Validation study of machine-learning chest radiograph software in primary and emergency medicine

AIM

To evaluate the performance of a machine learning based algorithm tool for chest radiographs (CXRs), applied to a consecutive cohort of historical clinical cases, in comparison to expert chest radiologists.

MATERIALS AND METHODS

The study comprised 1,960 consecutive CXR from primary care referrals and the emergency department (992 and 968 cases respectively), obtained in 2015 at a UK hospital. Two chest radiologists, each with >20 years of experience independently read all studies in consensus to serve as a reference standard. A chest artificial intelligence (AI) algorithm, Lunit INSIGHT CXR, was run on the CXRs, and results were correlated with those by the expert readers. The area under the receiver operating characteristic curve (AUROC) was calculated for the normal and 10 common findings: atelectasis, fibrosis, calcification, consolidation, lung nodules, cardiomegaly, mediastinal widening, pleural effusion, pneumothorax, and pneumoperitoneum.

RESULTS

The ground truth annotation identified 398 primary care and 578 emergency department datasets containing pathologies. The AI algorithm showed AUROC of 0.881-0.999 in the emergency department dataset and 0.881-0.998 in the primary care dataset. The AUROC for each of the findings between the primary care and emergency department datasets did not differ, except for pleural effusion (0.954 versus 0.988, p<0.001).

CONCLUSIONS

The AI algorithm can accurately and consistently differentiate normal from major thoracic abnormalities in both acute and non-acute settings, and can serve as a triage tool.

Coronary artery calcification is associated with mortality independent of pulmonary embolism severity: a retrospective cohort study

AIM

To assess coronary artery calcification (CAC) and vascular calcification in patients with pulmonary embolism (PE) and correlate this with mortality.

MATERIALS AND METHODS

PE severity was quantified using computed tomography pulmonary angiography (CTPA) in 400 consecutive cases using the modified Miller score (1-5, mild; 6-11, moderate; 12-16, severe). Right ventricle strain was assessed using the right/left ventricle diameter (RV/LV) ratio. CAC score (CACS) was assessed using a four-point scale (CACS mild 1-3, moderate 4-8, severe 9-12) for each vessel and summed to give the total CACS. Follow-up for mortality was obtained at 3 years.

RESULTS

PE severity was classified as mild in 48%, moderate in 21%, and severe in 32% of cases. The median modified Miller score was 6 (Interquartile range [IQR] 2, 14) and median total CACS was 2 (IQR 0, 7). All-cause mortality occurred in 128 (32%) patients. Patients with CAC were three times more likely to die than patients without CAC (Hazard ratio [HR] 2.96; 95% CI 1.84, 4.77; p<0.001), and patients with severe CAC were at the highest risk (HR 4.62; 95% CI 2.73, 7.83, p<0.001). Gender, modified Miller score and RV/LV ratio were not predictive of mortality. In multivariate analysis both CACS and age were independent predictors of 3-year all-cause mortality. Of the patients with CAC who died, the presence of coronary artery disease was only documented in 34 (32%).

CONCLUSION

CACS is an independent predictor of all-cause mortality in patients with PE, and has important implications for subsequent patient management.

Effects of Heparin Therapy on Fibrinolysis in Patients with Pulmonary Embolism

Previous investigations suggested that heparin administration to humans enhances the tissue type plasminogen activator (tPA) levels in blood, but it remains uncertain whether this effect induces fibrinolysis. We studied the effect of therapeutic levels of heparinization on plasma markers for fibrinolysis in patients suspected of pulmonary embolism (PE). Blood samples were taken from 49 consecutive patients; 28 had confirmed PE, 21 had PE excluded.

On admission, the plasma levels of plasmin-α2antiplasmin complexes and D-dimer were significantly higher in the patient group with PE compared to those in whom PE was excluded. After heparinization the tPA levels increased in both groups, showing that this effect was not dependent on the initial level of activity of fibrinolysis. In spite of this increment in tPA levels, the concentrations of plasmin-α2antiplasmin complexes and D-dimer decreased.

