Gregory J. Moore

A Comparison of Structural Brain Imaging Findings in Autism Spectrum Disorder and Attention-Deficit Hyperactivity Disorder

ASD and ADHD are regarded as distinct disorders in the current DSM-5. However, recent research and the RDoC initiative are recognizing considerable overlap in the clinical presentation of ASD, ADHD, and other neurodevelopmental disorders. In spite of numerous neuroimaging findings in ASD and ADHD, the extent to which either of the above views are supported remains equivocal. Here we compare structural MRI and DTI literature in ASD and ADHD. Our main findings reveal both distinct and shared neural features. Distinct expressions were in total brain volume (ASD: increased volume, ADHD: decreased volume), amygdala (ASD: overgrowth, ADHD: normal), and internal capsule (ASD: unclear, ADHD: reduced FA in DTI). Considerable overlap was noted in the corpus callosum and cerebellum (lower volume in structural MRI and decreased FA in DTI), and superior longitudinal fasciculus (reduced FA in DTI). In addition, we identify brain regions which have not been studied in depth and require more research. We discuss relationships between brain features and symptomatology. We conclude by addressing limitations of current neuroimaging research and offer approaches that account for clinical heterogeneity to better distinguish brain-behavior relationships.

Accurate segmentation of lung fields on chest radiographs using deep convolutional networks.

Accurate segmentation of lung fields on chest radiographs is the primary step for computer-aided detection of various conditions such as lung cancer and tuberculosis. The size, shape and texture of lung fields are key parameters for chest X-ray (CXR) based lung disease diagnosis in which the lung field segmentation is a significant primary step. Although many methods have been proposed for this problem, lung field segmentation remains as a challenge. In recent years, deep learning has shown state of the art performance in many visual tasks such as object detection, image classification and semantic image segmentation. In this study, we propose a deep convolutional neural network (CNN) framework for segmentation of lung fields. The algorithm was developed and tested on 167 clinical posterior-anterior (PA) CXR images collected retrospectively from picture archiving and communication system (PACS) of Geisinger Health System. The proposed multi-scale network is composed of five convolutional and two fully connected layers. The framework achieved IOU (intersection over union) of 0.96 on the testing dataset as compared to manual segmentation. The suggested framework outperforms state of the art registration-based segmentation by a significant margin. To our knowledge, this is the first deep learning based study of lung field segmentation on CXR images developed on a heterogeneous clinical dataset. The results suggest that convolutional neural networks could be employed reliably for lung field segmentation.

Inter-Method Discrepancies in Brain Volume Estimation May Drive Inconsistent Findings in Autism

Previous studies applying automatic preprocessing methods on Structural Magnetic Resonance Imaging (sMRI) report inconsistent neuroanatomical abnormalities in Autism Spectrum Disorder (ASD). In this study we investigate inter-method differences as a possible cause behind these inconsistent findings. In particular, we focus on the estimation of the following brain volumes: gray matter (GM), white matter (WM), cerebrospinal fluid (CSF), and total intra cranial volume (TIV). T1-weighted sMRIs of 417 ASD subjects and 459 typically developing controls (TDC) from the ABIDE dataset were estimated using three popular preprocessing methods: SPM, FSL, and FreeSurfer (FS). Brain volumes estimated by the three methods were correlated but had significant inter-method differences; except TIVSPM vs. TIVFS, all inter-method differences were significant. ASD vs. TDC group differences in all brain volume estimates were dependent on the method used. SPM showed that TIV, GM, and CSF volumes of ASD were larger than TDC with statistical significance, whereas FS and FSL did not show significant differences in any of the volumes; in some cases, the direction of the differences were opposite to SPM. When methods were compared with each other, they showed differential biases for autism, and several biases were larger than ASD vs. TDC differences of the respective methods. After manual inspection, we found inter-method segmentation mismatches in the cerebellum, sub-cortical structures, and inter-sulcal CSF. In addition, to validate automated TIV estimates we performed manual segmentation on a subset of subjects. Results indicate that SPM estimates are closest to manual segmentation, followed by FS while FSL estimates were significantly lower. In summary, we show that ASD vs. TDC brain volume differences are method dependent and that these inter-method discrepancies can contribute to inconsistent neuroimaging findings in general. We suggest cross-validation across methods and emphasize the need to develop better methods to increase the robustness of neuroimaging findings.

Neural substrates of a schizotypal spectrum in typically-developing children: Further evidence of a normal-pathological continuum

Schizophrenia represents the extreme end of a distribution of traits that extends well into the general population. Using a recently developed measure of psychotic-like traits in children, we examined the neural substrates of psychotic (and other psychiatric) symptoms using structural magnetic resonance imaging (MRI). Twenty-eight typically-developing children (14 males) between the ages of 6-17 years underwent a 3T MRI scan. Parents completed the Psychiatric and Schizotypal Inventory for Children. Results revealed that caudate, amygdala, hippocampal and middle temporal gyrus volumes were associated with quantitative dimensions of psychiatric traits. Furthermore, results suggest a differential a sexually-dimorphic pattern of brain-schizotypy associations. These findings highlight brain-behavior continuities between clinical conditions such as schizophrenia and normal trait variation in typical development.

