Machine learning in action: Revolutionizing intracranial hematoma detection and patient transport decision-making.

Objectives: Traumatic intracranial hematomas represent a critical clinical situation where early detection and management are of utmost importance. Machine learning has been recently used in the detection of neuroradiological findings. Hence, it can be used in the detection of intracranial hematomas and furtherly initiate a management cascade of patient transfer, diagnostics, admission, and emergency intervention. We aim, here, to develop a diagnostic tool based on artificial intelligence to detect hematomas instantaneously, and automatically start a cascade of actions that support the management protocol depending on the early diagnosis. Materials and Methods: A plot was designed as a staged model: The first stage of initiating and training the machine with the provisional evaluation of its accuracy and the second stage of supervised use in a tertiary care hospital and a third stage of its generalization in primary and secondary care hospitals. Two datasets were used: CQ500, a public dataset, and our dataset collected retrospectively from our tertiary hospital. Results: A mean dice score of 0.83 was achieved on the validation set of CQ500. Moreover, the detection of intracranial hemorrhage was successful in 94% of cases for the CQ500 test set and 93% for our local institute cases. Poor detection was present in only 6-7% of the total test set. Moderate false-positive results were encountered in 18% and major false positives reached 5% for the total test set. Conclusion: The proposed approach for the early detection of acute intracranial hematomas provides a reliable outset for generating an automatically initiated management cascade in high-flow hospitals.

Implementation and prospective real-time evaluation of a generalized system for in-clinic deployment and validation of machine learning models in radiology.

The medical imaging community has embraced Machine Learning (ML) as evidenced by the rapid increase in the number of ML models being developed, but validating and deploying these models in the clinic remains a challenge. The engineering involved in integrating and assessing the efficacy of ML models within the clinical workflow is complex. This paper presents a general-purpose, end-to-end, clinically integrated ML model deployment and validation system implemented at UCSF. Engineering and usability challenges and results from 3 use cases are presented. A generalized validation system based on free, open-source software (OSS) was implemented, connecting clinical imaging modalities, the Picture Archiving and Communication System (PACS), and an ML inference server. ML pipelines were implemented in NVIDIA's Clara Deploy framework with results and clinician feedback stored in a customized XNAT instance, separate from the clinical record but linked from within PACS. Prospective clinical validation studies of 3 ML models were conducted, with data routed from multiple clinical imaging modalities and PACS. Completed validation studies provided expert clinical feedback on model performance and usability, plus system reliability and performance metrics. Clinical validation of ML models entails assessing model performance, impact on clinical infrastructure, robustness, and usability. Study results must be easily accessible to participating clinicians but remain outside the clinical record. Building a system that generalizes and scales across multiple ML models takes the concerted effort of software engineers, clinicians, data scientists, and system administrators, and benefits from the use of modular OSS. The present work provides a template for institutions looking to translate and clinically validate ML models in the clinic, together with required resources and expected challenges.

Rapid artificial intelligence solutions in a pandemic-The COVID-19-20 Lung CT Lesion Segmentation Challenge.

Artificial intelligence (AI) methods for the automatic detection and quantification of COVID-19 lesions in chest computed tomography (CT) might play an important role in the monitoring and management of the disease. We organized an international challenge and competition for the development and comparison of AI algorithms for this task, which we supported with public data and state-of-the-art benchmark methods. Board Certified Radiologists annotated 295 public images from two sources (A and B) for algorithms training (n=199, source A), validation (n=50, source A) and testing (n=23, source A; n=23, source B). There were 1,096 registered teams of which 225 and 98 completed the validation and testing phases, respectively. The challenge showed that AI models could be rapidly designed by diverse teams with the potential to measure disease or facilitate timely and patient-specific interventions. This paper provides an overview and the major outcomes of the COVID-19 Lung CT Lesion Segmentation Challenge - 2020.

Integrating the OHIF Viewer into XNAT: Achievements, Challenges and Prospects for Quantitative Imaging Studies.

