X-ray Interpretation Research Articles

Development of Nanocrystalline Graphite from Lignin Sources

Carbon composites are attractive to a variety of high-impact applications, such as carbon fibers, batteries, and vehicle parts, due to their multifunctional properties. The properties of carbon are highly dependent on the allotrope the carbon takes and the functionality, impurities, and defects contained within the structure. The increase in demand for sustainable carbon sources in energy storage devices motivates interest in understanding synthesis parameters of lignin value-added products. Also, as the dependence on oil for fuel decreases, alternative sources for carbon in many applications will be needed. In this work, the thermochemical conversion of lignin powders from different feedstocks was evaluated via small and wide-angle X-ray scattering techniques to resolve the amorphous, disordered, and crystalline domains present in the lignin carbons. Scattering analyses indicated an evolution of hierarchical structures along with an increase in ordered domains as a function of carbonization temperature. Qualitative and quantitative methods were used to describe isotropic scattering intensity profiles at multiple length scales. The use of power law models in the mesoscopic region served as the basis to describe morphological changes related to structural features, for example, graphene stacking, degree of roughness, and surface fractals. Kraft softwood and switchgrass produced carbon powder with the most crystalline domains and the least surface roughness. Softwoods reached the highest degree of crystallinity followed by switchgrass samples and had less variability in particle sizes. These results suggest lignin carbons extracted from softwoods and switchgrass are viable substitutes for graphite. Interpretation of X-ray scattering data from lignin carbon powders elucidates feedstock- and processing-dependent morphological features across multiple length scales providing a straightforward framework to evaluate the feasibility of leveraging lignin carbons for producing tunable application-specific materials.

Point-of-Care Ultrasonography for Ankle Injuries in Children.

The aim of the study was to determine whether point-of-care ultrasound (US) can decrease x-rays in children with ankle injuries. Secondary objectives were to determine the test performance characteristics for ankle US, analyze diagnostic errors, and compare US with the Ottawa Ankle Rules (OAR). This was a prospective study of children younger than 21 years presenting to an emergency department with an ankle injury requiring x-rays. Pediatric emergency medicine physicians received a 1-hour training session, performed ankle US with a standardized scanning protocol of the distal tibia and fibula, and described the US as positive, negative, or equivocal for fracture. Ankle x-ray interpretation by a radiologist was the reference standard for fracture. One hundred twenty patients with a mean age of 13.5 (±4.0) years were enrolled. Nine patients (7.5%) had an ankle fracture on x-ray, and 56 patients (47%) had open physes. Ankle US would reduce x-rays by 81 (67.5%), missing 2 intra-articular, nondisplaced, tibial fractures in patients with open physes. Ankle US had a sensitivity of 78% (95% confidence interval [CI], 40%-97%), specificity of 71% (95% CI, 62%-79%), likelihood ratio for a positive test of 2.7 (95% CI, 1.7-4.3), and likelihood ratio for a negative test of 0.31 (95% CI, 0.09-1.07). The OAR would reduce x-rays by 21 (17.5%), missing one fracture. Ultrasound with OAR would reduce x-rays by 20 (17%) with no missed fractures. Point-of-care US has the potential to reduce x-rays for children with ankle injuries; however, nondisplaced, intra-articular tibial fractures may be missed. Ultrasound with OAR may reduce radiographs without missed fractures in this population.

Pre-registration UK diagnostic radiography student ability and confidence in interpretation of chest X-rays

