Identifying acute ischemic stroke patients within the thrombolytic treatment window using deep learning.

BACKGROUND AND PURPOSE: Treatment of acute ischemic stroke is heavily contingent upon time, as there is a strong relationship between time clock and tissue progression. Work has established imaging biomarker assessments as surrogates for time since stroke (TSS), namely, by comparing signal mismatch between diffusion-weighted imaging (DWI) and fluid-attenuated inversion recovery (FLAIR) imaging. Our goal was to develop an automatic technique for determining TSS from imaging that does not require subspecialist radiology expertise. METHODS: Using 772 patients (66 ± 9 years, 319 women), we developed and externally evaluated a deep learning network for classifying TSS from MR images and compared algorithm predictions to neuroradiologist assessments of DWI-FLAIR mismatch. Models were trained to classify TSS within 4.5 hours and performance metrics with confidence intervals were reported on both internal and external evaluation sets. RESULTS: Three board-certified neuroradiologists' DWI-FLAIR mismatch assessments, based on majority vote, yielded a sensitivity of .62, a specificity of .86, and a Fleiss' kappa of .46 when used to classify TSS. The deep learning method performed similarly to radiologists and outperformed previously reported methods, with the best model achieving an average evaluation accuracy, sensitivity, and specificity of .726, .712, and .741, respectively, on an internal cohort and .724, .757, and .679, respectively, on an external cohort. CONCLUSION: Our model achieved higher generalization performance on external evaluation datasets than the current state-of-the-art for TSS classification. These results demonstrate the potential of automatic assessment of onset time from imaging without the need for expertly trained radiologists.

Lamotrigine for Treating Anger in Veterans With Posttraumatic Stress Disorder.

ABSTRACT: Lamotrigine is an antiepileptic drug that was Food and Drug Administration approved in 2003 for use in the maintenance treatment of bipolar I disorder to delay the time to recurrence of new mood episodes. The mechanism by which lamotrigine achieves its therapeutic effect in the treatment of bipolar disorder is unknown. Here, we report on 2 Veterans with combat-related posttraumatic stress disorder (PTSD) endorsing significant anger, aggression, and agitation, who were treated with selective serotonin reuptake inhibitors, but whose residual symptoms of anger and aggression were ultimately successfully managed with lamotrigine augmentation. The authors would like to make mental health providers aware of the impact that lamotrigine may have on PTSD symptomology, especially when used to treat anger and aggression in patients with PTSD.

Intra-domain task-adaptive transfer learning to determine acute ischemic stroke onset time.

Treatment of acute ischemic strokes (AIS) is largely contingent upon the time since stroke onset (TSS). However, TSS may not be readily available in up to 25% of patients with unwitnessed AIS. Current clinical guidelines for patients with unknown TSS recommend the use of MRI to determine eligibility for thrombolysis, but radiology assessments have high inter-reader variability. In this work, we present deep learning models that leverage MRI diffusion series to classify TSS based on clinically validated thresholds. We propose an intra-domain task-adaptive transfer learning method, which involves training a model on an easier clinical task (stroke detection) and then refining the model with different binary thresholds of TSS. We apply this approach to both 2D and 3D CNN architectures with our top model achieving an ROC-AUC value of 0.74, with a sensitivity of 0.70 and a specificity of 0.81 for classifying TSS < 4.5 h. Our pretrained models achieve better classification metrics than the models trained from scratch, and these metrics exceed those of previously published models applied to our dataset. Furthermore, our pipeline accommodates a more inclusive patient cohort than previous work, as we did not exclude imaging studies based on clinical, demographic, or image processing criteria. When applied to this broad spectrum of patients, our deep learning model achieves an overall accuracy of 75.78% when classifying TSS < 4.5 h, carrying potential therapeutic implications for patients with unknown TSS.

Discovering and interpreting transcriptomic drivers of imaging traits using neural networks.