Advanced machine learning in action: identification of intracranial hemorrhage on computed tomography scans of the head with clinical workflow integration

Intracranial hemorrhage (ICH) requires prompt diagnosis to optimize patient outcomes. We hypothesized that machine learning algorithms could automatically analyze computed tomography (CT) of the head, prioritize radiology worklists and reduce time to diagnosis of ICH. 46,583 head CTs (~2 million images) acquired from 2007–2017 were collected from several facilities across Geisinger. A deep convolutional neural network was trained on 37,074 studies and subsequently evaluated on 9499 unseen studies. The predictive model was implemented prospectively for 3 months to re-prioritize “routine” head CT studies as “stat” on realtime radiology worklists if an ICH was detected. Time to diagnosis was compared between the re-prioritized “stat” and “routine” studies. A neuroradiologist blinded to the study reviewed false positive studies to determine whether the dictating radiologist overlooked ICH. The model achieved an area under the ROC curve of 0.846 (0.837–0.856). During implementation, 94 of 347 “routine” studies were re-prioritized to “stat”, and 60/94 had ICH identified by the radiologist. Five new cases of ICH were identified, and median time to diagnosis was significantly reduced (p < 0.0001) from 512 to 19 min. In particular, one outpatient with vague symptoms on anti-coagulation was found to have an ICH which was treated promptly with reversal of anticoagulation, resulting in a good clinical outcome. Of the 34 false positives, the blinded over-reader identified four probable ICH cases overlooked in original interpretation. In conclusion, an artificial intelligence algorithm can prioritize radiology worklists to reduce time to diagnosis of new outpatient ICH by 96% and may also identify subtle ICH overlooked by radiologists. This demonstrates the positive impact of advanced machine learning in radiology workflow optimization.A computer program that automatically analyzes brain images from patients undergoing CT scans of the head can reliably flag those with signs of hemorrhage. A team of researchers from Geisinger in Danville, Pennsylvania, USA, trained and tested a machine-learning algorithm using 46,583 computed tomography imaging studies of the head. Subsequently, they implemented the model into routine care for 3 months to help prioritize radiology worklists. Of 347 routine cases, the computer identified 94 as having an intracranial hemorrhage, two-thirds of which were confirmed by a radiologist, including five among patients who had a new diagnosis of a brain bleed. The algorithm reduced the average time in which a radiologist diagnosed these patients from around 8.5 h to just 19 min, demonstrating the positive impact of incorporating artificial intelligence into radiology workflow.

Automatic estimation of heart boundaries and cardiothoracic ratio from chest x-ray images

Cardiothoracic ratio (CTR) is a widely used radiographic index to assess heart size on chest X-rays (CXRs). Recent studies have suggested that also two-dimensional CTR might contain clinical information about the heart function. However, manual measurement of such indices is both subjective and time consuming. This study proposes a fast algorithm to automatically estimate CTR indices based on CXRs. The algorithm has three main steps: 1) model based lung segmentation, 2) estimation of heart boundaries from lung contours, and 3) computation of cardiothoracic indices from the estimated boundaries. We extended a previously employed lung detection algorithm to automatically estimate heart boundaries without using ground truth heart markings. We used two datasets: a publicly available dataset with 247 images as well as clinical dataset with 167 studies from Geisinger Health System. The models of lung fields are learned from both datasets. The lung regions in a given test image are estimated by registering the learned models to patient CXRs. Then, heart region is estimated by applying Harris operator on segmented lung fields to detect the corner points corresponding to the heart boundaries. The algorithm calculates three indices, CTR1D, CTR2D, and cardiothoracic area ratio (CTAR). The method was tested on 103 clinical CXRs and average error rates of 7.9%, 25.5%, and 26.4% (for CTR1D, CTR2D, and CTAR respectively) were achieved. The proposed method outperforms previous CTR estimation methods without using any heart templates. This method can have important clinical implications as it can provide fast and accurate estimate of cardiothoracic indices.

Conus Medullaris Level in Vertebral Columns With Lumbosacral Transitional Vertebra

BACKGROUND The estimated prevalence of lumbar or sacral transitional vertebrae (LSTV) in the population is 4% to 30%. Few small patient series have studied the normal level of the conus medullaris (CM) in individuals with LSTV. OBJECTIVE To determine, by using a large cohort of patients, whether individuals of all ages with LSTV have different CM positions in the spinal canal in comparison with the rest of the population with normal vertebral columns. METHODS We performed an institutional retrospective analysis of spinal magnetic resonance images on individuals with LSTV of all ages, sexes, and pathologies during a 10-year period. Fifty-seven percent of patients (n = 467) had a lumbarized vertebra and 43% had sacralized vertebra (n = 355). Mean age at the time of the study was 55 ± 19 years (range 1-97 years). Fifty-two percent were male and 48% were female. Sixty percent of subjects with a sacralized vertebra were female, and 54.5% of those with a lumbarized vertebra were male (P = .001). RESULTS The CM in individuals with a lumbarized vertebra was seen to be lower at L1-2 to L2s, than un those with a sacralized vertebra where most conuses were at T12-L1 to L1s (P ≤ 0.001). The CM level was similarly distributed among sexes and ages. CONCLUSION In our series, the CM level, when lumbarization occurred, was lower, with a mean level at L1-L2, whereas a more superior mean level at T12-L1 was seen when sacralization occurred. CM level was not influenced by sex, age, or pathology other than tethered cords.