In conclusion, although heparinization in patients with or without pulmonary embolism does lead to elevated tPA: Ag levels, this is not accompanied by enhanced fibrinolysis.

Multimodality quantitative assessments of myocardial perfusion using dynamic contrast enhanced magnetic resonance and 15O-labelled water positron emission tomography imaging

Kinetic modelling of myocardial perfusion imaging data allows the absolute quantification of myocardial blood flow (MBF) and can improve the diagnosis and clinical assessment of coronary artery disease (CAD). Positron emission tomography (PET) imaging is considered the reference standard technique for absolute quantification, whilst oxygen-15 (¹⁵O)-water has been extensively implemented for MBF quantification. Dynamic contrast enhanced magnetic resonance imaging (DCE-MRI) has also been used for MBF quantification and showed comparable diagnostic performance against (¹⁵O)-water PET studies. We investigated for the first time the diagnostic performance of two different PET MBF analysis softwares PMOD and Carimas, for obstructive CAD detection against invasive clinical standard methods in 20 patients with known or suspected CAD. Fermi and distributed parameter modelling-derived MBF quantification from DCE-MRI was also compared against (¹⁵O)-water PET, in a subgroup of 6 patients. The sensitivity and specificity for PMOD was significantly superior for obstructive CAD detection in both per vessel (0.83, 0.90) and per patient (0.86, 0.75) analysis, against Carimas (0.75, 0.65), (0.81, 0.70), respectively. We showed strong, significant correlations between MR and PET MBF quantifications (r=0.83-0.92). However, DP and PMOD analysis demonstrated comparable and higher haemodynamic differences between obstructive versus (no, minor or non)-obstructive CAD, against Fermi and Carimas analysis. Our MR method assessments against the optimum PET reference standard technique for perfusion analysis showed promising results in per segment level and can support further multi-modality assessments in larger patient cohorts. Further MR against PET assessments may help to determine their comparative diagnostic performance for obstructive CAD detection.

Imaging of cardiovascular risk in patients with Turner's syndrome

Turner's syndrome is a disorder defined by an absent or structurally abnormal second X chromosome and affects around 1 in 2000 newborn females. The standardised mortality ratio in Turner's syndrome is around three-times higher than in the general female population, mainly as a result of cardiovascular disorders. Most striking is the early age at which Turner's syndrome patients develop the life-threatening complications of cardiovascular disorders compared to the general population. The cardiovascular risk stratification in Turner's syndrome is challenging and imaging is not systematically used. The aim of this article is to review cardiovascular risks in this group of patients and discuss a systematic imaging approach for early identification of cardiovascular disorders in these patients.

Retinal imaging as a source of biomarkers for diagnosis, characterization and prognosis of chronic illness or long-term conditions

The black void behind the pupil was optically impenetrable before the invention of the ophthalmoscope by von Helmholtz over 150 years ago. Advances in retinal imaging and image processing, especially over the past decade, have opened a route to another unexplored landscape, the retinal neurovascular architecture and the retinal ganglion pathways linking to the central nervous system beyond. Exploiting these research opportunities requires multidisciplinary teams to explore the interface sitting at the border between ophthalmology, neurology and computing science. It is from the detail and depth of retinal phenotyping that novel metrics and candidate biomarkers are likely to emerge. Confirmation that in vivo retinal neurovascular measures are predictive of microvascular change in the brain and other organs is likely to be a major area of research activity over the next decade. Unlocking this hidden potential within the retina requires integration of structural and functional data sets, that is, multimodal mapping and longitudinal studies spanning the natural history of the disease process. And with further advances in imaging, it is likely that this area of retinal research will remain active and clinically relevant for many years to come. Accordingly, this review looks at state-of-the-art retinal imaging and its application to diagnosis, characterization and prognosis of chronic illness or long-term conditions.