MOTIVATION: Cancer heterogeneity is observed at multiple biological levels. To improve our understanding of these differences and their relevance in medicine, approaches to link organ- and tissue-level information from diagnostic images and cellular-level information from genomics are needed. However, these 'radiogenomic' studies often use linear or shallow models, depend on feature selection, or consider one gene at a time to map images to genes. Moreover, no study has systematically attempted to understand the molecular basis of imaging traits based on the interpretation of what the neural network has learned. These studies are thus limited in their ability to understand the transcriptomic drivers of imaging traits, which could provide additional context for determining clinical outcomes. RESULTS: We present a neural network-based approach that takes high-dimensional gene expression data as input and performs non-linear mapping to an imaging trait. To interpret the models, we propose gene masking and gene saliency to extract learned relationships from radiogenomic neural networks. In glioblastoma patients, our models outperformed comparable classifiers (>0.10 AUC) and our interpretation methods were validated using a similar model to identify known relationships between genes and molecular subtypes. We found that tumor imaging traits had specific transcription patterns, e.g. edema and genes related to cellular invasion, and 10 radiogenomic traits were significantly predictive of survival. We demonstrate that neural networks can model transcriptomic heterogeneity to reflect differences in imaging and can be used to derive radiogenomic traits with clinical value. AVAILABILITY AND IMPLEMENTATION: https://github.com/novasmedley/deepRadiogenomics. CONTACT: whsu@mednet.ucla.edu. SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

Predicting ischemic stroke tissue fate using a deep convolutional neural network on source magnetic resonance perfusion images.

Predicting infarct volume from magnetic resonance perfusion-weighted imaging can provide helpful information to clinicians in deciding how aggressively to treat acute stroke patients. Models have been developed to predict tissue fate, yet these models are mostly built using hand-crafted features (e.g., time-to-maximum) derived from perfusion images, which are sensitive to deconvolution methods. We demonstrate the application of deep convolution neural networks (CNNs) on predicting final stroke infarct volume using only the source perfusion images. We propose a deep CNN architecture that improves feature learning and achieves an area under the curve of 0.871 ± 0.024 , outperforming existing tissue fate models. We further validate the proposed deep CNN with existing 2-D and 3-D deep CNNs for images/video classification, showing the importance of the proposed architecture. Our work leverages deep learning techniques in stroke tissue outcome prediction, advancing magnetic resonance imaging perfusion analysis one step closer to an operational decision support tool for stroke treatment guidance.

A Machine Learning Approach for Classifying Ischemic Stroke Onset Time From Imaging.

Current clinical practice relies on clinical history to determine the time since stroke (TSS) onset. Imaging-based determination of acute stroke onset time could provide critical information to clinicians in deciding stroke treatment options, such as thrombolysis. The patients with unknown or unwitnessed TSS are usually excluded from thrombolysis, even if their symptoms began within the therapeutic window. In this paper, we demonstrate a machine learning approach for TSS classification using routinely acquired imaging sequences. We develop imaging features from the magnetic resonance (MR) images and train machine learning models to classify the TSS. We also propose a deep-learning model to extract hidden representations for the MR perfusion-weighted images and demonstrate classification improvement by incorporating these additional deep features. The cross-validation results show that our best classifier achieved an area under the curve of 0.765, with a sensitivity of 0.788 and a negative predictive value of 0.609, outperforming existing methods. We show that the features generated by our deep-learning algorithm correlate with the MR imaging features, and validate the robustness of the model on imaging parameter variations (e.g., year of imaging). This paper advances magnetic resonance imaging analysis one-step-closer to an operational decision support tool for stroke treatment guidance.

Classifying Acute Ischemic Stroke Onset Time using Deep Imaging Features.

Models have been developed to predict stroke outcomes (e.g., mortality) in attempt to provide better guidance for stroke treatment. However, there is little work in developing classification models for the problem of unknown time-since-stroke (TSS), which determines a patient's treatment eligibility based on a clinical defined cutoff time point (i.e., <4.5hrs). In this paper, we construct and compare machine learning methods to classify TSS<4.5hrs using magnetic resonance (MR) imaging features. We also propose a deep learning model to extract hidden representations from the MR perfusion-weighted images and demonstrate classification improvement by incorporating these additional imaging features. Finally, we discuss a strategy to visualize the learned features from the proposed deep learning model. The cross-validation results show that our best classifier achieved an area under the curve of 0.68, which improves significantly over current clinical methods (0.58), demonstrating the potential benefit of using advanced machine learning methods in TSS classification.

