Tengfei Li

Radiation dose reduction in four-dimensional computed tomography.

Four-dimensional (4D) CT is useful in many clinical situations, where detailed abdominal and thoracic imaging is needed over the course of the respiratory cycle. However, it usually delivers a larger radiation dose than the standard three-dimensional (3D) CT, since multiple scans at each couch position are required in order to provide the temporal information. Our purpose in this work is to develop a method to perform 4D CT scans at relatively low current, hence reducing the radiation exposure of the patients. To deal with the increased statistical noise caused by the low current, we proposed a novel 4D penalized weighted least square (4D-PWLS) smoothing method, which can incorporate both spatial and phase information. The 4D images at different phases were registered to the same phase via a deformable model, thereby, a regularization term combining temporal and spatial neighbors can be designed for the 4D-PWLS objective function. The proposed method was tested with phantom experiments and a patient study, and superior noise suppression and resolution preservation were observed. A quantitative evaluation of the benefit of the proposed method to 4D radiotherapy and 4D PET/CT imaging are under investigation.

Influence of population stratification on population-based marker-disease association analysis

Population‐based genetic association analysis may suffer from the failure to control for confounders such as population stratification (PS). There has been extensive study on the influence of PS on candidate gene‐disease association analysis, but much less attention has been paid to its influence on marker‐disease association analysis. In this paper, we focus on the Pearson χ2 test and the trend test for marker‐disease association analysis. The mean and variance of the test statistics are derived under presence of PS, so that the power and inflated type I error rate can be evaluated. It is shown that the bias and the variance distortion are not zero in the presence of both PS and penetrance heterogeneity (PH). Unlike candidate gene‐disease association analysis, when PS is present, the bias is not zero no matter whether PH is present or not. This work generalises the published results, where only the fully recessive penetrance model is considered and only the bias is calculated. It is shown that candidate gene‐disease association analysis can be treated as a special case of marker‐disease association analysis. Consequently, our results extend previous studies on candidate gene‐disease association analysis. A simulation study confirms the theoretical findings.

Genome-wide mediation analysis of psychiatric and cognitive traits through imaging phenotypes

Heritability is well documented for psychiatric disorders and cognitive abilities which are, however, complex, involving both genetic and environmental factors. Hence, it remains challenging to discover which and how genetic variations contribute to such complex traits. In this article, they propose to use mediation analysis to bridge this gap, where neuroimaging phenotypes were utilized as intermediate variables. The Philadelphia Neurodevelopmental Cohort was investigated using genome‐wide association studies (GWAS) and mediation analyses. Specifically, 951 participants were included with age ranging from 8 to 21 years. Two hundred and four neuroimaging measures were extracted from structural magnetic resonance imaging scans. GWAS were conducted for each measure to evaluate the SNP‐based heritability. Furthermore, mediation analyses were employed to understand the mechanisms in which genetic variants have influence on pathological behaviors implicitly through neuroimaging phenotypes, and identified SNPs that would not be detected otherwise. Our analyses found that rs10494561, located in the intron region within NMNAT2, was associated with the severity of the prodromal symptoms of psychosis implicitly, mediated through the volume of the left hemisphere of the superior frontal region ( P=2.38×10−8 ). The gene NMNAT2 is known to be associated with brainstem degeneration, and produce cytoplasmic enzyme which is mainly expressed in the brain. Another SNP rs2285351 was found in the intron region of gene IFT122 which may be potentially associated with human spatial orientation ability through the area of the left hemisphere of the isthmuscingulate region ( P=3.70×10−8 ). Hum Brain Mapp 38:4088–4097, 2017.

Radiomic analysis in prediction of Human Papilloma Virus status

Human Papilloma Virus (HPV) has been associated with oropharyngeal cancer prognosis. Traditionally the HPV status is tested through invasive lab test. Recently, the rapid development of statistical image analysis techniques has enabled precise quantitative analysis of medical images. The quantitative analysis of Computed Tomography (CT) provides a non-invasive way to assess HPV status for oropharynx cancer patients. We designed a statistical radiomics approach analyzing CT images to predict HPV status. Various radiomics features were extracted from CT scans, and analyzed using statistical feature selection and prediction methods. Our approach ranked the highest in the 2016 Medical Image Computing and Computer Assisted Intervention (MICCAI) grand challenge: Oropharynx Cancer (OPC) Radiomics Challenge, Human Papilloma Virus (HPV) Status Prediction. Further analysis on the most relevant radiomic features distinguishing HPV positive and negative subjects suggested that HPV positive patients usually have smaller and simpler tumors.

