Hai Lin

Deep learning for accurate diagnosis of liver tumor based on magnetic resonance imaging and clinical data

Fri, 16 Sep 2022 13:30:29 +0800

Background

Early-stage diagnosis and treatment can improve survival rates of liver cancer patients. Dynamic contrast-enhanced MRI provides the most comprehensive information for differential diagnosis of liver tumors. However, MRI diagnosis is affected by subjective experience, so deep learning may supply a new diagnostic strategy. We used convolutional neural networks (CNNs) to develop a deep learning system (DLS) to classify liver tumors based on enhanced MR images, unenhanced MR images, and clinical data including text and laboratory test results.

Methods

Using data from 1,210 patients with liver tumors (N = 31,608 images), we trained CNNs to get seven-way classifiers, binary classifiers, and three-way malignancy-classifiers (Model A-Model G). Models were validated in an external independent extended cohort of 201 patients (N = 6,816 images). The area under receiver operating characteristic (ROC) curve (AUC) were compared across different models. We also compared the sensitivity and specificity of models with the performance of three experienced radiologists.

Results

Deep learning achieves a performance on par with three experienced radiologists on classifying liver tumors in seven categories. Using only unenhanced images, CNN performs well in distinguishing malignant from benign liver tumors (AUC, 0.946; 95% CI 0.914–0.979 vs. 0.951; 0.919–0.982, P = 0.664). New CNN combining unenhanced images with clinical data greatly improved the performance of classifying malignancies as hepatocellular carcinoma (AUC, 0.985; 95% CI 0.960–1.000), metastatic tumors (0.998; 0.989–1.000), and other primary malignancies (0.963; 0.896–1.000), and the agreement with pathology was 91.9%.These models mined diagnostic information in unenhanced images and clinical data by deep-neural-network, which were different to previous methods that utilized enhanced images. The sensitivity and specificity of almost every category in these models reached the same high level compared to three experienced radiologists.

Conclusion

Trained with data in various acquisition conditions, DLS that integrated these models could be used as an accurate and time-saving assisted-diagnostic strategy for liver tumors in clinical settings, even in the absence of contrast agents. DLS therefore has the potential to avoid contrast-related side effects and reduce economic costs associated with current standard MRI inspection practices for liver tumor patients.

ALA-Net: Adaptive Lesion-Aware Attention Network for 3D Colorectal Tumor Segmentation

Fri, 16 Sep 2022 13:29:56 +0800

Accurate and reliable segmentation of colorectal tumors and surrounding colorectal tissues on 3D magnetic resonance images has critical importance in preoperative prediction, staging, and radiotherapy. Previous works simply combine multilevel features without aggregating representative semantic information and without compensating for the loss of spatial information caused by down-sampling. Therefore, they are vulnerable to noise from complex backgrounds and suffer from misclassification and target incompleteness-related failures. In this paper, we address these limitations with a novel adaptive lesion-aware attention network (ALA-Net) which explicitly integrates useful contextual information with spatial details and captures richer feature dependencies based on 3D attention mechanisms. The model comprises two parallel encoding paths. One of these is designed to explore global contextual features and enlarge the receptive field using a recurrent strategy. The other captures sharper object boundaries and the details of small objects that are lost in repeated down-sampling layers. Our lesion-aware attention module adaptively captures long-range semantic dependencies and highlights the most discriminative features, improving semantic consistency and completeness. Furthermore, we introduce a prediction aggregation module to combine multiscale feature maps and to further filter out irrelevant information for precise voxel-wise prediction. Experimental results show that ALA-Net outperforms state-of-the-art methods and inherently generalizes well to other 3D medical images segmentation tasks, providing multiple benefits in terms of target completeness, reduction of false positives, and accurate detection of ambiguous lesion regions.

DeepNFT: Towards Precise Neurofibrillary Tangle Detection via Improving Multi-scale Feature Fusion and Adversary

Fri, 16 Sep 2022 13:29:31 +0800

Detecting neurofibrillary tangles is an important procedure in the assessment of the intensity and distribution pattern of hippocampal tau pathology, which are the principal clinical phenotypes associated with Alzheimer’s disease. Existing deep learning based detectors still face a critical obstacle: the difficulty in detecting extremely small objects in high resolution images. In this paper, we propose a deep learning framework, named DeepNFT, which combines the multilevel feature aggregation pyramid network (MFAPN) and the adversarial feature generation module (AFGM) to acquire precise detection results with significantly reduced false positives. To prove its universality and robustness, DeepNFT has been validated on two datasets. Experiments show the significant performance gain of our proposed approach over state-of-the-art detectors. Ablation study shows our network components improve the performance of various backbones and detectors.