Introduction Chest X-rays are the most frequently requested X-ray imaging in English hospitals. This study aimed to assess final year UK radiography student’s confidence and ability in image interpretation of chest X-rays.
 Methods Thirty-three diagnostic radiography students were invited to assess their confidence and ability in interpreting chest x-rays from a bank of n=10 cases using multiple choice answers. Data analysis included 2x2 contingency tables, Kappa for inter-rater reliability, a Likert scale of confidence for each case, and questions to assess individual interpretation skills and ways to increase the learning of the subject.
 Results Twenty-three students participated in the study. The pooled accuracy achieved was 61% (95% CI 38.4-77.7; k=0.22). The degree of confidence and ability varied depending upon the student and the conditions observed. High confidence was noted with COVID-19 (n=12/23; 52%), lung metastasis (n=14/23; 61%), and pneumothorax (n=13/23; 57%). Low confidence was noted with conditions of consolidation (n=8/23; 35%), haemothorax (n=8/23; 35%), and surgical emphysema (n=8/23; 35%). From the sample n=11 (48%), participants stated they felt they had the knowledge to interpret chest X-rays required for a newly qualified radiographer.
 Conclusion The results demonstrated final year radiography student’s confidence and ability in image interpretation of chest X-rays. Student feedback indicated a preference for learning support through university lectures, online study resources, and time spent with reporting radiographers on clinical practice to improve ability and confidence in interpreting chest X-rays.

The Performance of EDXS at Elevated Sample Temperatures Using a MEMS-Based In Situ TEM Heating System

Since the development of MEMS heating holders, dynamic in-situ experiments at elevated temperatures may be complemented by X-ray spectrometry for chemical analysis. Although the amount of IR radiation is small when compared to furnace holders, the influence of IR radiation emitted from the heating device on the quality of the X-ray spectra is significant. In this work, we systematically examine the influence of infrared (IR) radiation generated by MEMS-based in situ heating systems (DENSsolutions single- and double-tilt holders) on the results and interpretation of energy-dispersive X-ray (EDX) spectra through simulation and measurement. Focal points of interest in this study are the influence of holder geometry, shadowing and orientation with respect to the different emission characteristics of IR and X-ray photons and their interaction with a side-entry and a multi-detector system. IR photons substantially contribute to count rates, dead time, electronic noise levels, energy resolution, and detection efficiency of semiconductor detectors. At higher sample temperatures, they ultimately limit the feasibility of EDXS for elemental characterization and especially the traceability of low-Z elements. This work provides a quantitative insight into the influence of all relevant parameters related to in situ heating experiments on the spectral quality. Bearing this in mind, we aim to provide a guide to optimizing in situ heating experiments with respect to chemical EDXS analysis.

Evidence of synergistic electrocatalysis at a cobalt oxide–graphene interface through nanochemical mapping of scanning transmission X-ray microscopy

Free-standing graphene membranes are a promising candidate for use as in-situ environmental windows in X-ray (electron) microscopy. In this study, graphene membranes were used as the working electrode, and cobalt nanoparticles (Co-NPs) were grown directly on top of the graphene through electrochemical deposition for interfacial variation. The electronic structure and the chemical bonding states of the Co-NPs and the graphene materials were examined by using the high spatial resolution and element-specific properties of scanning transmission X-ray microscopy. X-ray absorption spectra of C, O, and Co revealed that the Co-NP size increased in accordance with the oxidation state (Co0/2+/3+), depending on the configuration of carbon bonding (C–C/C–OH/HO–C=O/O–C(O)–O/C=O–like state). We conducted a spectral comparison of the dipped graphene and the electrodeposited Co–graphene sample, which revealed an increase in C–OH formation before Co-NPs growth. In addition to electron transfer and electrochemical reduction, the oxidation evolution from C–OH to HO–C=O (or a defect) and the O–C(O)–O (or C=O) state paralleled the increase in Co-NPs size. We curve-fitted the results to demonstrate the reduction in chemical structure from mixing Co2+/3+ to Co3+/2+/0, and to explain the interfacial modulation and the unique metal Co0 layer on the surface of the Co-NPs. Our results provide information for the design of a reliable graphene window and offer an example for the interpretation of experimental X-ray (electron) microscopy.