Medical Imaging Informatics.

Imaging is one of the most important sources of clinically observable evidence that provides broad coverage, can provide insight on low-level scale properties, is noninvasive, has few side effects, and can be performed frequently. Thus, imaging data provides a viable observable that can facilitate the instantiation of a theoretical understanding of a disease for a particular patient context by connecting imaging findings to other biologic parameters in the model (e.g., genetic, molecular, symptoms, and patient survival). These connections can help inform their possible states and/or provide further coherent evidence. The field of radiomics is particularly dedicated to this task and seeks to extract quantifiable measures wherever possible. Example properties of investigation include genotype characterization, histopathology parameters, metabolite concentrations, vascular proliferation, necrosis, cellularity, and oxygenation. Important issues within the field include: signal calibration, spatial calibration, preprocessing methods (e.g., noise suppression, motion correction, and field bias correction), segmentation of target anatomic/pathologic entities, extraction of computed features, and inferencing methods connecting imaging features to biological states.

Automatic Classification of Ultrasound Screening Examinations of the Abdominal Aorta.

Our work facilitates the identification of veterans who may be at risk for abdominal aortic aneurysms (AAA) based on the 2007 mandate to screen all veteran patients that meet the screening criteria. The main research objective is to automatically index three clinical conditions: pertinent negative AAA, pertinent positive AAA, and visually unacceptable image exams. We developed and evaluated a ConText-based algorithm with the GATE (General Architecture for Text Engineering) development system to automatically classify 1402 ultrasound radiology reports for AAA screening. Using the results from JAPE (Java Annotation Pattern Engine) transducer rules, we developed a feature vector to classify the radiology reports with a decision table classifier. We found that ConText performed optimally on precision and recall for pertinent negative (0.99 (0.98-0.99), 0.99 (0.99-1.00)) and pertinent positive AAA detection (0.98 (0.95-1.00), 0.97 (0.92-1.00)), and respectably for determination of non-diagnostic image studies (0.85 (0.77-0.91), 0.96 (0.91-0.99)). In addition, our algorithm can determine the AAA size measurements for further characterization of abnormality. We developed and evaluated a regular expression based algorithm using GATE for determining the three contextual conditions: pertinent negative, pertinent positive, and non-diagnostic from radiology reports obtained for evaluating the presence or absence of abdominal aortic aneurysm. ConText performed very well at identifying the contextual features. Our study also discovered contextual trigger terms to detect sub-standard ultrasound image quality. Limitations of performance included unknown dictionary terms, complex sentences, and vague findings that were difficult to classify and properly code.

Assessing Variability in Brain Tumor Segmentation to Improve Volumetric Accuracy and Characterization of Change.

Brain tumor analysis is moving towards volumetric assessment of magnetic resonance imaging (MRI), providing a more precise description of disease progression to better inform clinical decision-making and treatment planning. While a multitude of segmentation approaches exist, inherent variability in the results of these algorithms may incorrectly indicate changes in tumor volume. In this work, we present a systematic approach to characterize variability in tumor boundaries that utilizes equivalence tests as a means to determine whether a tumor volume has significantly changed over time. To demonstrate these concepts, 32 MRI studies from 8 patients were segmented using four different approaches (statistical classifier, region-based, edge-based, knowledge-based) to generate different regions of interest representing tumor extent. We showed that across all studies, the average Dice coefficient for the superset of the different methods was 0.754 (95% confidence interval 0.701-0.808) when compared to a reference standard. We illustrate how variability obtained by different segmentations can be used to identify significant changes in tumor volume between sequential time points. Our study demonstrates that variability is an inherent part of interpreting tumor segmentation results and should be considered as part of the interpretation process.

