Comparing the accuracy of several network-based COVID-19 prediction algorithms.

Researchers from various scientific disciplines have attempted to forecast the spread of coronavirus disease 2019 (COVID-19). The proposed epidemic prediction methods range from basic curve fitting methods and traffic interaction models to machine-learning approaches. If we combine all these approaches, we obtain the Network Inference-based Prediction Algorithm (NIPA). In this paper, we analyse a diverse set of COVID-19 forecast algorithms, including several modifications of NIPA. Among the algorithms that we evaluated, the original NIPA performed best at forecasting the spread of COVID-19 in Hubei, China and in the Netherlands. In particular, we show that network-based forecasting is superior to any other forecasting algorithm.

Tongue Tumor Detection in Hyperspectral Images Using Deep Learning Semantic Segmentation.

OBJECTIVE: The utilization of hyperspectral imaging (HSI) in real-time tumor segmentation during a surgery have recently received much attention, but it remains a very challenging task. METHODS: In this work, we propose semantic segmentation methods, and compare them with other relevant deep learning algorithms for tongue tumor segmentation. To the best of our knowledge, this is the first work using deep learning semantic segmentation for tumor detection in HSI data using channel selection, and accounting for more spatial tissue context, and global comparison between the prediction map, and the annotation per sample. Results, and Conclusion: On a clinical data set with tongue squamous cell carcinoma, our best method obtains very strong results of average dice coefficient, and area under the ROC-curve of [Formula: see text], and [Formula: see text], respectively on the original spatial image size. The results show that a very good performance can be achieved even with a limited amount of data. We demonstrate that important information regarding tumor decision is encoded in various channels, but some channel selection, and filtering is beneficial over the full spectra. Moreover, we use both visual (VIS), and near-infrared (NIR) spectrum, rather than commonly used only VIS spectrum; although VIS spectrum is generally of higher significance, we demonstrate NIR spectrum is crucial for tumor capturing in some cases. SIGNIFICANCE: The HSI technology augmented with accurate deep learning algorithms has a huge potential to be a promising alternative to digital pathology or a doctors' supportive tool in real-time surgeries.

Towards radiologist-level cancer risk assessment in CT lung screening using deep learning.

PURPOSE: Lung cancer is the leading cause of cancer mortality in the US, responsible for more deaths than breast, prostate, colon and pancreas cancer combined and large population studies have indicated that low-dose computed tomography (CT) screening of the chest can significantly reduce this death rate. Recently, the usefulness of Deep Learning (DL) models for lung cancer risk assessment has been demonstrated. However, in many cases model performances are evaluated on small/medium size test sets, thus not providing strong model generalization and stability guarantees which are necessary for clinical adoption. In this work, our goal is to contribute towards clinical adoption by investigating a deep learning framework on larger and heterogeneous datasets while also comparing to state-of-the-art models. METHODS: Three low-dose CT lung cancer screening datasets were used: National Lung Screening Trial (NLST, n = 3410), Lahey Hospital and Medical Center (LHMC, n = 3154) data, Kaggle competition data (from both stages, n = 1397 + 505) and the University of Chicago data (UCM, a subset of NLST, annotated by radiologists, n = 132). At the first stage, our framework employs a nodule detector; while in the second stage, we use both the image context around the nodules and nodule features as inputs to a neural network that estimates the malignancy risk for the entire CT scan. We trained our algorithm on a part of the NLST dataset, and validated it on the other datasets. Special care was taken to ensure there was no patient overlap between the train and validation sets. RESULTS AND CONCLUSIONS: The proposed deep learning model is shown to: (a) generalize well across all three data sets, achieving AUC between 86% to 94%, with our external test-set (LHMC) being at least twice as large compared to other works; (b) have better performance than the widely accepted PanCan Risk Model, achieving 6 and 9% better AUC score in our two test sets; (c) have improved performance compared to the state-of-the-art represented by the winners of the Kaggle Data Science Bowl 2017 competition on lung cancer screening; (d) have comparable performance to radiologists in estimating cancer risk at a patient level.

Optimization problems in correlated networks.