Pitfalls in X-ray absorption spectroscopy analysis and interpretation: A practical guide for general users

Despite the growing popularity of X-ray absorption spectroscopy (XAS) in scientific research, many researchers do not receive formalized training on this technique. Some of them learned from online resources, which only briefly introduce XAS and its applications. Here, this article aims to provide the overview of tips about the XAS analysis, general rules, as well as required information for presenting XAS data in publications, and some common mistakes in XAS data interpretations. Armed with these basics, the motivated aspiring XAS researchers will find existing resources more accessible and can progress much faster in understanding and using XAS.

Rates and clinical impact of discordant X-ray and CT imaging in transfers to a pediatric emergency department.

Children are often transferred to a Pediatric Emergency Department (PED) for definitive care after completion of diagnostic imaging. There is a paucity of data on the concordance rates of interpretation of imaging studies between referral and PED. Our objective is to describe the rates and clinical impact of discordant interpretation of X-rays and CT in children transferred to a PED. This was a retrospective cohort study of patients over a 12-month period from 12/1/2017-11/30/2018 with X-ray (XR) and CT performed prior to transfer to our PED. We compared referral radiology interpretations to those of pediatric radiologists to determine concordance. Encounters with discordant imaging interpretations were further evaluated for clinical impact (none, minor or major) based on need for additional laboratory workup, consultation, and changes in management and disposition. We analyzed 899 patient encounters. There were high rates of concordance in both XR and CT interpretation (668/743; 89.9%, 95% CI 0.87-0.91 and 205/235; 87.2%, 95% CI 0.82-0.91, respectively). XR discordance resulted in minor clinical impact in 34 patients (45%, 95% CI 0.35-0.57) and a major clinical impact in 28 patients (37%, 95% CI 0.27-0.49). CT discordance resulted in minor clinical impact in 10 patients (33%, 95% CI 0.19-0.51) of patients and major clinical impact in 15 patients (50%, 95% CI 0.33-0.67). The most common discordances with major clinical impact were related to pneumonia on XR chest and appendicitis or inflammatory bowel disease on CT abdomen. In patients transferred to the PED, concordance of XR and CT interpretations was high. A majority of discordant interpretations led to clinical impact meaningful to the patient and emergency medicine (EM) physician. Referring EM physicians might consider the benefit of pediatric radiology consultation upon transfer, especially for imaging diagnoses related to pneumonia, appendicitis, or inflammatory bowel disease.

Interpreting Uncertainty in Model Predictions for Covid-19 Diagnosis.

COVID-19, due to its accelerated spread has brought in the need to use assistive tools for faster diagnosis in addition to typical lab swab testing. Chest X-Rays for COVID cases tend to show changes in the lungs such as ground glass opacities and peripheral consolidations which can be detected by deep neural networks. However, traditional convolutional networks use point estimate for predictions, lacking in capture of uncertainty, which makes them less reliable for adoption. There have been several works so far in predicting COVID positive cases with chest X-Rays. However, not much has been explored on quantifying the uncertainty of these predictions, interpreting uncertainty, and decomposing this to model or data uncertainty. To address these needs, we develop a visualization framework to address interpretability of uncertainty and its components, with uncertainty in predictions computed with a Bayesian Convolutional Neural Network. This framework aims to understand the contribution of individual features in the Chest-X-Ray images to predictive uncertainty. Providing this as an assistive tool can help the radiologist understand why the model came up with a prediction and whether the regions of interest captured by the model for the specific prediction are of significance in diagnosis. We demonstrate the usefulness of the tool in chest x-ray interpretation through several test cases from a benchmark dataset.

Ab initio thermodynamics reveals the nanocomposite structure of ferrihydrite

Ferrihydrite is a poorly crystalline iron oxyhydroxide nanomineral that serves a critical role as the most bioavailable form of ferric iron for living systems. However, its atomic structure and composition remain unclear due in part to ambiguities in interpretation of X-ray scattering results. Prevailing models so far have not considered the prospect that at the level of individual nanoparticles multiple X-ray indistinguishable phases could coexist. Using ab initio thermodynamics we show that ferrihydrite is likely a nanocomposite of distinct structure types whose distribution depends on particle size, temperature, and hydration. Nanoparticles of two contrasting single-phase ferrihydrite models of Michel and Manceau are here shown to be thermodynamically equivalent across a wide range of temperature and pressure conditions despite differences in their structural water content. Higher temperature and water pressure favor the formation of the former, while lower temperature and water pressure favor the latter. For aqueous suspensions at ambient conditions, their coexistence is maximal for particle sizes up to 12 nm. The predictions inform and help resolve different observations in various experiments.