White matter changes in chronic alcoholic liver disease: Hypothesized association and putative biochemical mechanisms.

Advanced liver disease has long been associated with cerebral abnormalities. These abnormalities, termed acquired hepatocerebral degeneration, are typically visualized as T1 weighted hyperintensity on MRI in the deep gray matter of the basal ganglia. Recent reports, however, have demonstrated that a subset of patients with chronic alcoholic liver disease may also develop white matter abnormalities. Thus far, the morphology of these changes is not well characterized. Previous studies have described these changes as patchy, sporadic white matter abnormalities but have not posited localization of these changes to any particular white matter tracts. This paper hypothesizes that the white matter findings associated with advanced alcoholic liver disease localize to the corticocerebellar tracts. As an initial investigation of this hypothesis, 78 patients with a diagnosis of liver cirrhosis and an MRI showing clearly abnormal T1 weighted hyperintensity in the bilateral globus pallidus, characteristic of chronic liver disease, were examined for white matter signal abnormalities in the corticocerebellar tracts using FLAIR and T2 weighted images. The corticocerebellar tracts were subdivided into two regions: periventricular white matter (consisting of the sum of the centrum-semiovale and corona radiata), and lower white matter (consisting of the corona radiata, internal capsules, middle cerebral peduncles, middle cerebellar peduncles and cerebellum). As compared to matched controls, significantly greater signal abnormalities in both the periventricular white matter and lower white matter regions of the corticocerebellar tracts were observed in patients with known liver cirrhosis and abnormal T1 W hyperintensity in the globi pallidi. This difference was most pronounced in the lower white matter region of the corticocerebellar tract, with statistical significance of p<0.0005. Furthermore, the pathophysiologic mechanism underlying these changes remains unknown. This paper hypothesizes that the etiology of white matter changes associated with advanced liver disease may resemble that of white matter findings in subacute combined degeneration secondary to vitamin B12 deficiency. Specifically, significant evidence suggests that dysfunctional methionine metabolism as well as dysregulated cytokine production secondary to B12 deficiency play a major role in the development of subacute combined degeneration. Similar dysfunction of methionine metabolism and cytokine regulation is seen in alcoholic liver disease and is hypothesized in this paper to, at least in part, lead to white matter findings associated with alcoholic liver disease.

Association of calcified carotid atheromas visualized on panoramic images and aortic arch calcifications seen on chest radiographs of postmenopausal women.

BACKGROUND: Occult atherosclerotic disease is the leading cause of death among older women. The authors hypothesized that women with calcified carotid artery plaque (CCAP) visualized on panoramic images were more likely to have aortic arch calcifications (AAC) that were visible on chest radiographs (CRs), a risk indicator of experiencing cardiovascular events, than would matched cohorts who did not have atheromas. METHODS: The authors obtained the CRs of 36 female veterans (≥ 50 years) who had CCAP and atherogenically risk-matched them to those of 36 women without CCAP. A radiologist evaluated the CRs for AAC. Other study variables included age, ethnicity, body mass index and presence or absence of hypertension, diabetes and dyslipidemia. The authors computed descriptive and bivariate statistics. RESULTS: Women 60 years or older who had evidence of CCAP on their panoramic radiographs were significantly (P = .022; 95 percent confidence interval, 1.298-26.223) more likely to have evidence of AAC on their CRs than were similarly aged women who did not have evidence of CCAP. This association was not evident in women younger than 60 years. Among women who were both younger and older than 60 years, there was no evident association between the presence of CCAP and the severity (on a four point scale [0-3]) of AAC calcification. CONCLUSION: Prevalence of carotid plaque on panoramic images of women 60 years or older is significantly associated with presence of aortic arch calcifications on CRs. PRACTICAL IMPLICATIONS: Panoramic images of women 60 years or older must be evaluated for CCAP, given their association with AAC. Patients with atheromas should be referred to their physicians for further evaluation given the systemic implications.

Predicting discharge mortality after acute ischemic stroke using balanced data.

