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.     

Minimum clinically important difference of the health-related quality of life scales in adult spinal deformity calculated by latent class analysis: is it appropriate to use the same values for surgical and nonsurgical patients?

BACKGROUND CONTEXT

Health-related quality of life (HRQOL) parameters have been shown to be reliable and valid in patients with adult spinal deformity (ASD). Minimum clinically important difference (MCID) has become increasingly important to clinicians in evaluating patients with a threshold of improvement that is clinically relevant.

The effect of metoclopramide on gastric emptying in traumatic brain injury.

OBJECTIVE: Gastric paresis in traumatic brain injury (TBI) hinders the effectiveness of enteral support in this patient group. In this study we have investigated the effect of metoclopramide on gastric emptying in TBI patients. METHOD: In this prospective, randomized, controlled, double-blind study, 19 TBI patients with Glasgow Coma Scale scores of 3-11 were included. In all patients, enteral nutrition was commenced with a nasogastric feeding tube within 48 hours of trauma. Patients were randomized into two groups. In the metoclopramide (M) group, 10 mg metoclopramide was delivered intravenously three times daily for 5 days. In the control (C) group, an equal volume of saline was administered. Besides demographics, gastric emptying according to a paracetamol absorption test at days 0 and 5, time to reach target nutritional requirements, gastric residues, intolerance to feeding, nutritional complications, and clinical outcomes were recorded for each patient. RESULTS: The gastric residue rates were 2.7+/-7.4 mL and 8.1+/-17.7 mL per 100 patient days for groups C and M respectively (p=0.408). Similarly, feeding intolerance and complication rates did not significantly differ between groups C and M, (respectively p=0.543 and 0.930). Gastric emptying parameters also were similar between the study groups. CONCLUSION: We were unable to document any advantage to using metoclopramide in TBI patients. Simple intragastric enteral feeding with close monitoring of the possible complications seems to be sufficient with acceptable morbidity rates.     

Effect of in utero growth retardation on lung function at follow-up of prematurely born infants.

The aim of the study was to determine if prematurely born children who had suffered intra-uterine growth retardation (IUGR) had more severe lung function abnormalities than those born an appropriate weight for gestational age (AGA). Analysis of the lung function results of 119 infants (median (range) gestational age of 30 (23-35) weeks) was undertaken. In total, 31 of the infants had suffered IUGR and were born small for gestational age (SGA). Functional residual capacity and airways resistance (Raw) were measured at a median post-natal age of 10 (6-24) months. Specific airway conductance (sGaw) was calculated from thoracic gas volume and Raw. The SGA children were born at a greater gestational age and had a lower body weight at testing than the AGA children. Raw and sGaw differed between the SGA and AGA children. Regression analysis demonstrated that lung volumes were significantly related to body weight at testing, Raw was related to IUGR, maternal smoking and bronchopulmonary dysplasia, and sGaw to maternal smoking. In conclusion, these results suggest that prematurely born infants who have suffered intra-uterine growth retardation may be at increased risk of impaired lung function at follow-up.     

SENSORINEURAL HEARING LOSS IN PSEUDOEXFOLIATION SYNDROME

Purpose: to determine wheather an association exists between sensorineural hearing loss and pseudoexfoliation. Methods: Hearing thresholds were determined by using pure-tone au-diometry in 75 patients with pseudoexfoliation syndrome and in the control group of 75 subjects without pseudoexfoliation The groups were sex and age matched. Results: The frequency of     

CYSTOID MACULAR EDEMA ASSOCIATED WITH LATANOPROST

Purpose: To report clinically, anotomically and angiographically documented cystoid macular edema (CME) associated with the use of latanoprost in 2 uncomplicated pseudophakic eyes. Methods: Retrospective rewiev of 2 patients who had history of latanoprost use and uncomplicated cataract surgery, described     

Optimization of self-reference thermometry using complex field estimation.