MACHINE LEARNING BASED CHEST X-RAY ABNORMALITIES DETECATION SYSTEM

When you have a broken arm, radiologists help save the day —and the bone. These doctors diagnose and treat medical conditions using imaging techniques like CT and PET scans, MRIs, and, of course, X -rays. Yet, as it happens when working with such a wide variety of medical tools, radiologists face many daily challenges, perhaps the most difficult being the chest radiograph. The interpretation of chest X-rays can lead to medical misdiagnosis, even for the best practicing doctor. Computer-aided detection and diagnosis systems (CADe/CADx) would help reduce the pressure on doctors at metropolitan hospitals and improve diagnostic quality in rural areas. Existing methods of interpreting chest X-ray images classify them into a list of findings. There is currently no specification of their locations on the image which sometimes leads to inexplicable results. A solution for localizing findings on chest X-ray images is needed for providing doctors with more meaningful diagnostic assistance.

Effect of a comprehensive deep-learning model on the accuracy of chest x-ray interpretation by radiologists: a retrospective, multireader multicase study

Chest x-rays are widely used in clinical practice; however, interpretation can be hindered by human error and a lack of experienced thoracic radiologists. Deep learning has the potential to improve the accuracy of chest x-ray interpretation. We therefore aimed to assess the accuracy of radiologists with and without the assistance of a deep-learning model. In this retrospective study, a deep-learning model was trained on 821 681 images (284 649 patients) from five data sets from Australia, Europe, and the USA. 2568 enriched chest x-ray cases from adult patients (≥16 years) who had at least one frontal chest x-ray were included in the test dataset; cases were representative of inpatient, outpatient, and emergency settings. 20 radiologists reviewed cases with and without the assistance of the deep-learning model with a 3-month washout period. We assessed the change in accuracy of chest x-ray interpretation across 127 clinical findings when the deep-learning model was used as a decision support by calculating area under the receiver operating characteristic curve (AUC) for each radiologist with and without the deep-learning model. We also compared AUCs for the model alone with those of unassisted radiologists. If the lower bound of the adjusted 95% CI of the difference in AUC between the model and the unassisted radiologists was more than -0·05, the model was considered to be non-inferior for that finding. If the lower bound exceeded 0, the model was considered to be superior. Unassisted radiologists had a macroaveraged AUC of 0·713 (95% CI 0·645-0·785) across the 127 clinical findings, compared with 0·808 (0·763-0·839) when assisted by the model. The deep-learning model statistically significantly improved the classification accuracy of radiologists for 102 (80%) of 127 clinical findings, was statistically non-inferior for 19 (15%) findings, and no findings showed a decrease in accuracy when radiologists used the deep-learning model. Unassisted radiologists had a macroaveraged mean AUC of 0·713 (0·645-0·785) across all findings, compared with 0·957 (0·954-0·959) for the model alone. Model classification alone was significantly more accurate than unassisted radiologists for 117 (94%) of 124 clinical findings predicted by the model and was non-inferior to unassisted radiologists for all other clinical findings. This study shows the potential of a comprehensive deep-learning model to improve chest x-ray interpretation across a large breadth of clinical practice. Annalise.ai.

Chest radiographs and machine learning - Past, present and future.

SummaryDespite its simple acquisition technique, the chest X‐ray remains the most common first‐line imaging tool for chest assessment globally. Recent evidence for image analysis using modern machine learning points to possible improvements in both the efficiency and the accuracy of chest X‐ray interpretation. While promising, these machine learning algorithms have not provided comprehensive assessment of findings in an image and do not account for clinical history or other relevant clinical information. However, the rapid evolution in technology and evidence base for its use suggests that the next generation of comprehensive, well‐tested machine learning algorithms will be a revolution akin to early advances in X‐ray technology. Current use cases, strengths, limitations and applications of chest X‐ray machine learning systems are discussed.