Several models have been developed to predict stroke outcomes (e.g., stroke mortality, patient dependence, etc.) in recent decades. However, there is little discussion regarding the problem of between-class imbalance in stroke datasets, which leads to prediction bias and decreased performance. In this paper, we demonstrate the use of the Synthetic Minority Over-sampling Technique to overcome such problems. We also compare state of the art machine learning methods and construct a six-variable support vector machine (SVM) model to predict stroke mortality at discharge. Finally, we discuss how the identification of a reduced feature set allowed us to identify additional cases in our research database for validation testing. Our classifier achieved a c-statistic of 0.865 on the cross-validated dataset, demonstrating good classification performance using a reduced set of variables.

Unifying acute stroke treatment guidelines for a Bayesian belief network.

With the large number of clinical practice guidelines available, there is an increasing need for a comprehensive unified model for acute ischemic stroke treatment to assist in clinical decision making. We present a unified treatment model derived through review of existing clinical practice guidelines, meta-analyses, and clinical trials. Using logic from the treatment model, a Bayesian belief network was defined and fitted to data from our institution's observational quality improvement database for acute stroke patients. The resulting network validates known relationships between variables, treatment decisions and outcomes, and enables the exploration of new correlative relationships not defined in current guidelines.

Imaging-based observational databases for clinical problem solving: the role of informatics.

Imaging has become a prevalent tool in the diagnosis and treatment of many diseases, providing a unique in vivo, multi-scale view of anatomic and physiologic processes. With the increased use of imaging and its progressive technical advances, the role of imaging informatics is now evolving--from one of managing images, to one of integrating the full scope of clinical information needed to contextualize and link observations across phenotypic and genotypic scales. Several challenges exist for imaging informatics, including the need for methods to transform clinical imaging studies and associated data into structured information that can be organized and analyzed. We examine some of these challenges in establishing imaging-based observational databases that can support the creation of comprehensive disease models. The development of these databases and ensuing models can aid in medical decision making and knowledge discovery and ultimately, transform the use of imaging to support individually-tailored patient care.

Imaging informatics for consumer health: towards a radiology patient portal.

OBJECTIVE: With the increased routine use of advanced imaging in clinical diagnosis and treatment, it has become imperative to provide patients with a means to view and understand their imaging studies. We illustrate the feasibility of a patient portal that automatically structures and integrates radiology reports with corresponding imaging studies according to several information orientations tailored for the layperson. METHODS: The imaging patient portal is composed of an image processing module for the creation of a timeline that illustrates the progression of disease, a natural language processing module to extract salient concepts from radiology reports (73% accuracy, F1 score of 0.67), and an interactive user interface navigable by an imaging findings list. The portal was developed as a Java-based web application and is demonstrated for patients with brain cancer. RESULTS AND DISCUSSION: The system was exhibited at an international radiology conference to solicit feedback from a diverse group of healthcare professionals. There was wide support for educating patients about their imaging studies, and an appreciation for the informatics tools used to simplify images and reports for consumer interpretation. Primary concerns included the possibility of patients misunderstanding their results, as well as worries regarding accidental improper disclosure of medical information. CONCLUSIONS: Radiologic imaging composes a significant amount of the evidence used to make diagnostic and treatment decisions, yet there are few tools for explaining this information to patients. The proposed radiology patient portal provides a framework for organizing radiologic results into several information orientations to support patient education.

Complex imaging features of accidental cerebral intraventricular gadolinium administration.

Gadolinium diethylenetriamine pentaacetic acid (Gd-DTPA) is a contrast agent commonly used for enhancing MRI. In this paper, the authors report on 2 cases of postoperative inadvertent administration of Gd-DTPA directly into a ventriculostomy tubing side port that was mistaken for intravenous tubing. Both cases demonstrated a low signal on MRI throughout the ventricular system and dependent portions of the subarachnoid spaces, which was originally believed to be CSF with areas of T1 shortening in the nondependent portions of the subarachnoid spaces, and misinterpreted as basal leptomeningeal enhancement and meningitis. The authors propose that the appearance of profound T1 hypointensity within the ventricles and diffuse susceptibility artifact along the ependyma is pathognomonic of intraventricular Gd-DTPA and should be recognized.