Purpose: XNAT is an informatics software platform to support imaging research, particularly in the context of large, multicentre studies of the type that are essential to validate quantitative imaging biomarkers. XNAT provides import, archiving, processing and secure distribution facilities for image and related study data. Until recently, however, modern data visualisation and annotation tools were lacking on the XNAT platform. We describe the background to, and implementation of, an integration of the Open Health Imaging Foundation (OHIF) Viewer into the XNAT environment. We explain the challenges overcome and discuss future prospects for quantitative imaging studies. Materials and methods: The OHIF Viewer adopts an approach based on the DICOM web protocol. To allow operation in an XNAT environment, a data-routing methodology was developed to overcome the mismatch between the DICOM and XNAT information models and a custom viewer panel created to allow navigation within the viewer between different XNAT projects, subjects and imaging sessions. Modifications to the development environment were made to allow developers to test new code more easily against a live XNAT instance. Major new developments focused on the creation and storage of regions-of-interest (ROIs) and included: ROI creation and editing tools for both contour- and mask-based regions; a "smart CT" paintbrush tool; the integration of NVIDIA's Artificial Intelligence Assisted Annotation (AIAA); the ability to view surface meshes, fractional segmentation maps and image overlays; and a rapid image reader tool aimed at radiologists. We have incorporated the OHIF microscopy extension and, in parallel, introduced support for microscopy session types within XNAT for the first time. Results: Integration of the OHIF Viewer within XNAT has been highly successful and numerous additional and enhanced tools have been created in a programme started in 2017 that is still ongoing. The software has been downloaded more than 3700 times during the course of the development work reported here, demonstrating the impact of the work. Conclusions: The OHIF open-source, zero-footprint web viewer has been incorporated into the XNAT platform and is now used at many institutions worldwide. Further innovations are envisaged in the near future.

Federated learning for predicting clinical outcomes in patients with COVID-19.

Federated learning (FL) is a method used for training artificial intelligence models with data from multiple sources while maintaining data anonymity, thus removing many barriers to data sharing. Here we used data from 20 institutes across the globe to train a FL model, called EXAM (electronic medical record (EMR) chest X-ray AI model), that predicts the future oxygen requirements of symptomatic patients with COVID-19 using inputs of vital signs, laboratory data and chest X-rays. EXAM achieved an average area under the curve (AUC) >0.92 for predicting outcomes at 24 and 72 h from the time of initial presentation to the emergency room, and it provided 16% improvement in average AUC measured across all participating sites and an average increase in generalizability of 38% when compared with models trained at a single site using that site's data. For prediction of mechanical ventilation treatment or death at 24 h at the largest independent test site, EXAM achieved a sensitivity of 0.950 and specificity of 0.882. In this study, FL facilitated rapid data science collaboration without data exchange and generated a model that generalized across heterogeneous, unharmonized datasets for prediction of clinical outcomes in patients with COVID-19, setting the stage for the broader use of FL in healthcare.

Rapid Artificial Intelligence Solutions in a Pandemic - The COVID-19-20 Lung CT Lesion Segmentation Challenge.

Artificial intelligence (AI) methods for the automatic detection and quantification of COVID-19 lesions in chest computed tomography (CT) might play an important role in the monitoring and management of the disease. We organized an international challenge and competition for the development and comparison of AI algorithms for this task, which we supported with public data and state-of-the-art benchmark methods. Board Certified Radiologists annotated 295 public images from two sources (A and B) for algorithms training (n=199, source A), validation (n=50, source A) and testing (n=23, source A; n=23, source B). There were 1,096 registered teams of which 225 and 98 completed the validation and testing phases, respectively. The challenge showed that AI models could be rapidly designed by diverse teams with the potential to measure disease or facilitate timely and patient-specific interventions. This paper provides an overview and the major outcomes of the COVID-19 Lung CT Lesion Segmentation Challenge - 2020.

Federated Learning used for predicting outcomes in SARS-COV-2 patients.