Economic costs of accessing tuberculosis (TB) diagnostic services in Malawi: an analysis of patient costs from a randomised controlled trial of computer-aided chest x-ray interpretation

Background:Patients with tuberculosis (TB) symptoms in low-resource settings face convoluted diagnostic and treatment linkage pathways, incurring substantial health-seeking costs. In the context of a randomised trial looking at the impact of novel diagnostics such as computer-aided chest x-ray diagnosis (CAD4TB), we aimed to investigate the costs incurred by patients seeking TB diagnosis and whether optimised diagnostic interventions could result in a reduction in the cost faced by households.Methods:PROSPECT was a three-arm randomised trial conducted in a public primary health clinic in Blantyre, Malawi during 2018-2019 (trial arms: standard of care [SOC]; HIV testing [HIV]; HIV testing and CAD4TB [HIV/TB]). The direct and indirect costs incurred by 219 PROSPECT participants over the 56-day follow-up period were collected. Costs were deemed catastrophic if they exceeded 20% of annual household income. We compared mean costs and used generalised linear regression models to examine whether the interventions could result in a reduction in total costs.Results:The mean total cost incurred by all 219 participants was US$12.11 (standard error (SE): 1.86). The indirect and direct cost was US$8.47 (SE: 1.66) and US$3.64 (SE: 0.38), respectively. The mean total cost composed of 5.6% of the average annual household income. In total, 5% (9/180) of the participants with complete income data incurred catastrophic costs. Compared to SOC, there was no statistically significant difference in the mean total cost faced by those in the HIV (ratio: 0.77, 95% CI: 0.51, 1.19) and HIV/TB arms (ratio: 0.85, 95% CI: 0.53, 1.37).Conclusions:Despite the absence of user fees, patients seeking healthcare with TB symptoms incurred catastrophic costs. The optimised TB diagnostic interventions that were investigated in the PROSPECT study did not significantly reduce costs. TB diagnosis interventions should be implemented alongside social protection policies whilst ensuring healthcare facilities are accessible by the poor.

Laboratory Investigation on Physical-Mechanical Characteristics and Microstructure of a Clayey Gypsiferous Soil in the Presence of Chemical Accelerator

Gypsiferous soils are considered as problematic soils that cause damage to engineering infrastructures, particularly hydraulic structures. The exposition of gypsiferous soils to the steady flow leads to the dissolution of gypsum particles, significantly changing the soil engineering characteristics. In this paper, by employing physical, chemical, and mechanical experiments, the effect of raw gypsum with various percentages (0%, 3%, 6%, 9%, 12% and 15%) on the geotechnical parameters of a low plasticiy clay was investigated through leaching. Electrical conductivity test and ionic concentration analysis were used to assess the trend of gypsum dissolution. The results showed that leaching and increased gypsum content significantly degraded soil engineering characteristics, such as strength and compressibility. Acidic accelerator (diluted acetic acid) speeded up the chemical reaction rate and, at the same time, affected the geotechnical properties of the soil more than water. Mineralogical and microstructural studies (x-ray diffraction, scanning electron microscopy and energy-dispersive x-ray spectroscopy) were carried out to evaluate the interaction between gypsum and soil particles and verify laboratory test results. The x-ray diffraction patterns indicated the formation of new peaks such as calcium aluminate hydrated (CAH) and calcium silicate hydrated (CSH) and the altered structure of pure soil following the addition of raw gypsum. Scanning electron microscope images confirmed the presence of cementitious compounds and changes in the texture of gypsiferous soil in comparison with pure soil. Energy dispersive x-ray interpretation revealed changes in the chemical composition of soil blended with gypsum. The laboratory test results were corroborated by microstructural analyses.