BACKGROUND: Solving the shortest path and min-cut problems are key in achieving high-performance and robust communication networks. Those problems have often been studied in deterministic and uncorrelated networks both in their original formulations as well as in several constrained variants. However, in real-world networks, link weights (e.g., delay, bandwidth, failure probability) are often correlated due to spatial or temporal reasons, and these correlated link weights together behave in a different manner and are not always additive, as commonly assumed. METHODS: In this paper, we first propose two correlated link weight models, namely (1) the deterministic correlated model and (2) the (log-concave) stochastic correlated model. Subsequently, we study the shortest path problem and the min-cut problem under these two correlated models. RESULTS AND CONCLUSIONS: We prove that these two problems are NP-hard under the deterministic correlated model, and even cannot be approximated to arbitrary degree in polynomial time. However, these two problems are solvable in polynomial time under the (constrained) nodal deterministic correlated model, and can be solved by convex optimization under the (log-concave) stochastic correlated model.

From epidemics to information propagation: striking differences in structurally similar adaptive network models.

The continuous-time adaptive susceptible-infected-susceptible (ASIS) epidemic model and the adaptive information diffusion (AID) model are two adaptive spreading processes on networks, in which a link in the network changes depending on the infectious state of its end nodes, but in opposite ways: (i) In the ASIS model a link is removed between two nodes if exactly one of the nodes is infected to suppress the epidemic, while a link is created in the AID model to speed up the information diffusion; (ii) a link is created between two susceptible nodes in the ASIS model to strengthen the healthy part of the network, while a link is broken in the AID model due to the lack of interest in informationless nodes. The ASIS and AID models may be considered as first-order models for cascades in real-world networks. While the ASIS model has been exploited in the literature, we show that the AID model is realistic by obtaining a good fit with Facebook data. Contrary to the common belief and intuition for such similar models, we show that the ASIS and AID models exhibit different but not opposite properties. Most remarkably, a unique metastable state always exists in the ASIS model, while there an hourglass-shaped region of instability in the AID model. Moreover, the epidemic threshold is a linear function in the effective link-breaking rate in the AID model, while it is almost constant but noisy in the AID model.

Random line graphs and a linear law for assortativity.

For a fixed number N of nodes, the number of links L in the line graph H(N,L) can only appear in consecutive intervals, called a band of L. We prove that some consecutive integers can never represent the number of links L in H(N,L), and they are called a bandgap of L. We give the exact expressions of bands and bandgaps of L. We propose a model which can randomly generate simple graphs which are line graphs of other simple graphs. The essence of our model is to merge step by step a pair of nodes in cliques, which we use to construct line graphs. Obeying necessary rules to ensure that the resulting graphs are line graphs, two nodes to be merged are randomly chosen at each step. If the cliques are all of the same size, the assortativity of the line graphs in each step are close to 0, and the assortativity of the corresponding root graphs increases linearly from -1 to 0 with the steps of the nodal merging process. If we dope the constructing elements of the line graphs-the cliques of the same size-with a relatively smaller number of cliques of different size, the characteristics of the assortativity of the line graphs is completely altered. We also generate line graphs with the cliques whose sizes follow a binomial distribution. The corresponding root graphs, with binomial degree distributions, zero assortativity, and semicircle eigenvalue distributions, are equivalent to Erdos-Rényi random graphs.

Epidemic threshold and topological structure of susceptible-infectious-susceptible epidemics in adaptive networks.

The interplay between disease dynamics on a network and the dynamics of the structure of that network characterizes many real-world systems of contacts. A continuous-time adaptive susceptible-infectious-susceptible (ASIS) model is introduced in order to investigate this interaction, where a susceptible node avoids infections by breaking its links to its infected neighbors while it enhances the connections with other susceptible nodes by creating links to them. When the initial topology of the network is a complete graph, an exact solution to the average metastable-state fraction of infected nodes is derived without resorting to any mean-field approximation. A linear scaling law of the epidemic threshold τ(c) as a function of the effective link-breaking rate ω is found. Furthermore, the bifurcation nature of the metastable fraction of infected nodes of the ASIS model is explained. The metastable-state topology shows high connectivity and low modularity in two regions of the τ,ω plane for any effective infection rate τ>τ(c): (i) a "strongly adaptive" region with very high ω and (ii) a "weakly adaptive" region with very low ω. These two regions are separated from the other half-open elliptical-like regions of low connectivity and high modularity in a contour-line-like way. Our results indicate that the adaptation of the topology in response to disease dynamics suppresses the infection, while it promotes the network evolution towards a topology that exhibits assortative mixing, modularity, and a binomial-like degree distribution.