教学课程资源库 更多>>

关键词： Computer science   Artificial intelligence   Medical imaging  

摘要： Image segmentation is the most crucial step in image processing and analysis. It can divide an image into meaningfully descriptive components or pathological structures. The result of the image division helps analyze images and classify objects. Therefore, getting the most accurate segmented image is essential, especially in medical images. Segmentation methods can be divided into three categories: manual, semiautomatic, and automatic. Manual is the most general and straightforward approach. Manual segmentation is not only time-consuming but also is imprecise. However, automatic image segmentation techniques, such as thresholding and edge detection, are not accurate in the presence of artifacts like noise and texture. This research aims to show how to extract features and use traditional machine learning methods like a random forest to obtain the most accurate regions of interest in CT images. In addition, this study shows how to use a deep learning model to segment the wound area in raw pictures and then analyze the corresponding area in near-infrared images. This thesis first gives a brief review of current approaches to medical image segmentation and deep learning background. Furthermore, we describe different approaches to build a model for segmenting CT-Scan images and Wound Images. For the results, we achieve 97.4% accuracy in CT-image segmentation and 89.8% F1-Score For wound image segmentation.

关键词： Electrical engineering   Artificial intelligence   Computer science  

关键词： Artificial intelligence   Mathematics   Computer science   Applied mathematics  

摘要： We consider constrained policy optimization in Reinforcement Learning (RL), where the constraints are in form of marginals on state visitations and global action executions. Given these distributions, we formulate policy optimization as unbalanced optimal transport over the set of occupancy measures. We propose a general purpose RL objective based on Bregman divergence and optimize it using Dykstra's algorithm. The approach admits a large scale algorithm for when the state or action space is large and only samples from the marginals are available. We discuss applications of our approach and provide demonstrations to show the effectiveness of our algorithm.

关键词： Databases   Software   Genomics   Metabolites   Computer science   Genetics  

摘要： Background: Natural products (NP) from plants and microbes are a rich source for bioactive compounds essential for human life. A large part of agriculture, lifestyle and healthcare practice relies on metabolites derived from natural sources. To examine the biosynthetic potential of organisms and to guide NP discovery efforts, people increasingly utilise metabolomic, transcriptomic and genomic approaches. The co-location of metabolic genes in microbial genomes (termed Biosynthetic Gene Cluster or BGC) paves a way for an inexpensive and high throughput survey of natural products. While a focused scope analysis that targets a specific family of known compound chemistry was proven successful to optimize the compound’s utility, a truly global overview which will open our eyes to the actual extent of novel chemistries lies unexplored in nature is still hampered by the limitation of (high quality) data, techniques and bioinformatic tools that are currently available. Results: A computational prediction tool PlantiSMASH was made to enable the exploration of putative plant BGCs, which combines genomic and transcriptomic data to give insights into plant secondary metabolism and evolution. To support large scale annotation and analysis of BGCs, a reference database of known BGC (MIBiG) was markedly improved both in quality and quantity, providing a 73% data increase over its initial release version. A large-scale study of BGC and Gene Cluster Family (GCF) diversity across taxa was done, enabled by the development of a novel bioinformatics tool which can process 1.2 million BGCs within ten days of computing time. Finally, an online database of more than 25,000 GCFs was released for the first time, giving means to the community to do crowdsourced curation, which in turn would come back and be useful in the annotation and discovery of putative or novel BGCs of their own. Conclusions: The works presented in this thesis provide the foundation for a global diversity-informed NP disc

关键词： Artificial intelligence   Oncology   Computer science   Information technology   Health care management  

摘要： Purpose: Traditional pre-treatment patient specific QA is known for its high workload for physicist, ineffectiveness at identifying clinically relevant dosimetric uncertainties of treatment plans, and incompatibility with on-line adaptive radiotherapy. Our purpose is to develop a trajectory file based PSQA procedure that allows for a virtual pre-treatment QA that can effectively evaluate the performance and robustness of a treatment plan via a DVH based analysis and can be carried out with online adaptive radiotherapy. For this purpose, we have developed a machine learning model that can predict discrepancy in machine parameters between delivery and treatment plan on a Varian TrueBeam linear accelerator. Methods: Trajectory log files and DICOM-RT plan files of 30 IMRT plans and 75 VMAT plans from four Varian TrueBeam linear accelerators were collected for analysis. The discrepancy in machine parameters is divided into “conversion error” (from converting DICOM-RT to deliverable machine trajectory) and “delivery error” (difference in machine parameters recorded in trajectory files). Correlation matrices were obtained to determine the linear correlation between actual discrepancy and mechanical parameters, such as MLC velocity, MLC acceleration, control point, dose rate, gravity vector, gantry velocity, and gantry acceleration. Multiple regression algorithms were used to develop machine learning models to predict the total discrepancy in machine parameters and its components based on mechanical parameters. The fully trained models were validated with an independent validation dataset and treatment plans constructed with varying degrees of complexity approaching the limitations of the linear accelerator. Results: For both IMRT and VMAT, the RMS of conversion error (0.1528 mm) was 4 times greater than the RMS of delivery error (0.0367 mm). A high correlation existed between MLC velocity and both components of discrepancies for IMRT (R2 ∈ [0.61, 0.75]) and VMAT [0.75, 0.8

关键词： Computer science   Mathematics  

摘要： We consider the problem of multi-agent optimization in the presence of Byzantine workers. Specifically, we focus on the case of parallelized stochastic gradient descent utilizing n workers, up to f < n/2 of which are Byzantine with the goal of corrupting optimization. We propose four schemes based off of a known fault-tolerant algorithm, ReactiveRedundantSGD, implement all five techniques, and use them in gradient descent optimization experiments. Through these experiments, we verify the exact fault-tolerance of the original algorithm and show that its speed can be consistently improved by three schemes' mechanisms: adaptive redundancy, rollback recovery, and lightweight gradient filtering.

