Search

In medicine, any abnormalities in the number of chromosomes or the structure of chromosomes may cause the newborn baby to suffer from genetic diseases, such as Edward syndrome and so on. Chromosome karyotype analysis is the most important and common method for prenatal diagnosis to determine whether a newborn baby has chromosome defects refers to segment chromosome instances from stained cell images and arrange chromosome instances according to their categories. However, due to the non-rigid nature of chromosomes, chromosome instances may overlap and adhere to each other, which makes the task of segmenting chromosome instances time-consuming and error-prone. This paper proposes a novel chromosome instance segmentation method that includes three stages. First, we segment a given stained cell image into several segments using geometric connectivity. Second, a machine learning method is proposed to distinguish chromosome of individual instances and clusters. Finally, a deep learning-based method is applied to separate chromosome instances from clusters. It shows that the proposed method achieves 97.61% instance segmentation accuracy in a hold-out clinical dataset with 162 cell images consisting of 7,452 chromosome instances, which is a promising result in clinical application. The innovation of this work is to combine geometry and deep learning to handle tasks for different stages of chromosome instance segmentation issue. The benefit of this innovation is that it can obtain a much better performance than existing geometric-based methods with a small number of training samples. Meanwhile, the segmentation performance of the proposed method is superior to existing methods fully based on deep learning.

Background: In medicine, chromosome karyotyping analysis plays a crucial role in prenatal diagnosis for diagnosing whether a fetus has severe defects or genetic diseases. However, chromosome instance segmentation is the most critical obstacle to automatic chromosome karyotyping analysis due to the complicated morphological characteristics of chromosome clusters, restricting chromosome karyotyping analysis to highly depend on skilled clinical analysts.Method: In this paper, we build a clinical dataset and propose multiple segmentation baselines to tackle the chromosome instance segmentation problem of various overlapping and touching chromosome clusters. First, we construct a clinical dataset for deep learning-based chromosome instance segmentation models by collecting and annotating 1,655 privacy-removal chromosome clusters. After that, we design a chromosome instance labeled dataset augmentation CILA algorithm for the clinical dataset to improve the generalization performance of deep learning-based models. Last, we propose a chromosome instance segmentation framework and implement multiple baselines for the proposed framework based on various instance segmentation models. Experiments evaluated on the clinical dataset show that the best baseline of the proposed framework based on the Mask-RCNN model yields an outstanding result with 77% mAP, 97.5% AP50, and 95.5% AP75 segmentation precision, and 95.38% accuracy, which exceeds results reported in current chromosome instance segmentation methods.

Karyotyping is a vital cytogenetics technique widely applied in prenatal diagnosis and genetic screening. Heavily dependent on the experience of the cytogeneticist and easily affected by the attention, karyotype analysis is a time-consuming and error-prone task, and incorrect karyotypes may result in misdiagnosis conclusions. This paper proposes an effective identification framework for incorrect karyotypes based on deep learning technology. Firstly, a chromosome classifier is trained and utilized to classify chromosome instances in karyotypes performed manually by cytogeneticists. Afterward, when the categories of chromosome instances classified by the classifier are not identical to those categories classified by cytogeneticists, the proposed framework identifies these corresponding karyotypes as unreliable. Finally, the expert team review these unreliable karyotypes and confirmed their correctness. Extensive experiments show that the proposed framework achieves 100% recall and 88.89% F1 score on incorrect karyotypes, which demonstrates the advancement and promising effectiveness of the proposed framework to address the issue of incorrect karyotypes.

In medicine, chromosome karyotype analysis is vital for genetic disease diagnoses, such as Edward syndrome, Patau syndrome, and Down syndrome. However, the chromosome karyotype analysis is usually manually done by extensive experienced clinical analysts, which is tedious and time-consuming. Consequently, automatic or partial automatic chromosome karyotyping is essential to alleviate clinical analysts’ jobs. This paper proposes a chromosome cluster identification framework based on geometric chromosome features and machine learning algorithms to identify chromosome clusters needed to further process in the automated instance segmentation task. In the proposed framework, we first collect multiple dimensions of chromosome geometric features into a feature tuple, including object area, bounding box area, convex area, extent, solidity, perimeter, equivalent diameter, eccentricity, major axis length, minor axis length, and minor-major axis ratio. Second, we classify these chromosome feature tuples utilizing different machine learning classification algorithms. The experiment results show that our proposed method yields 96.21 ± 0.91% classification accuracy and 0.9832 ± 0.0111AUC (Area Under The Curve) value in the clinical dataset. The highlight of this paper is that the performance of the proposed approach has exceeded the existing geometric features threshold-based methods and multiple end-to-end deep learning-based baselines. Moreover, our proposed method can be deployed on any devices and platforms with a Python running environment, which significantly improves application flexibility. To facilitate peers in reproducing and employing our work, we release the code and the corresponding clinical dataset on Github.

