis etiologies and causes of death, reveal prognostic factors, and contribute to clinical practice. Materials and Methods: A retrospective cross-sectional study was conducted with data from 33 peritoneal dialysis and 37 hemodialysis patients collected between October 2018 and February 2020. Survival and mortality were retrospectively tracked over 70 months (October 2018-June 2024). Chest X-ray measurements (cardiothoracic index, pulmonary vascular pedicle width, right pulmonary artery diameter, diaphragmatic height) and biochemical parameters (urea, albumin, creatinine, parathormone, ferritin, hemoglobin, arterial blood gas, potassium) were analyzed for their impact on survival. Statistical analyses included descriptive statistics, chi-square test, Fisher's exact test, Bayesian analysis, McNemar test, Kaplan-Meier survival analysis, Cox regression, Bayesian correlation test, linear regression analysis (scatter plot), and ROC analysis. SPSS 20.0 was used for data analysis, with p < 0.05 considered statistically significant. Results: Hypertension, type 2 diabetes, and urogenital disorders were the main dialysis etiologies. Peritonitis (38.5%) and cardiovascular diseases (47.4%) were the leading causes of death in peritoneal and hemodialysis patients, respectively. Significant chest X-ray differences included pulmonary vascular pedicle width and pulmonary artery diameter in hemodialysis and diaphragm height in peritoneal dialysis. Kaplan-Meier showed no survival difference between methods. Cox regression identified age, intact parathormone levels, iPTH/PVPW ratio, and clinical status as survival and mortality factors. The iPTH/PVPW ratio cut-off for mortality prediction was ≤6.8. Conclusions: Age, intact parathormone levels, pulmonary vascular pedicle width, and clinical status significantly impact survival in dialysis patients. Management of hypertension and diabetes, management and follow-up of urogenital disorders, infection control, patient education, and regular cardiovascular check-ups may improve survival rates. Additionally, the iPTH/PVPW ratio can predict mortality risk.

Asunto(s)