A Stacked Generalization of 3D Orthogonal Deep Learning Convolutional Neural Networks for Improved Detection of White Matter Hyperintensities in 3D FLAIR Images

BACKGROUND AND PURPOSE:Accurate and reliable detection of white matter hyperintensities and their volume quantification can provide valuable clinical information to assess neurologic disease progression. In this work, a stacked generalization ensemble of orthogonal 3D convolutional neural networks, StackGen-Net, is explored for improving automated detection of white matter hyperintensities in 3D T2-FLAIR images.MATERIALS AND METHODS:Individual convolutional neural networks in StackGen-Net were trained on 2.5D patches from orthogonal reformatting of 3D-FLAIR (n = 21) to yield white matter hyperintensity posteriors. A meta convolutional neural network was trained to learn the functional mapping from orthogonal white matter hyperintensity posteriors to the final white matter hyperintensity prediction. The impact of training data and architecture choices on white matter hyperintensity segmentation performance was systematically evaluated on a test cohort (n = 9). The segmentation performance of StackGen-Net was compared with state-of-the-art convolutional neural network techniques on an independent test cohort from the Alzheimer’s Disease Neuroimaging Initiative-3 (n = 20).RESULTS:StackGen-Net outperformed individual convolutional neural networks in the ensemble and their combination using averaging or majority voting. In a comparison with state-of-the-art white matter hyperintensity segmentation techniques, StackGen-Net achieved a significantly higher Dice score (0.76 [SD, 0.08], F1-lesion (0.74 [SD, 0.13]), and area under precision-recall curve (0.84 [SD, 0.09]), and the lowest absolute volume difference (13.3% [SD, 9.1%]). StackGen-Net performance in Dice scores (median = 0.74) did not significantly differ (P = .22) from interobserver (median = 0.73) variability between 2 experienced neuroradiologists. We found no significant difference (P = .15) in white matter hyperintensity lesion volumes from StackGen-Net predictions and ground truth annotations.CONCLUSIONS:A stacked generalization of convolutional neural networks, utilizing multiplanar lesion information using 2.5D spatial context, greatly improved the segmentation performance of StackGen-Net compared with traditional ensemble techniques and some state-of-the-art deep learning models for 3D-FLAIR.

White matter hyperintensities (WMHs) correspond to pathologic features of axonal degeneration, demyelination, and gliosis observed within cerebral white matter.¹ Clinically, the extent of WMHs in the brain has been associated with cognitive impairment, Alzheimer’s disease and vascular dementia, and increased risk of stroke.^2,3 The detection and quantification of WMH volumes to monitor lesion burden evolution and its correlation with clinical outcomes have been of interest in clinical research.^4,5 Although the extent of WMHs can be visually scored,⁶ the categoric nature of such scoring systems makes quantitative evaluation of disease progression difficult. Manually segmenting WMHs is tedious, prone to inter- and intraobserver variability, and is, in most cases, impractical. Thus, there is an increased interest in developing fast, accurate, and reliable computer-aided automated techniques for WMH segmentation.Convolutional neural network (CNN)-based approaches have been successful in several semantic segmentation tasks in medical imaging.⁷ Recent works have proposed using deep learning–based methods for segmenting WMHs using 2D-FLAIR images.^8-11 More recently, a WMH segmentation challenge¹² was also organized (http://wmh.isi.uu.nl/) to facilitate comparison of automated segmentation of WMHs of presumed vascular origin in 2D multislice T2-FLAIR images. Architectures that used an ensemble of separately trained CNNs showed promising results in this challenge, with 3 of the top 5 winners using ensemble-based techniques.¹²Conventional 2D-FLAIR images are typically acquired with thick slices (3–4 mm) and possible slice gaps. Partial volume effects from a thick slice are likely to affect the detection of smaller lesions, both in-plane and out-of-plane. 3D-FLAIR images, with isotropic resolution, have been shown to achieve higher resolution and contrast-to-noise ratio¹³ and have shown promising results in MS lesion detection using 3D CNNs.¹⁴ Additionally, the isotropic resolution enables viewing and evaluation of the images in multiple planes. This multiplanar reformatting of 3D-FLAIR without the use of interpolating kernels is only possible due to the isotropic nature of the acquisition. Network architectures that use information from the 3 orthogonal views have been explored in recent works for CNN-based segmentation of 3D MR imaging data.¹⁵ The use of data from multiple planes allows more spatial context during training without the computational burden associated with full 3D training.¹⁶ The use of 3 orthogonal views simultaneously mirrors how humans approach this segmentation task.Ensembles of CNNs have been shown to average away the variances in the solution and the choice of model- and configuration-specific behaviors of CNNs.¹⁷ Traditionally, the solutions from these separately trained CNNs are combined by averaging or using a majority consensus. In this work, we propose the use of a stacked generalization framework (StackGen-Net) for combining multiplanar lesion information from 3D CNN ensembles to improve the detection of WMH lesions in 3D-FLAIR. A stacked generalization¹⁸ framework learns to combine solutions from individual CNNs in the ensemble. We systematically evaluated the performance of this framework and compared it with traditional ensemble techniques, such as averaging or majority voting, and state-of-the-art deep learning techniques.     