Iterative reconstruction and individualized automatic tube current selection reduce radiation dose while maintaining image quality in 320-multidetector computed tomography coronary angiography

AIM

To assess the effect of two iterative reconstruction algorithms (AIDR and AIDR3D) and individualized automatic tube current selection on radiation dose and image quality in computed tomography coronary angiography (CTCA).

MATERIALS AND METHODS

In a single-centre cohort study, 942 patients underwent electrocardiogram-gated CTCA using a 320-multidetector CT system. Images from group 1 (n = 228) were reconstructed with a filtered back projection algorithm (Quantum Denoising Software, QDS+). Iterative reconstruction was used for group 2 (AIDR, n = 379) and group 3 (AIDR3D, n = 335). Tube current was selected based on body mass index (BMI) for groups 1 and 2, and selected automatically based on scout image attenuation for group 3. Subjective image quality was graded on a four-point scale (1 = excellent, 4 = non-diagnostic).

RESULTS

There were no differences in age (p = 0.975), body mass index (p = 0.435), or heart rate (p = 0.746) between the groups. Image quality improved with iterative reconstruction and automatic tube current selection [1.3 (95% confidence intervals (CI): 1.2-1.4), 1.2 (1.1-1.2) and 1.1 (1-1.2) respectively; p < 0.001] and radiation dose decreased [274 (260-290), 242 (230-253) and 168 (156-180) mGy cm, respectively; p < 0.001].

CONCLUSION

The application of the latest iterative reconstruction algorithm and individualized automatic tube current selection can substantially reduce radiation dose whilst improving image quality in CTCA.

Hyperpolarised 3He MRI versus HRCT in COPD and normal volunteers: PHIL trial

The aim of the present study was to apply hyperpolarised (HP) (3)He magnetic resonance imaging (MRI) to identify patients with chronic obstructive pulmonary disease (COPD) and alpha(1)-antitrypsin deficiency (alpha(1)-ATD) from healthy volunteers and compare HP (3)He MRI findings with high-resolution computed tomography (HRCT) in a multicentre study. Quantitative measurements of HP (3)He MRI (apparent diffusion coefficient (ADC)) and HRCT (mean lung density (MLD)) were correlated with pulmonary function tests. A prospective three centre study enrolled 122 subjects with COPD (either acquired or genetic) and age-matched never-smokers. All diagnostic studies were completed in 94 subjects (52 with COPD; 13 with alpha(1)-ATD; 29 healthy subjects; 63 males; and 31 females; median age 62 yrs). The consensus assessment of radiologists, blinded for other test results, estimated nonventilated lung volume (HP (3)He MRI) and percentage diseased lung (HRCT). Quantitative evaluation of all data for each centre consisted of ADC (HP (3)He MRI) and MLD measurements (HRCT), and correlation with forced expiratory volume in 1 s (FEV(1))/forced vital capacity (FVC) indicating airway obstruction, and the diffusing capacity of the lung for carbon monoxide (D(L,CO)) indicating alveolar destruction. Using lung function tests as a reference, regional analysis of HP (3)He MRI and HRCT correctly categorised normal volunteers in 100% and 97%, COPD in 42% and 69% and alpha(1)-ATD in 69% and 85% of cases, respectively. Direct comparison of HP (3)He MRI and CT revealed 23% of subjects with moderate/severe structural abnormalities had only mild ventilation defects. In comparison with lung function tests, ADC was more effective in separating COPD patients from healthy subjects than MLD (p<0.001 versus 0.038). ADC measurements showed better correlation with D(L,CO) than MLD (r = 0.59 versus 0.29). Hyperpolarised (3)He MRI correctly categorised patients with COPD and normal volunteers. It offers additional functional information, without the use of ionising radiation whereas HRCT gives better morphological information. We showed the feasibility of a multicentre study using different magnetic resonance systems.