Medical image segmentation is one of the most important tasks in machine learning. In order to help with the doctors, we focus on computer-aided automatic computer tomography (CT) segmentation of liver tumors. Instead of using a single model during the segmentation process, we argue that the filter made by the model of object detection before segmentation could improve the accuracy of segmentation. From the motivation above, we proposed a multi-stage deep convolutional neural network (DCNN) model, which can be used for segmentation of the tumors in the livers. To further improve the accuracy of the object detection model, we proposed an iterative training method which boosts the performance of our model and yield more accurate localization of the tumors. Finally, the evaluation of proposed method on our private dataset from hospital and public dataset from MICCAI-LiTS2017 suggested that our model is superior to the current state-of-the-art models.

Although anomaly detection is an urgent and significant task in various applications, there are few good enough methods to meet the requirements. The challenge of anomaly detection lies in how to distinguish the diverse or even unknown anomalies. Therefore, previous studies have focused on imagelevel anomaly detection tasks, and mostly based on unsupervised methods. These methods only consider the characteristics of normal samples and are confined to the “normal” category, ignoring information between samples. In particular, the limitations of unsupervised methods are more obvious for pixellevel abnormalities such as the chromosomal structural anomalies investigated in this paper, because the similarity between normal and abnormal samples is relatively high. In this work, we propose a supervised anomaly detection approach Classification - Enhanced GANomaly (abbreviated as CS-GANomaly). By fusing the category information with the sample, the spatial distribution of the sample has regional characteristics and then combining the anomaly detection strategies to carry out multidimension anomaly detection. The CS-GANomaly adopts the GAN framework for adversarial training and adapts the GAN output to anomaly detection. The experimental results prove the superiority of our approach, whether it is a pixel-level anomaly in abnormal chromosome dataset or image-level anomaly in the benchmark dataset.

In this paper, we present a garbage image classification framework to tackle the waste sorting problem which besets residents around the world every day. The proposed framework consists of two modules, a convolutional neural network backbone transferred from the ImageNet classification task and a customized network header designed for the garbage image classification task. To friendly deploy into mobile devices, the proposed method makes an artful tradeoff between classification accuracy and running efficiency. The proposed framework yields 95.62% online classification accuracy on the test dataset provided by Huawei cloud with 95 ms per image inference time occupying 897 Megabytes GPU memory.

Chromosome classification is an important way of clinical genetic diagnosis. However, it is usually done manually by skilled cytologists, which requires extensive experience, domain knowledge, and tedious manual work. In this paper, we tackle the task of chromosome classification utilizing deep learning techniques based on aggregated residual architecture. We proposed a novel chromosome classification method named MixNet. Firstly, we design a deep learning-based chromosome classification framework that consists of a chromosome encoding backbone and an adaptive header. Secondly, we design the chromosome backbone utilizing the aggregated residual architecture and propose the adaptive header by aggregating pooling layers to classify chromosome latent features. Lastly, we propose a training algorithm to fine-tune MixNet. The experimental results demonstrate that our proposed method yields a mean classification accuracy of 98.73% with 0.81 standard deviations on our G-band clinical dataset and a mean classification accuracy of 96.50% with 1.87 standard deviations on public Q-band dataset. To our best knowledge, it is the state-of-the-art result of the public Q-band dataset.