'Federated Learning' (FL) is a method to train Artificial Intelligence (AI) models with data from multiple sources while maintaining anonymity of the data thus removing many barriers to data sharing. During the SARS-COV-2 pandemic, 20 institutes collaborated on a healthcare FL study to predict future oxygen requirements of infected patients using inputs of vital signs, laboratory data, and chest x-rays, constituting the "EXAM" (EMR CXR AI Model) model. EXAM achieved an average Area Under the Curve (AUC) of over 0.92, an average improvement of 16%, and a 38% increase in generalisability over local models. The FL paradigm was successfully applied to facilitate a rapid data science collaboration without data exchange, resulting in a model that generalised across heterogeneous, unharmonized datasets. This provided the broader healthcare community with a validated model to respond to COVID-19 challenges, as well as set the stage for broader use of FL in healthcare.

Gaussian Process Regressions for Inverse Problems and Parameter Searches in Models of Ventricular Mechanics.

Patient specific models of ventricular mechanics require the optimization of their many parameters under the uncertainties associated with imaging of cardiac function. We present a strategy to reduce the complexity of parametric searches for 3-D FE models of left ventricular contraction. The study employs automatic image segmentation and analysis of an image database to gain geometric features for several classes of patients. Statistical distributions of geometric parameters are then used to design parametric studies investigating the effects of: (1) passive material properties during ventricular filling, and (2) infarct geometry on ventricular contraction in patients after a heart attack. Gaussian Process regression is used in both cases to build statistical models trained on the results of biophysical FEM simulations. The first statistical model estimates unloaded configurations based on either the intraventricular pressure or the end-diastolic fiber strain. The technique provides an alternative to the standard fixed-point iteration algorithm, which is more computationally expensive when used to unload more than 10 ventricles. The second statistical model captures the effects of varying infarct geometries on cardiac output. For training, we designed high resolution models of non-transmural infarcts including refinements of the border zone around the lesion. This study is a first effort in developing a platform combining HPC models and machine learning to investigate cardiac function in heart failure patients with the goal of assisting clinical diagnostics.

Assessment of liver fibrosis using fast strain-encoded MRI driven by inherent cardiac motion.

PURPOSE: An external driver-free MRI method for assessment of liver fibrosis offers a promising noninvasive tool for diagnosis and monitoring of liver disease. Lately, the heart's intrinsic motion and MR tagging have been utilized for the quantification of liver strain. However, MR tagging requires multiple breath-hold acquisitions and substantial postprocessing. In this study, we propose the use of a fast strain-encoded (FSENC) MRI method to measure the peak strain (Sp ) in the liver's left lobe, which is in close proximity and caudal to the heart. Additionally, we introduce a new method of measuring heart-induced shear wave velocity (SWV) inside the liver. METHODS: Phantom and in vivo experiments (11 healthy subjects and 11 patients with liver fibrosis) were conducted. Reproducibility experiments were performed in seven healthy subjects. RESULTS: Peak liver strain, Sp , decreased significantly in fibrotic liver compared with healthy liver (6.46% ± 2.27% vs 12.49% ± 1.76%; P < 0.05). Heart-induced SWV increased significantly in patients compared with healthy subjects (0.15 ± 0.04 m/s vs 0.63 ± 0.32 m/s; P < 0.05). Reproducibility analysis yielded no significant difference in Sp (P = 0.47) or SWV (P = 0.56). CONCLUSION: Accelerated external driver-free noninvasive assessment of left liver lobe strain and SWV is feasible using strain-encoded MRI. The two measures significantly separate healthy subjects from patients with fibrotic liver. Magn Reson Med 74:106-114, 2015. © 2014 Wiley Periodicals, Inc.

Inter-study reproducibility of cardiovascular magnetic resonance tagging.