Carotid artery stenosis: wide variability in reporting formats--a review of 127 Veterans Affairs medical centers.

PURPOSE: To determine whether radiology reports describe clinically significant carotid arterial stenosis in a consistent format that is actionable by ordering clinicians. MATERIALS AND METHODS: This study was HIPAA compliant. Informed consent was waived. Institutional review board approval was obtained for this retrospective chart review, which included radiology reports of carotid artery imaging for patients hospitalized with ischemic stroke at 127 Veterans Affairs medical centers in 2006-2007. "Clinically significant results" were defined as results with at least 50% stenosis or at least moderate stenosis, excluding complete occlusion. How often clinically significant results were reported as an exact percentage stenosis (such as 60%), range (such as 50%-69%), or category (such as moderate) was determined. Among results reported as a range, how often the range bracketed clinical thresholds of 50% and 70% (typically used to determine appropriateness of carotid arterial revascularization) was determined. RESULTS: Among 2675 patients, there were 6618 carotid imaging results, of which 1015 (15%) were considered clinically significant. Among 695 clinically significant results at ultrasonography (US), 348 (50%) were described as a range, and another 314 (45%) were reported as an exact percentage stenosis. Among the 348 clinically significant US results reported as a range, 259 (74%) bracketed the thresholds of 50% or 70%. For magnetic resonance angiographic results, 48% (106 of 221) qualitatively described clinically significant results as a category, 38% (84 of 221) as an exact percentage stenosis, and 14% (31 of 221) as a range. CONCLUSION: In this national health care system, the manner in which clinically significant carotid arterial stenosis was reported varied widely.

Context-based electronic health record: toward patient specific healthcare.

Due to the increasingly data-intensive clinical environment, physicians now have unprecedented access to detailed clinical information from a multitude of sources. However, applying this information to guide medical decisions for a specific patient case remains challenging. One issue is related to presenting information to the practitioner: displaying a large (irrelevant) amount of information often leads to information overload. Next-generation interfaces for the electronic health record (EHR) should not only make patient data easily searchable and accessible, but also synthesize fragments of evidence documented in the entire record to understand the etiology of a disease and its clinical manifestation in individual patients. In this paper, we describe our efforts toward creating a context-based EHR, which employs biomedical ontologies and (graphical) disease models as sources of domain knowledge to identify relevant parts of the record to display. We hypothesize that knowledge (e.g., variables, relationships) from these sources can be used to standardize, annotate, and contextualize information from the patient record, improving access to relevant parts of the record and informing medical decision making. To achieve this goal, we describe a framework that aggregates and extracts findings and attributes from free-text clinical reports, maps findings to concepts in available knowledge sources, and generates a tailored presentation of the record based on the information needs of the user. We have implemented this framework in a system called Adaptive EHR, demonstrating its capabilities to present and synthesize information from neurooncology patients. This paper highlights the challenges and potential applications of leveraging disease models to improve the access, integration, and interpretation of clinical patient data.

Basal ganglia structures differentially contribute to verbal fluency: evidence from Human Immunodeficiency Virus (HIV)-infected adults.

BACKGROUND: The basal ganglia (BG) are involved in executive language functions (i.e., verbal fluency) through their connections with cortical structures. The caudate and putamen receive separate inputs from prefrontal and premotor cortices, and may differentially contribute to verbal fluency performance. We examined BG integrity in relation to lexico-semantic verbal fluency performance among older HIV infected adults. METHOD: 20 older (50+ years) HIV+ adults underwent MRI and were administered measures of semantic and phonemic fluency. BG (caudate, putamen) regions of interest were extracted. RESULTS: Performance on phonemic word generation significantly predicted caudate volume, whereas performance on phonemic switching predicted putamen volume. CONCLUSIONS: These findings suggest a double dissociation of BG involvement in verbal fluency tasks with the caudate subserving word generation and the putamen associated with switching. As such, verbal fluency tasks appear to be selective to BG function.