A pinboard by
Botian Xu

Research Assistant, University of Southern California


Develop a learning-based fully automated lesion segmentation algorithm in sickle cell disease

The aim of this project is to develop a fully automated lesion segmentation system, instead of traditional manual/semi-auto segmentation, in sickle cell anemia children; a subset of the Silent Cerebral Infarct Multi-Center Clinical Trail magnetic resonance imaging (MRI) dataset. SIT is a National Institute of Neurological Disorders and Stroke study to determine the effectiveness of blood transfusion for prevention of stroke in children with sickle cell disease (SCD). MRI data was collected at 29 clinical centers worldwide. Participants ranged from 5 to 14 years old having sickle cell anemia. The data from all 197 children was analyzed by neuroradiologists, and more than 750 silent cerebral infarcts (SCIs) in total were identified from the MRI images. In contrast to overt stroke, SCI is not associated with prominent neurological impairment. However, children with a SCI are at higher risk for poor academic achievement and lower IQ, as compared with children with SCD having normal results on magnetic resonance imaging (MRI). In addition, SCI has been linked to a 14-fold increase in stroke. MRI makes it possible to unveil SCI, in the form of abnormal signals on T2-weighted MRI images. However, manually delineating the contours of each lesion would be cumbersome and laborious, leading to subject bias. In recent years, computer-aided segmentation algorithms have been developed, allowing more objective and faster segmentation. Despite this, the accuracy of these methods relies on the dataset or the type of the MRI sequences. In addition, current learning-based methods (artificial intelligence techniques) performed well on large lesion load segmentation, but are not sensitive enough to detect small lesions. Given the large dataset with small lesion load, we will test k-nearest neighbor (kNN) as a starting point and set a baseline in order to evaluate the performance of more complex algorithms. We will design convolutional neural networks (CNNs) to further decrease errors. We will project the MRI and the segmented results to the common imaging space and build a lesion distribution map, which also have applications and benefits beyond patients with SCD. SCIs are observed in blood transfusion independent anemia patients such as thalassemia. SCIs are also common in elderly patients with age-related vascular disease. Developing a distribution map of SCI could unearth the onset and severity of lesions, and may prove translational across these populations.


Interventions for preventing silent cerebral infarcts in people with sickle cell disease.