A GPU-based Radio Wave Propagation Prediction with Progressive Processing on Point Cloud

Thu, 15 Sep 2022 16:54:40 +0800

Point cloud data (PCD) can record a high-resolution model of the environment much more efficiently than the conventional geometrical mesh. This feature makes PCD suitable for the radio channel prediction of a new environment, especially for the millimeter-wave (mmWave). Since the multiple propagations and visibility computation using PCD are time consuming, progressive processing on PCD is introduced to make the prediction compatible with GPU-based frameworks such as OptiX and CUDA. Compared with the surface reconstruction from PCD, progressive data from the processing reserves more domain features such as diffraction wedge and the label of planar or nonplanar points. The numerical result of an outdoor environment shows that the proposed method has a close agreement with measurement, as well as two impressive speedups, 49.8 for the ray tracing and field calculation and 18.8 for the total prediction, compared with the original method using PCD.

Visual exploration of software evolution via topic modeling

Thu, 15 Sep 2022 16:49:56 +0800

For various reasons, such as new requirements, architecture refactoring, and bug fixing, software projects often evolve to yield better quality and performance. All changes produced during the development process are reflected in the source code, which provides an opportunity to explore software evolution. In this paper, we propose a visual analytics system to support evolution analysis based on topic modeling. We focus on three aspects: (1) when significant changes to source code occur, (2) how software features evolve, and (3) why software evolution occurs. Each source file is regarded as a document and represented by its topic vector. The files of each two successive versions are classified into four types to quantify version differences, and the number of topic-associated files is denoted as the topic assignment to characterize feature evolution. Finally, we inspect the causes of software evolution through the visual comparison between versions. Two case studies on JavaScript libraries demonstrate the usefulness and effectiveness of our system.

Graph convolutional network-based semi-supervisedfeature classiﬁcation of volumes

Thu, 15 Sep 2022 16:01:45 +0800

Feature classiﬁcation has always been one of the research hotspots in scientiﬁc visualization.However, conventional interactive feature classiﬁcation methods rely on prior knowledge and typicallyrequire trial and error, whereas feature classiﬁcation based on data mining is generally based on localfeatures; therefore, obtaining good results with traditional methods is difﬁcult. In this paper, we ﬁrst map avolume to the super-voxel graph using a 3D extension of the simple linear iterative clustering algorithm andthen construct a graph convolutional neural network to implement node classiﬁcation in a semi-supervisedway, i.e., a small number of user-labeled super-voxels. We transform the feature classiﬁcation of a volumeinto the classiﬁcation task of nodes of a super-voxel graph, which is a novel approach and broadens theapplication scope of graph neural netwo rk to volumes. Experiments on different volumes have demonstratedthe strong learning ability and reasoning ability of the proposed method.

Exploring Evolution of Dynamic Networks via Diachronic Node Embeddings

Thu, 15 Sep 2022 15:54:01 +0800

Dynamic networks evolve with their structures changing over time. It is still a challenging problem to efficiently explore the evolution of dynamic networks in terms of both their structural and temporal properties. In this paper, we propose a visual analytics methodology to interactively explore the temporal evolution of dynamic networks in the context of their structure. A novel diachronic node embedding method is first proposed to learn latent representations of the structural and temporal features of nodes in a vector space. Diachronic node embeddings are then used to discover communities with similar structural proximity and temporal evolution patterns. A visual analytics system is designed to enable users to visually explore the evolutions of nodes, communities, and the network as a whole in terms of their structural and temporal properties. We evaluate the effectiveness of our method using artificial and real-world dynamic networks and comparisons with previous methods.

Solving the Scattering of Combined Conductor-Dielectric Bodies Based on EPA

Thu, 15 Sep 2022 15:42:32 +0800

This work introduces a new method for analyzing the scattering from combined conductor-dielectric bodies based on equivalent principle algorithm (EPA). The object is decomposed into several different parts. In each part, it can be an inhomogeneous dielectric or a perfect electric conductor (PEC) of arbitrary shape. Equivalent surfaces (ES) are built for the interactions between each parts. And the current translation from the object to ES can be accelerated by the transformed multilevel fast multipole algorithm (MLFMA). Numerical result demonstrates the efficiency and accuracy of the proposed method.

Electromagnetic Characteristics Optimization Based on Shape Deformation

Thu, 15 Sep 2022 15:42:18 +0800

A feasible approach to optimize the characteristics of electromagnetic (EM) objects through shape deformation is proposed. We provide a complete and detailed procedure of the optimization and investigate the shape deformation formulations for some typical EM objects. Numerical results demonstrate the validity and performance of the optimization approach.

Deep Learning with Attention Mechanism for Electromagnetic Inverse Scattering

Fri, 15 Jul 2022 17:04:51 +0800

The inverse scattering problems (ISPs) aim for reconstructing the unknown targets from the measured scattered fields. Due to its highly nonlinear and ill-posed properties, solving ISPs is a challenging task. In this paper, a new deep learning technique is investigated to improve the solution of electromagnetic inverse scattering problems (ISPs). The basic idea of this technique is to introduce the attention mechanism to U-Net. By doing this, the deep learning model can automatically learn to focus on target areas of the measured scattered field matrix. In this way, the proposed technique can offer higher accuracy and faster convergence compared with the traditional deep learning networks. Numerical results validate the efficacy of the proposed technique.