Manejo de la enfermedad de Parkinson y otros trastornos del movimiento en mujeres en edad fértil: Parte 1

IntroductionThe main challenge of Parkinson's disease in women of childbearing age is managing symptoms and drugs during pregnancy and breastfeeding. The increase in the age at which women are having children makes it likely that these pregnancies will become more common in future.ObjectivesThis study aims to define the clinical characteristics of women of childbearing age with Parkinson's disease and the factors affecting their lives, and to establish a series of guidelines for managing pregnancy in these patients.ResultsThis consensus document was developed through an exhaustive literature search and a discussion of the available evidence by a group of movement disorder experts from the Spanish Society of Neurology.ConclusionsParkinson's disease affects all aspects of sexual and reproductive health in women of childbearing age. Pregnancy should be well planned to minimise teratogenic risk. A multidisciplinary approach should be adopted in the management of these patients in order to take all relevant considerations into account.     

Self-care activities,glycaemic control and health-related quality of life of patients with type 2 diabetes in a tertiary hospital in Nigeria

Background and aimsDiabetes is one of the global health emergencies of the 21st century. This study sought to assess self-care behaviours, glycaemic control and health-related quality of life (HRQoL) of type 2 diabetes patients at a Nigerian hospital, determine factors associated with HRQoL, and assess the inter-relationship among these outcomes.MethodsThe Summary of Diabetes Self-Care Activities (SDSCA) measure was used to assess self-care behaviours; HRQoL was assessed with the Appraisal of Diabetes Scale (ADS) and SF-12v2 Health Survey. Glycaemic control was assessed with fasting blood sugar (FBS). The questionnaires were distributed to 250 eligible patients attending Endocrinology clinic at the University of Uyo Teaching Hospital, Akwa Ibom State, Nigeria.ResultsMean scores (±SD) of self-care behaviours were 4.0 ± 1.3, 3.9 ± 1.0, 2.4 ± 1.2, 1.0 ± 1.1, and 2.5 ± 2.1 for general diet, specific diet, exercise, blood glucose monitoring, and foot care respectively. Mean FBS was 7.1 ± 2.1 mmol/L. Mean ADS score was 18.1 ± 4.2. The physical and mental component summary scores were 47.84 ± 8.21 and 45.66 ± 8.74 respectively. HRQoL was significantly associated with patient socio-demographic (except gender) and clinical variables (p < 0.05). Self-care activities, FBS and HRQoL were significantly inter-correlated (r_s ?0.31 to 0.68; p < 0.01, p < 0.001).ConclusionsPatients were more adherent to diet recommendations than to other aspects of self-care. Though most patients had adequate glucose control, the presence of diabetes had a negative impact on the mental component of health. HRQoL was affected by most of the patient-related variables. Interventions to improve adherence to self-care are recommended to achieve good glycaemic control and improve HRQoL.     