Referenceless, or self-reference, thermometry is a technique for mapping temperature differences in the region of interest (ROI) using the baseline phase estimated by extrapolating the field in the surrounding region for estimation (RFE) and subtracting the estimated baseline from the measured field. In the present work a self-reference technique based on complex field estimation using 2D polynomials comprising complex-valued coefficients was proposed and optimized. Numerical simulations with a Gaussian-profiled phase distribution demonstrated that the ROI radius had to be 2.3-2.5 times the standard deviation (SD) of the Gaussian function in order to keep the error below 8% of the peak phase change. The area ratio between the ROI and the RFE had to be larger than 2.0 to maintain the error level. Based on the simulations, and phantom and volunteer experiments, the complex-based method with independently optimized polynomial orders for the two spatial dimensions was compared with the phase-based method using the similar-order optimization strategy. The complex-based method appeared to be useful when phase unwrapping was not removed. Otherwise, the phase-based method yielded equivalent results with less polynomial orders.     

Intracoronary shunt reduces postoperative troponin leaks: a prospective randomized study.

OBJECTIVE: The purpose of this study was to evaluate whether intracoronary shunt usage reduced the myocardial damage on the basis of the cardiac markers when compared with the shuntless anastomosis in off-pump coronary artery bypass grafting (OPCABG) surgery of isolated left anterior descending artery lesions. METHODS: Forty patients who had stable angina with isolated left anterior descending (LAD) coronary artery lesion undergoing OPCABG surgery were randomized into two groups. Shunt group consisted of 20 patients who had OPCABG using intracoronary shunt, whereas the shuntless group consisted of 20 patients who underwent OPCABG without using intracoronary shunt. Cardiac troponin I, CK, and CK-MB before and 24h after the surgery were assessed in the groups. RESULTS: There were no deaths in the study. The two groups were similar with respect to sex and age. Duration of LIMA-LAD anastomosis was significantly higher in the shunt group (p=0.01). There was no significant difference between the groups concerning the preoperative and postoperative CK and CK-MB levels. The preoperative troponin I levels of the groups were not different (p=0.238; NS), whereas postoperative levels of this marker was significantly higher in the shuntless group (p=0.003). CONCLUSION: Intracoronary shunt reduced the postoperative troponin I levels significantly, so it may be indicated in the patients who are thought to be susceptible to transient ischemia.     

Fast decomposition of water and lipid using a GRASE technique with the IDEAL algorithm.

Three-point Dixon techniques achieve good lipid-water separation by estimating the phase due to field inhomogeneities. Recently it was demonstrated that the combination of an iterative algorithm (iterative decomposition of water and fat with echo asymmetry and least-squares estimation (IDEAL)) with a fast spin-echo (FSE) three-point Dixon method yielded robust lipid-water decomposition. As an alternative to FSE, the gradient- and spin-echo (GRASE) technique has been developed for efficient data collection. In this work we present a method for lipid-water separation by combining IDEAL with the GRASE technique. An approach to correct for errors in the lipid-water decomposition caused by phase distortions due to the switching of the readout gradient polarities inherent to GRASE is presented. The IDEAL-GRASE technique is demonstrated in phantoms and in vivo for various applications, including pelvic, musculoskeletal, and (breath-hold) cardiac imaging.     

Susceptibilities of Mycobacterium tuberculosis to isoniazid and rifampin on blood agar

In this study, blood agar was used instead of 7H10 agar for the susceptibility testing of 34 clinical isolates of Mycobacterium tuberculosis to isoniazid (INH) and rifampin (RIF) in accordance with the NCCLS. The BACTEC 460 TB system (Becton Dickinson, Sparks, Md.) was used as a "gold standard." Results for both media were in agreement for RIF and INH at 100 and 94.1%, respectively. For INH, the specificity, sensitivity, positive predictive value, and negative predictive value were found to be 71.4, 100, 93.1, and 100%, respectively, while these values were 100% for RIF. In addition, the results of the susceptibility test performed with blood agar were obtained on day 14 of incubation. In conclusion, results were obtained much earlier with blood agar (2 weeks) than with 7H10 agar (3 weeks), and the results of this study suggest that blood agar may be used as an alternative medium for the susceptibility testing of M. tuberculosis to INH and RIF.