BACKGROUND: The aim of this study is to determine the test-retest reliability of the measurement of regional myocardial function by cardiovascular magnetic resonance (CMR) tagging using spatial modulation of magnetization. METHODS: Twenty-five participants underwent CMR tagging twice over 12 ± 7 days. To assess the role of slice orientation on strain measurement, two healthy volunteers had a first exam, followed by image acquisition repeated with slices rotated ±15 degrees out of true short axis, followed by a second exam in the true short axis plane. To assess the role of slice location, two healthy volunteers had whole heart tagging. The harmonic phase (HARP) method was used to analyze the tagged images. Peak midwall circumferential strain (Ecc), radial strain (Err), Lambda 1, Lambda 2, and Angle α were determined in basal, mid and apical slices. LV torsion, systolic and early diastolic circumferential strain and torsion rates were also determined. RESULTS: LV Ecc and torsion had excellent intra-, interobserver, and inter-study intra-class correlation coefficients (ICC range, 0.7 to 0.9). Err, Lambda 1, Lambda 2 and angle had excellent intra- and interobserver ICC than inter-study ICC. Angle had least inter-study reproducibility. Torsion rates had superior intra-, interobserver, and inter-study reproducibility to strain rates. The measurements of LV Ecc were comparable in all three slices with different short axis orientations (standard deviation of mean Ecc was 0.09, 0.18 and 0.16 at basal, mid and apical slices, respectively). The mean difference in LV Ecc between slices was more pronounced in most of the basal slices compared to the rest of the heart. CONCLUSIONS: Intraobserver and interobserver reproducibility of all strain and torsion parameters was excellent. Inter-study reproducibility of CMR tagging by SPAMM varied between different parameters as described in the results above and was superior for Ecc and LV torsion. The variation in LV Ecc measurement due to altered slice orientation is negligible compared to the variation due to slice location. TRIAL REGISTRATION: This trial is registered as NCT00005487 at National Heart, Lung and Blood institute.

Regional and global biventricular function in pulmonary arterial hypertension: a cardiac MR imaging study.

PURPOSE: To determine whether chronic pulmonary arterial pressure (PAP) elevation affects regional biventricular function and whether regional myocardial function may be reduced in pulmonary arterial hypertension (PAH) patients with preserved global right ventricular (RV) function. MATERIALS AND METHODS: After informed consent, 35 PAH patients were evaluated with right heart catheterization and cardiac magnetic resonance (MR) imaging and compared with 13 healthy control subjects. Biventricular segmental, section, and mean ventricular peak systolic longitudinal strain (E(LL)), as well as left ventricular (LV) circumferential and RV tangential strains were compared between PAH patients and control subjects and correlated with global function and catheterization of the right heart indexes. Spearman ρ correlation with Bonferroni correction was used. Multiple linear regression analysis was performed to determine predictors for regional myocardial function. RESULTS: In the RV of PAH patients, longitudinal contractility was reduced at the basal, mid, and apical levels, and tangential contractility was reduced at the midventricular level. Mean RV E(LL) positively correlated with mean PAP (r = 0.62, P < .0014) and pulmonary vascular resistance index (PVRI) (r = 0.77, P < .0014). Mean PAP was a predictor of mean RV E(LL) (ß = .19, P = .005) in a multiple linear regression analysis. In the LV, reduced LV longitudinal and circumferential contractility were noted at the base. LV anteroseptal E(LL) positively correlated with increased mean PAP (r = 0.5, P = .03) and septal eccentricity index (r = 0.5, P = .01). In a subgroup of PAH patients with normal global RV function, significantly reduced RV longitudinal contractility was noted at basal and mid anterior septal insertions, as well as the mid anterior RV wall (P < .05 for all). CONCLUSION: In PAH patients, reduced biventricular regional function is associated with increased RV afterload (mean PAP and PVRI). Cardiac MR imaging helps identify regional RV dysfunction in PAH patients with normal global RV function. SUPPLEMENTAL MATERIAL: http://radiology.rsna.org/lookup/suppl/doi:10.1148/radiol.12111599/-/DC1.

Finding the optimal compression level for strain-encoded (SENC) breast MRI; simulations and phantom experiments.