Abstract: Sickle cell disease (SCD) is one of the commonest severe monogenic disorders in the world, due to the inheritance of two abnormal haemoglobin (beta globin) genes. SCD can cause severe pain, significant end-organ damage, pulmonary complications, and premature death. Silent cerebral infarcts are the commonest neurological complication in children and probably adults with SCD. Silent cerebral infarcts also affect academic performance, increase cognitive deficits and may lower intelligence quotient.To assess the effectiveness of interventions to reduce or prevent silent cerebral infarcts in people with SCD.We searched for relevant trials in the Cochrane Library, MEDLINE (from 1946), Embase (from 1974), the Transfusion Evidence Library (from 1980), and ongoing trial databases; all searches current to 19 September 2016. We searched the Cochrane Cystic Fibrosis and Genetic Disorders Group Trials Register: 06 October 2016.Randomised controlled trials comparing interventions to prevent silent cerebral infarcts in people with SCD. There were no restrictions by outcomes examined, language or publication status.We used standard Cochrane methodological procedures.We included five trials (660 children or adolescents) published between 1998 and 2016. Four of the five trials were terminated early. The vast majority of participants had the haemoglobin (Hb)SS form of SCD. One trial focused on preventing silent cerebral infarcts or stroke; three trials were for primary stroke prevention and one trial dealt with secondary stroke prevention.Three trials compared the use of regular long-term red blood cell transfusions to standard care. Two of these trials included children with no previous long-term transfusions: one in children with normal transcranial doppler (TCD) velocities; and one in children with abnormal TCD velocities. The third trial included children and adolescents on long-term transfusion.Two trials compared the drug hydroxyurea and phlebotomy to long-term transfusions and iron chelation therapy: one in primary prevention (children), and one in secondary prevention (children and adolescents).The quality of the evidence was moderate to very low across different outcomes according to GRADE methodology. This was due to trials being at high risk of bias because they were unblinded; indirectness (available evidence was only for children with HbSS); and imprecise outcome estimates. Long-term red blood cell transfusions versus standard care Children with no previous long-term transfusions and higher risk of stroke (abnormal TCD velocities or previous history of silent cerebral infarcts) Long-term red blood cell transfusions may reduce the incidence of silent cerebral infarcts in children with abnormal TCD velocities, risk ratio (RR) 0.11 (95% confidence interval (CI) 0.02 to 0.86) (one trial, 124 participants, low-quality evidence); but make little or no difference to the incidence of silent cerebral infarcts in children with previous silent cerebral infarcts on magnetic resonance imaging and normal or conditional TCDs, RR 0.70 (95% CI 0.23 to 2.13) (one trial, 196 participants, low-quality evidence).No deaths were reported in either trial.Long-term red blood cell transfusions may reduce the incidence of: acute chest syndrome, RR 0.24 (95% CI 0.12 to 0.49) (two trials, 326 participants, low-quality evidence); and painful crisis, RR 0.63 (95% CI 0.42 to 0.95) (two trials, 326 participants, low-quality evidence); and probably reduces the incidence of clinical stroke, RR 0.12 (95% CI 0.03 to 0.49) (two trials, 326 participants, moderate-quality evidence).Long-term red blood cell transfusions may improve quality of life in children with previous silent cerebral infarcts (difference estimate -0.54; 95% confidence interval -0.92 to -0.17; one trial; 166 participants), but may have no effect on cognitive function (least squares means: 1.7, 95% CI -1.1 to 4.4) (one trial, 166 participants, low-quality evidence). Transfusions continued versus transfusions halted: children and adolescents with normalised TCD velocities (79 participants; one trial)Continuing red blood cell transfusions may reduce the incidence of silent cerebral infarcts, RR 0.29 (95% CI 0.09 to 0.97 (low-quality evidence).We are very uncertain whether continuing red blood cell transfusions has any effect on all-cause mortality, Peto odds ratio (OR) 8.00 (95% CI 0.16 to 404.12); or clinical stroke, RR 0.22 (95% CI 0.01 to 4.35) (very low-quality evidence).The trial did not report: comparative numbers for SCD-related adverse events; quality of life; or cognitive function. Hydroxyurea and phlebotomy versus transfusions and chelation Primary prevention, children (121 participants; one trial)We are very uncertain whether switching to hydroxyurea and phlebotomy has any effect on: silent cerebral infarcts (no infarcts); all-cause mortality (no deaths); risk of stroke (no strokes); or SCD-related complications, RR 1.52 (95% CI 0.58 to 4.02) (very low-quality evidence). Secondary prevention, children and adolescents with a history of stroke (133 participants; one trial)We are very uncertain whether switching to hydroxyurea and phlebotomy has any effect on: silent cerebral infarcts, Peto OR 7.28 (95% CI 0.14 to 366.91); all-cause mortality, Peto OR 1.02 (95%CI 0.06 to 16.41); or clinical stroke, RR 14.78 (95% CI 0.86 to 253.66) (very low-quality evidence).Switching to hydroxyurea and phlebotomy may increase the risk of SCD-related complications, RR 3.10 (95% CI 1.42 to 6.75) (low-quality evidence).Neither trial reported on quality of life or cognitive function.We identified no trials for preventing silent cerebral infarcts in adults, or in children who do not have HbSS SCD.Long-term red blood cell transfusions may reduce the incidence of silent cerebral infarcts in children with abnormal TCD velocities, but may have little or no effect on children with normal TCD velocities. In children who are at higher risk of stroke and have not had previous long-term transfusions, long-term red blood cell transfusions probably reduce the risk of stroke, and other SCD-related complications (acute chest syndrome and painful crises).In children and adolescents at high risk of stroke whose TCD velocities have normalised, continuing red blood cell transfusions may reduce the risk of silent cerebral infarcts. No treatment duration threshold has been established for stopping transfusions.Switching to hydroxyurea with phlebotomy may increase the risk of silent cerebral infarcts and SCD-related serious adverse events in secondary stroke prevention.All other evidence in this review is of very low-quality.

Pub.: 14 May '17, Pinned: 02 Jul '17

Hemoglobin and mean platelet volume predicts diffuse T1-MRI white matter volume decrease in sickle cell disease patients.