A label-fusion-aided convolutional neural network for isointense infant brain tissue segmentation

The extremely low tissue contrast in white matter during an infants isointense stage (6–8 months) of brain development presents major difficulty when segmenting brain image regions for analysis. We sought to develop a label-fusion-aided deep-learning approach for automatically segmenting isointense infant brain images into white matter, gray matter and cerebrospinal fluid using T1- and T2-weighted magnetic resonance images. A key idea of our approach is to apply the fully convolutional neural network (FCNN) to individual brain regions determined by a traditional registration-based segmentation method instead of training a single model for the whole brain. This provides more refined segmentation results by capturing more region-specific features. We show that this method outperforms traditional joint label fusion and FCNN-only methods in terms of Dice coefficients using the dataset from iSEG MICCAI Grand Challenge 2017.

Statistical disease mapping for heterogeneous neuroimaging studies

Most cancers and neuro-related diseases (e.g., autism and stroke) display significant phenotypic and genetic heterogeneity. Characterizing such heterogeneity could transform our understanding of the etiology of these conditions and inspire new approaches to urgently needed preventions, diagnoses, and treatments. However, existing statistical methods face major challenges in delineating such heterogeneity at both group and individual levels. The aim of this paper is to propose a novel statistical disease mapping (SDM) framework to address some of these challenges. We develop an efficient estimation method to estimate unknown parameters in SDM and individual and group disease maps. Both simulation studies and real data analysis on the ADNI PET dataset indicate that our SDM can not only effectively detect diseased regions in each patient, but also provide a group disease map analysis of Alzheimer (AD) subgroups.

Large-scale neuroimaging and genetic study reveals genetic architecture of brain white matter microstructure

Microstructural changes of white matter (WM) tracts are known to be associated with various neuropsychiatric disorders/diseases. Heritability of structural changes of WM tracts has been examined using diffusion tensor imaging (DTI) in family-based studies for different age groups. The availability of genetic and DTI data from recent large population-based studies offers opportunity to further improve our understanding of genetic contributions. Here, we analyzed the genetic architecture of WM tracts using DTI and single-nucleotide polymorphism (SNP) data of unrelated individuals in the UK Biobank (n ~ 8000). The DTI parameters were generated using the ENIGMA-DTI pipeline. We found that DTI parameters are substantially heritable on most WM tracts. We observed a highly polygenic or omnigenic architecture of genetic influence across the genome as well as the enrichment of SNPs in active chromatin regions. Our bivariate analyses showed strong genetic correlations for several pairs of WM tracts as well as pairs of DTI parameters. We performed voxel-based analysis to illustrate the pattern of genetic effects on selected parts of the tract-based spatial statistics skeleton. Comparing the estimates from the UK Biobank to those from small population-based studies, we illustrated that sufficiently large sample size is essential for genetic architecture discovery in imaging genetics. We confirmed this finding with a simulation study.Background Individual variations of white matter (WM) tracts are known to be associated with various cognitive and neuropsychiatric traits. Diffusion tensor imaging (DTI) and genome-wide single-nucleotide polymorphism (SNP) data from 17,706 UK Biobank participants offer opportunity to identify novel genetic variants of WM tracts and explore the genetic overlap with other brain-related complex traits. Method We analyzed the genetic architecture of 110 tract-based DTI parameters, carried out genome-wide association studies (GWAS) and performed post-GWAS analyses, including association lookups, gene-based association analysis, functional gene mapping, and genetic correlation estimation. Results DTI parameters are substantially heritable for all WM tracts (mean heritability 48.7%). We observed a highly polygenic architecture of genetic influence across the genome (p-value=1.67*10−05) as well as the enrichment of genetic effects for active SNPs annotated by central nervous system cells (p-value=8.95*10−12). GWAS identified 213 independent significant SNPs associated with 90 DTI parameters (696 SNP-level and 205 locus-level associations; p-value<4.5*10−10, adjusted for testing multiple phenotypes). Gene-based association study prioritized 112 significant genes, most of which are novel. More importantly, association lookups found that many of the novel SNPs and genes of DTI parameters have previously been implicated with cognitive and mental health traits. Conclusions The present study identifies many new genetic variants at SNP, locus and gene levels for integrity of brain WM tracts and provides the overview of pleiotropy with cognitive and mental health traits.