Influence of Group on Individual Subject Maps in SPM Voxel Based Morphometry

Voxel based morphometry (VBM) is a widely utilized neuroimaging technique for spatially normalizing brain structural MRI (sMRI) onto a common template. The DARTEL technique of VBM takes into account the spatial intensity distribution of sMRIs to construct a study specific group template. The group template is then used to create final individual normalized tissue maps (FINTM) for each subject in the group. In this study, we investigate the effect of group on FINTM, i.e., we evaluate the variability of a constant subjects FINTM when other subjects in the group are iteratively changed. We examine this variability under the following scenarios: (1) when the demographics of the iterative groups are similar, (2) when the average age of the iterative groups is increased, and (3) when the number of subjects with a brain disorder (here we use subjects with autism) is increased. Our results show that when subject demographics of the group remains similar the mean standard deviation (SD) of FINTM gray matter (GM) of the constant subject was around 0.01. As the average age of the group is increased, mean SD of GM increased to around 0.03 and at certain brain locations variability was as high as 0.23. A similar increase in variability was observed when the number of autism subjects in the group was increased where mean SD was around 0.02. Further, we find that autism vs. control GM differences are in the range of −0.05 to +0.05 for more than 97% of the voxels and note that the magnitude of the differences are comparable to GM variability. Finally, we report that opting not to modulate during normalization or increasing the size of the smoothing kernel can decrease FINTM variability but at the loss of subject-specific features. Based on the findings of this study, we outline precautions that should be considered by investigators to reduce the impact of group on FINTM and thereby derive more meaningful group differences when comparing two cohorts of subjects.

Implementation of a patient-facing genomic test report in the electronic health record using a web-application interface

BackgroundGenomic medicine is emerging into clinical care. Communication of genetic laboratory results to patients and providers is hampered by the complex technical nature of the laboratory reports. This can lead to confusion and misinterpretation of the results resulting in inappropriate care. Patients usually do not receive a copy of the report leading to further opportunities for miscommunication. To address these problems, interpretive reports were created using input from the intended end users, patients and providers. This paper describes the technical development and deployment of the first patient-facing genomic test report (PGR) within an electronic health record (EHR) ecosystem using a locally developed standards-based web-application interface.MethodsA patient-facing genomic test report with a companion provider report was configured for implementation within the EHR using a locally developed software platform, COMPASS™. COMPASS™ is designed to manage secure data exchange, as well as patient and provider access to patient reported data capture and clinical display tools. COMPASS™ is built using a Software as a Service (SaaS) approach which exposes an API that apps can interact with.ResultsAn authoring tool was developed that allowed creation of patient-specific PGRs and the accompanying provider reports. These were converted to a format that allowed them to be presented in the patient portal and EHR respectively using the existing COMPASS™ interface thus allowing patients, caregivers and providers access to individual reports designed for the intended end user.ConclusionsThe PGR as developed was shown to enhance patient and provider communication around genomic results. It is built on current standards but is designed to support integration with other tools and be compatible with emerging opportunities such as SMART on FHIR. This approach could be used to support genomic return of results as the tool is scalable and generalizable.

Experience with Integrating Diagnostic Decision Support Software with Electronic Health Records: Benefits versus Risks of Information Sharing

Introduction: Reducing misdiagnosis has long been a goal of medical informatics. Current thinking has focused on achieving this goal by integrating diagnostic decision support into electronic health records. Methods: A diagnostic decision support system already in clinical use was integrated into electronic health record systems at two large health systems, after clinician input on desired capabilities. The decision support provided three outputs: editable text for use in a clinical note, a summary including the suggested differential diagnosis with a graphical representation of probability, and a list of pertinent positive and pertinent negative findings (with onsets). Results: Structured interviews showed widespread agreement that the tool was useful and that the integration improved workflow. There was disagreement among various specialties over the risks versus benefits of documenting intermediate diagnostic thinking. Benefits were most valued by specialists involved in diagnostic testing, who were able to use the additional clinical context for richer interpretation of test results. Risks were most cited by physicians making clinical diagnoses, who expressed concern that a process that generated diagnostic possibilities exposed them to legal liability. Discussion and Conclusion: Reconciling the preferences of the various groups could include saving only the finding list as a patient-wide resource, saving intermediate diagnostic thinking only temporarily, or adoption of professional guidelines to clarify the role of decision support in diagnosis.