Efficacy and safety of gadodiamide (Gd-DTPA-BMA) in renal 3D-magnetic resonance angiography (MRA): a phase II study

PURPOSE

To determine the most efficacious dose of gadodiamide for three-dimensional (3D) contrast-enhanced (CE) magnetic resonance angiography (MRA) of the renal arteries on a patient level based on the sensitivity in detecting the main hemodynamically relevant (> or =50% or occlusion) renal artery stenosis (RAS) using intra-arterial digital subtraction angiography (IA DSA) as the gold standard.

MATERIALS AND METHODS

This prospective, randomized, double-blind, parallel-group, multicenter study included 273 patients referred to IA DSA for suspected RAS. Patients underwent 3D CE MRA after injection of 0.01, 0.05, 0.1, or 0.2mmol/kg of body weight gadodiamide (0.5mmol/ml). The images were assessed for location and degree of RAS by independent blinded readers (MRA: three readers, IA DSA: one reader). Hypothesis testing for a significant trend in sensitivity across dose groups was based on the one-sided Cochran-Armitage style trend test for each independent MRA reader.

RESULTS

The lowest dose group (0.01mmol/kg) proved non-efficacious in detecting hemodynamically relevant (i.e., > or =50% or occlusion) RAS. A statistically significant dose trend (p<0.001) was shown for each of the three independent readers. Depending on reader, the sensitivity obtained with 0.05, 0.1, and 0.2mmol/kg was 63.9-86.1%, 75.8-91.4% and 80.6-90.6%, the specificity was 66.7-73.9%, 59.3-75.0%, and 59.3-75.0% and accuracy was 67.8-78.9%, 75.4-77.4%, and 76.3-81.0%, for the three dose groups, respectively. There were eight non-severe adverse events (AEs). Three serious AEs occurring in one patient were judged not related to gadodiamide by the on-site investigator.

CONCLUSION

A significant dose trend between the four doses examined was observed. The lowest dose (0.01mmol/kg) differed significantly from those of the other three doses. Based on the analysis of the primary and secondary endpoints, 0.1mmol/kg gadodiamide appears to be the most suitable dose in diagnosing hemodynamically relevant RAS. The present study also demonstrated gadodiamide to be safe and well tolerated.

The clinical impact of a combined gamma camera/CT imaging system on somatostatin receptor imaging of neuroendocrine tumours

AIM

With a combined gamma camera/CT imaging system, CT images are obtained which are inherently registered to the emission images and can be used for the attenuation correction of SPECT and for mapping the functional information from these nuclear medicine tomograms onto anatomy. The aim of this study was to evaluate the clinical impact of SPECT/CT using such a system for somatostatin receptor imaging (SRI) of neuroendocrine tumours.

MATERIALS AND METHODS

SPECT/CT imaging with (111)In-Pentetreotide was performed on 29 consecutive patients, the majority of whom had carcinoid disease. All SPECT images were first reported in isolation and then re-reported with the addition of the CT images for functional anatomical mapping (FAM).

RESULTS

Fifteen of the 29 SPECT images were reported as abnormal, and in 11 of these abnormal images (73%) FAM was found to either establish a previously unknown location (7/11) or change the location (4/11) of at least one lesion. The revised location could be independently confirmed in 64% of these cases. Confirmation of location was not possible in the other patients due to either a lack of other relevant investigations, or the fact that lesions seen in the SPECT images were not apparent in the other investigations. FAM affected patient management in 64% of the cases where the additional anatomical information caused a change in the reported location of lesions.

CONCLUSION

These results imply that FAM can improve the reporting accuracy for SPECT SRI with significant impact on patient management.