Breast cancer is the most common cancer among women and the second highest cause of cancer-related death. Diagnostic magnetic resonance imaging (MRI) is recommended to screen high-risk patients. Strain-Encoded (SENC) can improve MRI's specificity by detecting and differentiating masses according to their stiffness. Previous phantom and ex-vivo studies have utilized SENC to detect cancerous masses. However, SENC required a 30% compression of the tissue, which may not be feasible for in-vivo imaging. In this work, we use finite element method simulations and phantom experiments to determine the minimum compression required to detect and classify masses. Results show that SENC is capable of detecting stiff masses at compression level of 7%, though higher compression is needed in order to differentiate between normal tissue and benign or malignant masses. With on-line SENC calculations implemented on the scanner console, we propose to start with small compressions for maximum patient comfort, then progress to larger compressions if any masses are detected.

Age-related changes in aortic arch geometry: relationship with proximal aortic function and left ventricular mass and remodeling.

OBJECTIVES: We sought to define age-related geometric changes of the aortic arch and determine their relationship to central aortic stiffness and left ventricular (LV) remodeling. BACKGROUND: The proximal aorta has been shown to thicken, enlarge in diameter, and lengthen with aging in humans. However, no systematic study has described age-related longitudinal and transversal remodeling of the aortic arch and their relationship with LV mass and remodeling. METHODS: We studied 100 subjects (55 women, 45 men, average age 46 ± 16 years) free of overt cardiovascular disease using magnetic resonance imaging to determine aortic arch geometry (length, diameters, height, width, and curvature), aortic arch function (local aortic distensibility and arch pulse wave velocity [PWV]), and LV volumes and mass. Radial tonometry was used to calculate central blood pressure. RESULTS: Aortic diameters and arch length increased significantly with age. The ascending aorta length increased most, with age leading to aortic arch widening and decreased curvature. These geometric changes of the aortic arch were significantly related to decreased ascending aortic distensibility, increased aortic arch PWV (p < 0.001), and increased central blood pressures (p < 0.001). Increased ascending aortic diameter, lengthening, and decreased curvature of the aortic arch (unfolding) were all significantly associated with increased LV mass and concentric remodeling independently of age, sex, body size, and central blood pressure (p < 0.01). CONCLUSIONS: Age-related unfolding of the aortic arch is related to increased proximal aortic stiffness in individuals without cardiovascular disease and associated with increased LV mass and mass-to-volume ratio independent of age, body size, central pressure, and cardiovascular risk factors.

Improved hardware for higher spatial resolution strain-encoded (SENC) breast MRI for strain measurements.

RATIONALE AND OBJECTIVES: Early detection of breast lesions using mammography has resulted in lower mortality rates. However, some breast lesions are mammography occult, and magnetic resonance imaging (MRI) is recommended, but it has lower specificity. It is possible to achieve higher specificity by using strain-encoded (SENC) MRI and/or magnetic resonance elastography. SENC breast MRI can measure the strain properties of breast tissue. Similarly, magnetic resonance elastography is used to measure the elasticity (ie, shear stiffness) of different tissue compositions interrogating the tissue mechanical properties. Reports have shown that malignant tumors are three to 13 times stiffer than normal tissue and benign tumors. MATERIALS AND METHODS: The investigators have developed a SENC breast hardware device capable of periodically compressing the breast, thus allowing for longer scanning time and measuring the strain characteristics of breast tissue. This hardware enables the use of SENC MRI with high spatial resolution (1 × 1 × 5 mm(3)) instead of fast SENC imaging. Simple controls and multiple safety measures were added to ensure accurate, repeatable, and safe in vivo experiments. RESULTS: Phantom experiments showed that SENC breast MRI has higher signal-to-noise ratio and contrast-to-noise ratio than fast SENC imaging under different scanning resolutions. Finally, the SENC breast device reproducibility measurements resulted in a difference of <1 mm with a 1% strain difference. CONCLUSIONS: SENC breast magnetic resonance images have higher signal-to-noise ratio and contrast-to-noise ratios than fast SENC images. Thus, combining SENC breast strain measurements with diagnostic breast MRI to differentiate benign from malignant lesions could potentially increase the specificity of diagnosis in the clinical setting.