Machine Learning Applications in Head and Neck Radiation Oncology: Lessons From Open-Source Radiomics Challenges

Radiomics leverages existing image datasets to provide non-visible data extraction via image post-processing, with the aim of identifying prognostic, and predictive imaging features at a sub-region of interest level. However, the application of radiomics is hampered by several challenges such as lack of image acquisition/analysis method standardization, impeding generalizability. As of yet, radiomics remains intriguing, but not clinically validated. We aimed to test the feasibility of a non-custom-constructed platform for disseminating existing large, standardized databases across institutions for promoting radiomics studies. Hence, University of Texas MD Anderson Cancer Center organized two public radiomics challenges in head and neck radiation oncology domain. This was done in conjunction with MICCAI 2016 satellite symposium using Kaggle-in-Class, a machine-learning and predictive analytics platform. We drew on clinical data matched to radiomics data derived from diagnostic contrast-enhanced computed tomography (CECT) images in a dataset of 315 patients with oropharyngeal cancer. Contestants were tasked to develop models for (i) classifying patients according to their human papillomavirus status, or (ii) predicting local tumor recurrence, following radiotherapy. Data were split into training, and test sets. Seventeen teams from various professional domains participated in one or both of the challenges. This review paper was based on the contestants feedback; provided by 8 contestants only (47%). Six contestants (75%) incorporated extracted radiomics features into their predictive model building, either alone (n = 5; 62.5%), as was the case with the winner of the “HPV” challenge, or in conjunction with matched clinical attributes (n = 2; 25%). Only 23% of contestants, notably, including the winner of the “local recurrence” challenge, built their model relying solely on clinical data. In addition to the value of the integration of machine learning into clinical decision-making, our experience sheds light on challenges in sharing and directing existing datasets toward clinical applications of radiomics, including hyper-dimensionality of the clinical/imaging data attributes. Our experience may help guide researchers to create a framework for sharing and reuse of already published data that we believe will ultimately accelerate the pace of clinical applications of radiomics; both in challenge or clinical settings.

Heritability of Regional Brain Volumes in Large-Scale Neuroimaging and Genetic Studies

Brain genetics is an active research area. The degree to which genetic variants impact variations in brain structure and function remains largely unknown. We examined the heritability of regional brain volumes (P ~ 100) captured by single-nucleotide polymorphisms (SNPs) in UK Biobank (n ~ 9000). We found that regional brain volumes are highly heritable in this study population and common genetic variants can explain up to 80% of their variabilities (median heritability 34.8%). We observed omnigenic impact across the genome and examined the enrichment of SNPs in active chromatin regions. Principal components derived from regional volume data are also highly heritable, but the amount of variance in brain volume explained by the component did not seem to be related to its heritability. Heritability estimates vary substantially across large-scale functional networks, exhibit a symmetric pattern across left and right hemispheres, and are consistent in females and males (correlation = 0.638). We repeated the main analysis in Alzheimers Disease Neuroimaging Initiative (n ~ 1100), Philadelphia Neurodevelopmental Cohort (n ~ 600), and Pediatric Imaging, Neurocognition, and Genetics (n ~ 500) datasets, which demonstrated that more stable estimates can be obtained from the UK Biobank.

TPCNN: Two-phase patch-based convolutional neural network for automatic brain tumor segmentation and survival prediction

The aim of this paper is to integrate some advanced statistical methods with modern deep learning methods for tumor segmentation and survival time prediction in the BraTS 2017 challenge. The goals of the BraTS 2017 challenge are to utilize multi-institutional pre-operative MRI scans to segment out different tumor subregions and then to use tumor information to predict patient’s overall survival. We build a two-phase patch-based convolutional neural network (TPCNN) model to classify all the pixels in the brain and further refine the segmentation results by using XGBoost and a post-processing procedure. The segmentation results are then used to extract various informative radiomic features for prediction of the survival time by using the XGBoost method.