Hyperpolarized 3-helium MR imaging of the lungs: testing the concept of a central production facility

The aim of this study was to test the feasibility of a central production facility with distribution network for implementation of hyperpolarized 3-helium MRI. The 3-helium was hyperpolarized to 50-65% using a large-scale production facility based at a university in Germany. Using a specially designed transport box, containing a permanent low-field shielded magnet and dedicated iron-free glass cells, the hyperpolarized 3-helium gas was transported via airfreight to a university in the UK. At this location, the gas was used to perform in vivo MR experiments in normal volunteers and patients with chronic obstructive lung diseases. Following initial tests, the transport (road-air-road cargo) was successfully arranged on six occasions (approximately once per month). The duration of transport to imaging averaged 18 h (range 16-20 h), which was due mainly to organizational issues such as working times and flight connections. During the course of the project, polarization at imaging increased from 20% to more than 30%. A total of 4 healthy volunteers and 8 patients with chronic obstructive pulmonary disease were imaged. The feasibility of a central production facility for hyperpolarized 3-helium was demonstrated. This should enable a wider distribution of gas for this novel technology without the need for local start-up costs.

Radiological imaging as the basis for a simulation software of ventilation in the tracheo-bronchial tree

The inhaled route is a promising new way for administering drugs to the human body. Flow and particle deposition in the human respiratory tract depends on the individual's anatomy as well as on the drug composition. A European Framework V Program supported project is currently developing a simulation tool for assessment of drug distribution and deposition. This tool relies heavily on the input of radiological data sets, which are obtained in humans. Both high temporal and spatial resolutions are required, and CT and MRI (including hyperpolarized helium-3 MRI) are applied. The radiological data are integrated into computation fluid dynamics software, which is capable of assessing air-flow profiles and compartmental behaviours. This is complemented by pharmacokinetic models, which should result in a simulation tool that will be of use for the theoretical design of new inhaled therapies. This article describes the special imaging requirements of each region of the respiratory tract and the feasibility of these sophisticated radiological techniques with a view of using these data in a simulation model of the lung.

MR imaging of the lungs with hyperpolarized helium-3 gas transported by air

Hyperpolarized noble gas MRI shows promise in the functional imaging of the pulmonary air spaces. The production of hyperpolarized (HP) gas requires specialized laser optical pumping apparatus, which is not likely to be home built in the majority of clinical MRI radiology centres. There are two routes through which HP gas will be made available to hospitals for clinical use: either the apparatus will be installed locally at a considerable expense to the centre, or a central facility will produce the gas and then deliver it to remote MRI sites as and when required. In this study, the feasibility of transporting large quantities of HP gas for in vivo MR imaging from a remote production facility in Mainz, Germany, by airfreight to Sheffield, UK, was successfully demonstrated.

Magnetic resonance angiography for the diagnosis of renal artery stenosis: a meta-analysis

AIM

To review the published literature comparing the diagnostic accuracy of magnetic resonance angiography (MRA) with and without gadolinium in diagnosing renal artery stenosis, using catheter angiography as reference.

MATERIALS AND METHODS

A meta-analysis was performed of English language articles identified by computer search using PubMed/MEDLINE, followed by extensive bibliography review from 1985 to May 2001. Inclusion criteria were: (1) blinded comparison with catheter angiography; (2)indication for MRA stated; (3) clear descriptions of imaging techniques; and (4) interval between MRA and catheter angiography < 3 months and only the largest of all studies from one centre was selected in the analysis.

RESULTS

A total of 39 studies were identified, of which 25 met the inclusion criteria. The number of patients included in the meta-analysis was 998: 499 with non-enhanced MRA and 499 with gadolinium-enhanced MRA. The sensitivity and specificity of non-enhanced MRA were 94% (95% CI: 90-97%) and 85% (95% CI: 82-87%), respectively. For gadolinium-enhanced MRA sensitivity was 97% (95% CI: 93-98%) and specificity was 93% (95% CI: 91-95%). Thus, specificity and positive predictive value were significantly better for gadolinium-enhanced MRA (P < 0.001). Accessory renal arteries were depicted better by gadolinium-enhanced MRA (82%; 95% CI: 75-87%) than non-gadolinium MRA (49%; 95% CI: 42-60%) (P < 0.001).

CONCLUSIONS

Gadolinium-enhanced MRA may replace arteriography in most patients with suspected renal artery stenosis, and has major advantages in that it is non-invasive, avoids ionizing radiation and uses a non-nephrotoxic contrast agent.