Abstract: Sickle cell disease (SCD) is a life-threatening genetic condition. Patients suffer from chronic systemic and cerebral vascular disease that leads to early and cumulative neurological damage. Few studies have quantified the effects of this disease on brain morphometry and even fewer efforts have been devoted to older patients despite the progressive nature of the disease. This study quantifies global and regional brain volumes in adolescent and young adult patients with SCD and racially matched controls with the aim of distinguishing between age related changes associated with normal brain maturation and damage from sickle cell disease. T1 weighted images were acquired on 33 clinically asymptomatic SCD patients (age = 21.3 ± 7.8; F = 18, M = 15) and 32 racially matched control subjects (age = 24.4 ± 7.5; F = 22, M = 10). Exclusion criteria included pregnancy, previous overt stroke, acute chest, or pain crisis hospitalization within one month. All brain volume comparisons were corrected for age and sex. Globally, grey matter volume was not different but white matter volume was 8.1% lower (p = 0.0056) in the right hemisphere and 6.8% (p = 0.0068) in the left hemisphere in SCD patients compared with controls. Multivariate analysis retained hemoglobin (β = 0.33; p = 0.0036), sex (β = 0.35; p = 0.0017) and mean platelet volume (β = 0.27; p = 0.016) as significant factors in the final prediction model for white matter volume for a combined r(2) of 0.37 (p < 0.0001). Lower white matter volume was confined to phylogenetically younger brain regions in the anterior and middle cerebral artery distributions. Our findings suggest that there are diffuse white matter abnormalities in SCD patients, especially in the frontal, parietal and temporal lobes, that are associated with low hemoglobin levels and mean platelet volume. The pattern of brain loss suggests chronic microvascular insufficiency and tissue hypoxia as the causal mechanism. However, longitudinal studies of global and regional brain morphometry can help us give further insights on the pathophysiology of SCD in the brain.

Pub.: 26 May '17, Pinned: 02 Jul '17

White matter hyperintensity and stroke lesion segmentation and differentiation using convolutional neural networks

Abstract: The accurate assessment of White matter hyperintensities (WMH) burden is of crucial importance for epidemiological studies to determine association between WMHs, cognitive and clinical data. The manual delineation of WMHs is tedious, costly and time consuming. This is further complicated by the fact that other pathological features (i.e. stroke lesions) often also appear as hyperintense. Several automated methods aiming to tackle the challenges of WMH segmentation have been proposed, however cannot differentiate between WMH and strokes. Other methods, capable of distinguishing between different pathologies in brain MRI, are not designed with simultaneous WMH and stroke segmentation in mind. In this work we propose to use a convolutional neural network (CNN) that is able to segment hyperintensities and differentiate between WMHs and stroke lesions. Specifically, we aim to distinguish between WMH pathologies from those caused by stroke lesions due to either cortical, large or small subcortical infarcts. As far as we know, this is the first time such differentiation task has explicitly been proposed. The proposed fully convolutional CNN architecture, is comprised of an analysis path, that gradually learns low and high level features, followed by a synthesis path, that gradually combines and up-samples the low and high level features into a class likelihood semantic segmentation. Quantitatively, the proposed CNN architecture is shown to outperform other well established and state-of-the-art algorithms in terms of overlap with manual expert annotations. Clinically, the extracted WMH volumes were found to correlate better with the Fazekas visual rating score. Additionally, a comparison of the associations found between clinical risk-factors and the WMH volumes generated by the proposed method, were found to be in line with the associations found with the expert-annotated volumes.

Pub.: 03 Jun '17, Pinned: 02 Jul '17

A novel voxel-wise lesion segmentation technique on 3.0-T diffusion MRI of hyperacute focal cerebral ischemia at 1 h after permanent MCAO in rats.

Abstract: To assess hyperacute focal cerebral ischemia in rats on 3.0-Tesla diffusion-weighted imaging (DWI), we developed a novel voxel-wise lesion segmentation technique that overcomes intra- and inter-subject variation in apparent diffusion coefficient (ADC) distribution. Our novel technique involves the following: (1) intensity normalization including determination of the optimal type of region of interest (ROI) and its intra- and inter-subject validation, (2) verification of focal cerebral ischemic lesions at 1 h with gross and high-magnification light microscopy of hematoxylin-eosin (H&E) pathology, (3) voxel-wise segmentation on ADC with various thresholds, and (4) calculation of dice indices (DIs) to compare focal cerebral ischemic lesions at 1 h defined by ADC and matching H&E pathology. The best coefficient of variation was the mode of the left hemisphere after normalization using whole left hemispheric ROI, which showed lower intra- (2.54 ± 0.72%) and inter-subject (2.67 ± 0.70%) values than the original. Focal ischemic lesion at 1 h after middle cerebral artery occlusion (MCAO) was confirmed on both gross and microscopic H&E pathology. The 83 relative threshold of normalized ADC showed the highest mean DI (DI = 0.820 ± 0.075). We could evaluate hyperacute ischemic lesions at 1 h more reliably on 3-Tesla DWI in rat brains.

Pub.: 10 Jun '17, Pinned: 02 Jul '17