关键词： Radio frequency identification   Internet of Things   Global positioning systems--GPS   Random access memory   Computer science   Electrical engineering   Engineering  

关键词： Computer science   Artificial intelligence   Health sciences  

摘要： The primary goal of Machine learning (ML) models in the prediction of medical conditions is to accurately predict (classify) the occurrence of a disease, or therapy. Many ML models, traditional and deep, have been utilized for the prediction of disease diagnosis, or prediction of the most optimal therapeutic approach. Almost all categories of medical conditions were subject to ML analysis. When creating predictive ML algorithms in medicine, it is pivotal to consider what problems are intended to be solved and how much and what types of training data are available. For challenging prediction (classification) problems, the understanding of disease pathogenesis makes the selection of an adequate ML model and accurate prediction more likely. The hypothesis of the research was to demonstrate that the optimal and adequate selection of model inputs as well as the selection and design of adequate ML methods improves the prediction accuracy of occurrence of diseases and their outcomes. The effectiveness and accuracy of created deep learning and traditional methods have been analyzed and compared. The impact of different medical conditions and different medical domains on optimal selection and performance of ML models was also studied. The effectiveness of advanced ML models was tested on four different diseases: Alzheimer’s disease (AD), Diabetes Mellitus type 2 (DM2), Influenza, and Colorectal cancer (CRC). The objective of the first part of the thesis (AD study) was to determine could prediction of AD from Electronic medical records (EMR) data alone be significantly improved by applying domain knowledge in positive dataset selection rather than setting naïve filters. Selected Clinically Relevant Positive (SCRP) datasets were used as inputs to a Long-Short-Term Memory (LSTM) Recurrent Neural Network (RNN) deep learning model to predict will the patient develop AD. The LSTM RNN method performed significantly better when learning from the SCRP dataset than when datasets were

关键词： Sparsity   Gene expression   Bar codes   Optimization   Correlation analysis   Algorithms   Clustering   Geometry   Artificial intelligence   Bioinformatics   Computer science   Genetics   Mathematics  

摘要： Convex and nonconvex optimization problems are frequently encountered in many research fields. In this thesis, we study several models with different model structure assumptions under the smooth optimization framework. In many applications, data are two-dimensional and represented as data matrices where each row represents an observation, whereas each column represents a feature of each observation. In Chapter 2, we propose a general method that can simultaneously exploit the correlations among both the rows and columns of the data matrix to efficiently compute the affinities in a missing data setting. We introduce a multi-scale metric that captures the geometry of the complete data matrix and represents the row and column similarities under an optimization framework. The multi-scale metric can serve as affinity weights in many applications even when data are incomplete. We present experimental results to illustrate the effectiveness of the method on common machine learning problems and show the metric is general and can be easily adapted to other applications. In Chapter 3, we consider the problem of low-rank plus sparse matrix decomposition. It is proved that this problem can be solved via heuristically solving a convex relaxation problem where the nuclear norm and l1 norm are utilized to induce low-rank and sparsity. Many algorithms exploit first-order methods that involve the proximal operator of the nuclear norm and require conducting the singular value decomposition (SVD). In each iteration, the dominant cost is computing an SVD of the same size as the input data, which has a superlinear *** need to compute a sequence of full or partial SVD severely limits the applicability of these algorithms to large-scale problems. To remedy this, we introduce the low-rank bi-factorization structure and characterize the nuclear norm in terms of the Frobenius norm. With this low-rank bi-factorization, we develop a more scalable algorithm that avoids the computational bu

关键词： Computer science   Genetics   Statistics   Schizophrenia   Human genome--Research   Schizophrenia--Etiology  

摘要： Over the past 20 years, Genome-Wide Association Studies (GWAS) have identified thousands of variants in the genome linked to genetic diseases. However, these associations often reveal little about underlying genetic etiology, which for many phenotypes is thought to be highly heterogeneous. This work investigates statistical methods to move beyond conventional GWAS methods to both improve estimation of associations and to extract additional etiological insights from known associations, with a focus on schizophrenia. This thesis addresses the above aim through three primary topics: First, we describe ***, a web platform to crowdsource the collection of genomic data with user consent and active participation, thereby rapidly increasing sample sizes and power required for GWAS. Second, we describe methods to characterize the latent genomic contributors to heterogeneity in GWAS phenotypes. We develop a Z-score test to detect heterogeneity using correlations between variants among affected individuals, and we develop a contrastive tensor decomposition to explicitly characterize subtype-specific SNP effects independently of confounding heterogeneity such as ancestry. Using these methods we provide evidence of significant heterogeneity in GWAS cohorts for schizophrenia. Lastly, a major avenue of investigation beyond GWAS is identifying the genes through which associated SNPs mechanistically affect the presentation of phenotypes. We develop a method to improve estimation of expression quantitative trait loci by joint inference over gene expression reference data and GWAS data, incorporating insights from the liability threshold model. These methods will advance ongoing efforts to explain the complex etiology of genetic diseases as well as improve the accuracy of disease prediction models based on these insights.