Clinical aspects of using artificial intelligence for the interpretation of chest X-rays

The review considers the possible use of artificial intelligence for the interpretation of chest X-rays by analyzing 45 publications. Experimental and commercial diagnostic systems for pulmonary tuberculosis, pneumonia, neoplasms and other diseases have been analyzed.

Chest x-ray automated triage: A semiologic approach designed for clinical implementation, exploiting different types of labels through a combination of four Deep Learning architectures

Background and objectivesThe multiple chest x-ray datasets released in the last years have ground-truth labels intended for different computer vision tasks, suggesting that performance in automated chest x-ray interpretation might improve by using a method that can exploit diverse types of annotations. This work presents a Deep Learning method based on the late fusion of different convolutional architectures, that allows training with heterogeneous data with a simple implementation, and evaluates its performance on independent test data. We focused on obtaining a clinically useful tool that could be successfully integrated into a hospital workflow. Materials and methodsBased on expert opinion, we selected four target chest x-ray findings, namely lung opacities, fractures, pneumothorax and pleural effusion. For each finding we defined the most suitable type of ground-truth label, and built four training datasets combining images from public chest x-ray datasets and our institutional archive. We trained four different Deep Learning architectures and combined their outputs with a late fusion strategy, obtaining a unified tool. The performance was measured on two test datasets: an external openly-available dataset, and a retrospective institutional dataset, to estimate performance on the local population. ResultsThe external and local test sets had 4376 and 1064 images, respectively, for which the model showed an area under the Receiver Operating Characteristics curve of 0.75 (95%CI: 0.74–0.76) and 0.87 (95%CI: 0.86–0.89) in the detection of abnormal chest x-rays. For the local population, a sensitivity of 86% (95%CI: 84–90), and a specificity of 88% (95%CI: 86–90) were obtained, with no significant differences between demographic subgroups. We present examples of heatmaps to show the accomplished level of interpretability, examining true and false positives. ConclusionThis study presents a new approach for exploiting heterogeneous labels from different chest x-ray datasets, by choosing Deep Learning architectures according to the radiological characteristics of each pathological finding. We estimated the tool's performance on the local population, obtaining results comparable to state-of-the-art metrics. We believe this approach is closer to the actual reading process of chest x-rays by professionals, and therefore more likely to be successful in a real clinical setting.

Development and Current Status of a Chest X-ray Interpretation Exercise System for Lung Cancer Screening

Development and Current Status of a Chest X-ray Interpretation Exercise System for Lung Cancer Screening

Technical and clinical study of x-ray-based surface echo probe tracking using an attached fiducial apparatus.

Several types of structural heart intervention (SHI) use information from multiple imaging modalities to complete an interventional task. For example, in transcatheter aortic valve replacement (TAVR), placement and deployment of a bioprosthetic aortic valve in the aorta is primarily guided by x-ray fluoroscopy (XRF), and echocardiography provides visualization of cardiac anatomy and blood flow. However, simultaneous interpretation of independent x-ray and echo displays remains a challenge for the interventionalist. The purpose of this work was to develop a novel echo/x-ray co-registration solution in which volumetric transthoracic echo (TTE) is transformed to the x-ray coordinate system by tracking the three-dimensional (3D) pose of a probe fiducial attachment from its appearance in two-dimensional (2D) x-ray images. A fiducial attachment for a commercial TTE probe consisting of rings of high-contrast ball bearings was designed and fabricated. The 3D pose (position and orientation) of the fiducial attachment is estimated from a 2D x-ray image using an algorithm in which a virtual point cloud model of the attachment is iteratively rotated, translated, and forward-projected onto the image until the average sum-of-squares of grayscale values at the projected points is minimized. Fiducial registration error (FRE) and target registration error (TRE) of this approach were evaluated in phantom studies using TAVR-relevant gantry orientations and four standard acoustic windows for the TTE probe. A patient study was conducted to assess the clinical suitability of the fiducial attachment prototype during TTE imaging of patients undergoing SHI. TTE image quality for the task of guiding a transcatheter procedure was evaluated in a reviewer study. The 3D FRE ranged from 0.32±0.03mm (mean±SD) to 1.31±0.05mm, depending on C-arm orientation and probe acoustic window. The 3D TRE ranged from 1.06±0.03mm to 2.42±0.06mm. Fiducial pose estimation was stable when >75% of the fiducial markers were visible in the x-ray image. A panel of reviewers graded the presentation of heart valves in TTE images from 48 SHI patients. While valve presentation did not differ significantly between acoustic windows (P>0.05), the mitral valve did achieve a significantly higher image quality compared to the aortic and tricuspid valves (P<0.001). Overall, reviewers perceived sufficient image quality in 76.5% of images of the mitral valve, 54.9% of images of the aortic valve, and 48.6% of images of the tricuspid valve. Fiducial-based tracking of a commercial TTE probe is compatible with clinical SHI workflows and yields 3D target registration error of less than 2.5mm for a variety of x-ray gantry geometries and echo probe acoustic windows. Although TTE image quality with respect to target valve anatomy was sufficient for the majority of cases examined, prescreening of patients for sufficient TTE quality would be helpful.

