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.     

105Changes in Nitric Oxide Levels After Deep Dermal injury in the Female,Red Duroc Pig (FRDP)

Introduction : Hypertrophic scar is a devastating sequel to burns and other tangential skin injuries. It follows deep dermal injuries and does not occur after superficial injuries. Nitric oxide (NO) plays many important roles in wound healing from inflammation to scar remodeling. Studies have shown that expression of nitric oxide synthase and nitric oxide production are decreased in human hypertrophic scar. However little is known about NO involvement in the early stages of hypertrophic scarring, because of the lack of an animal model. It was recently reported that the female red Duroc pig (FRDP) makes thick scar, which is similar to human hypertrophic scar. We hypothesized that NO production in wounds on the female, red Duroc pig is similar to that of human hypertrophic scar and that NO involvement in deep wounds is different from that in superficial wounds. Methods : Superficial (0.015” to 0.030”) and deep (0.045” to 0.060”) wounds were created on the backs of four FRDPs. Biopsies were collected at weeks 1.5, 4, 8 and 21 post wounding including samples of uninjured skin. Nitric oxide levels were measured with the Griess reaction assay and normalized with tissue protein level. Results : Superficial wounds healed with an invisible scar whereas the deep wounds healed with scar resembling mild hypertrophic scar. The thickness of the scars from the deep wounds was significantly greater than uninjured skin and healed superficial wounds (p < 0.01). NO levels were increased at 1.5 weeks in deep wounds compared to superficial wounds and uninjured skin (p < 0.05). At 8 weeks, NO levels in deep wounds had returned to the level of uninjured tissue and superficial wounds. By 21 weeks, NO levels had decreased significantly when compared to superficial wounds (p < 0.01). There were no differences in NO levels between uninjured skin and superficial wounds at any time point (p > 0.05). Conclusions : NO production is similar in late, deep wounds on the female, red Duroc pig to that reported in the literature for human hypertrophic scar further validating this animal model. NO production is quite different after deep wounds as compared to superficial wounds in the FRDP. Early elevation in nitric oxide production might account for excessive inflammation in deep wounds that become thick scars in the FRDP. Nitric oxide regulators and effects at early stages of scar formation should be elucidated further and the FRDP appears to be a useful model.     

TREATMENT OF PALMOPLANTER HYPERHIDROSIS BY IONTOPHORESIS

Twenty-seven patients with idiopathic palmoplanter hyperhidrosis were treated with Iontotherapy over a one year period. In twenty-four cases there was a good response but maintenance therapy was required every 3-4 weeks.KEY WORDS: Iontophoresis, Palmoplanter hyperhidrosis     

B19 parvovirus replicates in circulating cells of acutely infected patients

B19 parvovirus is the etiologic agent of fifth disease and transient aplastic crisis. In natural infections, B19 antigen and DNA have been detected in sera early in the course of aplastic crisis and only rarely in fifth disease. We have found B19 DNA in circulating cells of infected patients by DNA dot blot with a virus-specific probe: in four of four sickle cell patients with aplastic crisis, in one asymptomatic sibling, and in one normal adult with fifth disease. Only two of the sera showed B19 DNA. High-molecular weight intermediate forms were detected by Southern analysis of DNA extracted from cells, thus indicating active replication of virus in cells rather than passive adsorption to their surface membranes. Separation of cells into high- and low-density fractions resulted in a concentration of the virus DNA in the granulocytic fraction.     

Chronic synovitis and HLA B27 in patients with severe haemophilia

Chronic synovitis affects about 10% of patients with severe haemophilia in India. This disease has some features in common with ankylosing spondylitis, which has been linked to HLA B27. We therefore aimed to test whether there is an association between HLA B27 and chronic synovitis. We studied 473 patients with severe haemophilia (33 of whom had chronic synovitis), and 1175 healthy controls using a standard serological technique and the reverse line strip assay. 64% (21 of 33) of patients with haemophilia and chronic synovitis were positive for HLA B27, compared with 5% (23 of 440) of those with severe haemophilia, but not chronic synovitis (odds ratio 31.6 [95% CI 9.28-39.38], p<0.0001), and 9% (100 of 1175) of healthy controls (18.81 [9.6-27.7], p<0.0001). We conclude that there is a strong association between HLA B27 and chronic synovitis in Indian patients with severe haemophilia and screening in this population could allow treatment and prevention of the complication.     

Identification of glycoprotein Ib as a target for autoantibody in idiopathic (autoimmune) thrombocytopenic purpura

An antibody (DIL) from a patient with idiopathic thrombocytopenic purpura (ITP) was shown to have autospecificity on the basis of reactions with autologous platelets that were identical to those obtained with platelets from normal subjects. DIL antibody also reacted strongly in an immunofluorescence test with platelets from a patient with Glanzmann's thrombasthenia, but failed to react with platelets from a patient with the Bernard-Soulier syndrome who was known to be deficient in glycoprotein Ib (GPIb). Purified GPIb and control platelets, but not Bernard-Soulier platelets, inhibited the lytic activity of DIL. Using the GPIb-specific monoclonal antibody AP1 and one-dimensional rocket electrophoresis into gels containing rabbit antihuman platelet membrane antibody, it was shown that staphylococcal protein A-Sepharose beads coated with DIL antibody selectively remove GPIb from solubilized platelet preparations. By crossed immunoelectrophoresis it was found that DIL recognizes a determinant on GPIb on the membrane side of the cleavage site of the platelet calcium- activated protease (calpain). These studies provide direct evidence for binding of a platelet autoantibody to a determinant on GPIb relatively close to the site of insertion of this protein into the platelet membrane.