Voxel2vec

Tue, 12 Jul 2022 16:33:15 +0800

Relationships in scientific data are intricate and complex, such as the numerical and spatial distribution relations of features in univariate data, the scalar-value combinations’ relations in multivariate data, and the association of volumes in time-varying and ensemble data. This paper presents voxel2vec, a novel unsupervised representation learning model, to learn distributed representations of scalar values in a low-dimensional vector space. The basic assumption is that if two scalar values/scalar-value combinations have similar contexts, they usually have high similarity in terms of features. By representing scalar values as symbols, voxel2vec learns the similarity of scalar values in the context of spatial distribution and then we can explore the overall association between volumes by transfer prediction. We demonstrate the usefulness and effectiveness of voxel2vec by comparing it with the isosurface similarity map of univariate data and applying the learned distributed representations to feature classification for multivariate data and association analysis for time-varying and ensemble data.

Enhanced solution to the surface-volume-surface EFIE for arbitrary metal-dielectric composite objects

Sat, 01 Jan 2022 00:00:00 +0000

Graph convolutional network-based semi-supervised feature classification of volumes

Sat, 01 Jan 2022 00:00:00 +0000

SatFormer: Saliency-Guided Abnormality-Aware Transformer for Retinal Disease Classification in Fundus Image

Sat, 01 Jan 2022 00:00:00 +0000

ScalarGCN: scalar-value association analysis of volumes based on graph convolutional network

Sat, 01 Jan 2022 00:00:00 +0000

voxel2vec: A Natural Language Processing Approach to Learning Distributed Representations for Scientific Data

Sat, 01 Jan 2022 00:00:00 +0000

A GPU-based radio wave propagation prediction with progressive processing on point cloud

Fri, 01 Jan 2021 00:00:00 +0000

Accuracy improvement of the algebraic fast methods for the volume-surface integral equation

Fri, 01 Jan 2021 00:00:00 +0000

ALA-Net: Adaptive Lesion-Aware Attention Network for 3D Colorectal Tumor Segmentation

Fri, 01 Jan 2021 00:00:00 +0000

An automatic tooth reconstruction method based on multimodal data

Fri, 01 Jan 2021 00:00:00 +0000

DeepNFT: Towards Precise Neurofibrillary Tangle Detection via Improving Multi-scale Feature Fusion and Adversary

Fri, 01 Jan 2021 00:00:00 +0000

ELF near-field propagation of a vertical electric dipole due to lightning discharges

Fri, 01 Jan 2021 00:00:00 +0000

IsoExplorer: an isosurface-driven framework for 3D shape analysis of biomedical volume data

Fri, 01 Jan 2021 00:00:00 +0000

SPAN: Subgraph Prediction Attention Network for Dynamic Graphs

Fri, 01 Jan 2021 00:00:00 +0000

Visual exploration of dependency graph in source code via embedding-based similarity

Fri, 01 Jan 2021 00:00:00 +0000

Visual exploration of software evolution via topic modeling

Fri, 01 Jan 2021 00:00:00 +0000

CephaNN: A Multi-Head Attention Network for Cephalometric Landmark Detection

Wed, 01 Jan 2020 00:00:00 +0000

Exploring Evolution of Dynamic Networks via Diachronic Node Embeddings

Wed, 01 Jan 2020 00:00:00 +0000

Iterative MLFMA-MADBT technique for analysis of antenna mounted on large platforms

Wed, 01 Jan 2020 00:00:00 +0000

Learning Dynamic Context Graph Embedding

Wed, 01 Jan 2020 00:00:00 +0000

Solving the scattering of combined conductor-dielectric bodies based on EPA

Wed, 01 Jan 2020 00:00:00 +0000

Time-varying volume visualization: a survey

Wed, 01 Jan 2020 00:00:00 +0000

Visual analytics of urban transportation from a bike-sharing and taxi perspective

Wed, 01 Jan 2020 00:00:00 +0000

An automatic algorithm for repairing dental models based on contours

Tue, 01 Jan 2019 00:00:00 +0000

An Interactive Visual Analytics System for Incremental Classification Based on Semi-supervised Topic Modeling

Tue, 01 Jan 2019 00:00:00 +0000

CephaNet: An Improved Faster R-CNN for Cephalometric Landmark Detection

Tue, 01 Jan 2019 00:00:00 +0000

Dynamic Network Embeddings for Network Evolution Analysis

Tue, 01 Jan 2019 00:00:00 +0000

Electromagnetic characteristics optimization based on shape deformation

Tue, 01 Jan 2019 00:00:00 +0000

FeatureFlow: exploring feature evolution for time-varying volume data

Tue, 01 Jan 2019 00:00:00 +0000

Multiple GPU parallel strategy of beam tracing for propagation prediction in large-scale scenes

Tue, 01 Jan 2019 00:00:00 +0000

Multivariate spatial data visualization: a survey

Tue, 01 Jan 2019 00:00:00 +0000

Visual analytics of taxi trajectory data via topical sub-trajectories

Tue, 01 Jan 2019 00:00:00 +0000

Voxer - a platform for creating, customizing, and sharing scientific visualizations

Tue, 01 Jan 2019 00:00:00 +0000

A co-analysis framework for exploring multivariate scientific data