Training focal lung pathology detection using an eye movement modeling example.

Purpose: Published reports suggest that nonoptimal visual search behavior is associated with false negatives in chest x-ray interpretation. Eye movement modeling example (EMME)-based training interventions, that is, interventions showing models of visual search to trainees, have been shown to improve visual search as well as task accuracy. Approach: We examined the detection of focal lung pathology on chest x-rays before and after two different EMME training interventions that have been shown to be efficient: (i)an EMME showing moving fixations on a blurred background (spotlight group) and (ii)an EMME showing moving fixations on a nonblurred background (circle group). These two experimental groups were compared to a control group that was only provided with the correct location of pathologies on the chest x-rays. Results: Performance outcomes showed improved detection sensitivity and specificity in all groups (also the control group). It appears that repetitive exposure to pathologies on chest x-rays with feedback resulted in enhanced pattern recognition. In addition, visual search strategies became more efficient during post-tests. Conclusion: Repetitive exposure to a focal lung pathology detection task with feedback improves overall performance. However, the specific EMME training interventions did not add any further advantages. Similar training interventions can be provided online to assess feasibility and reproducibility of such (or similar) training formats. Such training can, for example, reduce the number of false negative errors, especially for novices.

Learning Using Partially Available Privileged Information and Label Uncertainty: Application in Detection of Acute Respiratory Distress Syndrome.

Acute respiratory distress syndrome (ARDS) is a fulminant inflammatory lung injury that develops in patients with critical illnesses, affecting 200,000 patients in the United States annually. However, a recent study suggests that most patients with ARDS are diagnosed late or missed completely and fail to receive life-saving treatments. This is primarily due to the dependency of current diagnosis criteria on chest x-ray, which is not necessarily available at the time of diagnosis. In machine learning, such an information is known as Privileged Information - information that is available at training but not at testing. However, in diagnosing ARDS, privileged information (chest x-rays) are sometimes only available for a portion of the training data. To address this issue, the Learning Using Partially Available Privileged Information (LUPAPI) paradigm is proposed. As there are multiple ways to incorporate partially available privileged information, three models built on classical SVM are described. Another complexity of diagnosing ARDS is the uncertainty in clinical interpretation of chest x-rays. To address this, the LUPAPI framework is then extended to incorporate label uncertainty, resulting in a novel and comprehensive machine learning paradigm - Learning Using Label Uncertainty and Partially Available Privileged Information (LULUPAPI). The proposed frameworks use Electronic Health Record (EHR) data as regular information, chest x-rays as partially available privileged information, and clinicians' confidence levels in ARDS diagnosis as a measure of label uncertainty. Experiments on an ARDS dataset demonstrate that both the LUPAPI and LULUPAPI models outperform SVM, with LULUPAPI performing better than LUPAPI.