Glioma-Induced Disruption of Resting-State Functional Connectivity and Amplitude of Low-Frequency Fluctuations in the Salience Network

BACKGROUND AND PURPOSE:Cognitive challenges are prevalent in survivors of glioma, but their neurobiology is incompletely understood. The purpose of this study was to investigate the effect of glioma presence and tumor characteristics on resting-state functional connectivity and amplitude of low-frequency fluctuations of the salience network, a key neural network associated with cognition.MATERIALS AND METHODS:Sixty-nine patients with glioma (mean age, 48.74 [SD, 14.32] years) who underwent resting-state fMRI were compared with 31 healthy controls (mean age, 49.68 [SD, 15.54] years). We identified 4 salience network ROIs: left/right dorsal anterior cingulate cortex and left/right anterior insula. Average salience network resting-state functional connectivity and amplitude of low-frequency fluctuations within the 4 salience network ROIs were computed.RESULTS:Patients with gliomas showed decreased overall salience network resting-state functional connectivity (P = .001) and increased amplitude of low-frequency fluctuations in all salience network ROIs (P < .01) except in the left dorsal anterior cingulate cortex. Compared with controls, patients with left-sided gliomas showed increased amplitude of low-frequency fluctuations in the right dorsal anterior cingulate cortex (P = .002) and right anterior insula (P < .001), and patients with right-sided gliomas showed increased amplitude of low-frequency fluctuations in the left anterior insula (P = .002). Anterior tumors were associated with decreased salience network resting-state functional connectivity (P < .001) and increased amplitude of low-frequency fluctuations in the right anterior insula, left anterior insula, and right dorsal anterior cingulate cortex. Patients with high-grade gliomas had decreased salience network resting-state functional connectivity compared with healthy controls (P < .05). The right anterior insula showed increased amplitude of low-frequency fluctuations in patients with grade II and IV gliomas compared with controls (P < .01).CONCLUSIONS:By demonstrating decreased resting-state functional connectivity and an increased amplitude of low-frequency fluctuations related to the salience network in patients with glioma, this study adds to our understanding of the neurobiology underpinning observable cognitive deficits in these patients. In addition to more conventional functional connectivity, amplitude of low-frequency fluctuations is a promising functional-imaging biomarker of tumor-induced vascular and neural pathology.

Detrimental effects of cancer on cognitive function and, consequently, on the quality of life are emerging as a key focus of cancer survivorship both in research and clinical practice.^1,2 Brain tumors have been shown to affect memory, processing, and attention in patients; however, their underlying neurobiology is incompletely understood.³ Using resting-state functional MR imaging (rsfMRI) to evaluate changes in cognitive resting-state networks may provide a better understanding of the pathology underlying the observable cognitive disruptions in gliomas, the most common primary brain tumor in adults.A “triple network model” of neurocognitive pathology has been proposed, which encompasses the default mode network, involved in mind wandering; the central executive network, involved in decision-making; and the salience network (SN), implicated in modulating activation of the default mode network and central executive network by detecting the presence of salient stimuli.^4-8 While previous rsfMRI research has largely focused on tumor-induced changes in the default mode network,^9,10 our study examined the less-studied SN, a network rooted in the anterior insula and the dorsal anterior cingulate cortex.⁶Prior studies evaluating gliomas and SN resting-state functional connectivity (RSFC) provided conflicting results in small patient samples: Maesawa et al¹⁰ found no significant differences in the SN in 12 patients, while Liu et al¹¹ more recently found decreased SN connectivity in 13 patients. Gliomas impact the integrity of the neurovascular unit to varying degrees, resulting in neurovascular uncoupling that has been reported to confound fMRI interpretations in patients with brain tumors.^12-14 Additionally, research has reported neuronal plasticity manifested by structural reorganization and functional remodeling of neural networks in patients with gliomas with possible alterations in clinically observable cognitive manifestations.^15-17 An rsfMRI metric, the amplitude of low-frequency fluctuations (ALFF), has recently shown promise as a biomarker for brain plasticity and hemodynamic characterization, including neurovascular uncoupling in patients with gliomas.^15-19The purpose of this study was to investigate the effect of glioma presence and tumor characteristics on overall RSFC and regional normalized ALFF within the SN in a large patient population. We hypothesized that there would be decreased average SN RSFC and altered ALFF in patients with gliomas compared with healthy controls. Recent studies have acknowledged that gliomas have variable effects on network integrity based on lesion location and proximity to network ROIs,^20-22 and unilateral gliomas can be associated with plasticity in both the ipsilateral and contralateral hemispheres.^11,17 Research also supports differences in resting-state network reorganization in aggressive high-grade gliomas compared with slower-growing low-grade gliomas.^20,23 Therefore, we also hypothesized that there would be differences in average SN RSFC and regional ALFF in patients based on the anterior-versus-posterior location, hemispheric side, and grade of glioma.