Method: This paper focuses on chromosome classification because it is critical for chromosome karyotyping. In recent years, deep learning-based methods are the most promising methods for solving the tasks of chromosome classification. Although the deep learning-based Inception architecture has yielded state-of-the-art performance in the 2015 ILSVRC challenge, it has not been used in chromosome classification tasks so far. Therefore, we develop an automatic chromosome classification approach named CIR-Net based on Inception-ResNet which is an optimized version of Inception. However, the classification performance of origin Inception-ResNet on the insufficient chromosome dataset still has a lot of capacity for improvement. Further, we propose a simple but effective augmentation method called CDA for improving the performance of CIR-Net. Results: The experimental results show that our proposed method achieves 95.98% classification accuracy on the clinical G-band chromosome dataset whose training dataset is insufficient. Moreover, the proposed augmentation method CDA improves more than 8.5% (from 87.46% to 95.9%) classification accuracy compared to other methods. In this paper, the experimental results demonstrate that our proposed method is recent the most effective solution for solving clinical chromosome classification problems in chromosome auto-karyotyping on the condition of the insufficient training dataset. Code and Dataset are available at https://github.com/CloudDataLab/CIR-Net. © 2020 IEEE. Personal use of this material is permitted. Permission from IEEE must be obtained for all other uses, in any current or future media, including reprinting/republishing this material for advertising or promotional purposes, creating new collective works, for resale or redistribution to servers or lists, or reuse of any copyrighted component of this work in other works.

Chromosome karyotyping analysis plays a crucial role in prenatal diagnosis for diagnosing whether a fetus has severe defects or genetic diseases. However, due to the complicated morphological characteristics of various types of chromosome clusters, chromosome instance segmentation is the most challenging stage of chromosome karyotyping analysis, leading chromosome karyotyping analysis to highly dependent on skilled clinical analysts. Since most of the chromosome instance segmentation efforts are currently devoted to segmenting chromosome instances from different types of chromosome clusters, type identification of chromosome clusters is a vital anterior task for chromosome instance segmentation. Firstly, this paper proposes an automatic approach for chromosome cluster identification using recent transfer learning techniques. The proposed framework is based on ResNeXt weakly-supervised learning (WSL) pre-trained backbone and a task-specific network header. Secondly, this paper proposes a fast training methodology that tunes our framework from coarse-to-fine gradually. Extensive evaluations on a clinical dataset consisting of 6592 clinical chromosome samples show that the proposed framework achieves 94.09% accuracy, 92.79% sensitivity, and 98.03% specificity. Such performance is superior to the best baseline model that we obtain 92.17% accuracy, 89.1% sensitivity, and 97.42% specificity. To foster research and application in the chromosome cluster type identification, we make our clinical dataset and code available via GitHub.

With the gradual progress of cloud computing, big data, network virtualization and other network technology, exponential growth of data makes traditional network show its weakness completely. Such as inefficiency of data forwarding, poor agility of network deployment and so on. SDN(Software Defined Networking) separates data forwarding from logic control and makes network programmable to improve utilization of network resources, and then, there are efficiency of data forwarding and agility of network deployment. Compared to traditional network, SDN network is more in line with the future network for its characteristic. However, SDN is facing some severe problems too. Like flow table congestion and network security problems. In SDN network, OpenFlow switches can only contain 3000 flow entries which means flow table resources are really rare. Thus, we come out with a strategy of flow table optimizing. We choose a number of switches based on betweenness centrality as the station on "bus route" and issue the least flow entry to realize the whole network working.

SDN(Software Defined Networking) is a new network architecture, which decouples the control plane from data plane and operates the globle network with elaborate abstraction. The control plane advocates a centralized approach of network control, which provides both flexibility and high efficiency. However, there are lots of problems in this architecture and the bottleneck problem is the network's performance. Existing OpenFlow based flow expiry mechanisms relys on a fixed timeout after which the switch proactively removes the flow entries from its flow table which may causes unnecessary flow table occupation. Therefore, in this paper, a strategy of dynamically adjusting timeout time is proposed and the program is implemented. It can dynamically adjust the timeout time according to the actual situation of equipment and data flow, and can better adapt to the network with different data flows. The experiment shows that the strategy can improve the network's performance and the mechanism designed in this paper is proved to be effective.

The development of Internet and Web systems in recent years has made it difficult and challenging to deal with large-scale data. What we usually need to process is incremental data, the scale of which will enlarge with the changing of time. Nowadays, many general queries over data is traditionally based on the full amount of raw data, which becomes a great challenge to performance when the data increases substantially. As these data will not be updated after generated, this paper proposes a query model, called hybrid-granularity model, that data and queries can be preprocessed to form intermediate result sets of different granularities. With query transformation, submitted queries can take advantage of intermediate preprocessing results, to obtain the required final results. In this paper, we also describe query transformation and the methods to seek the solution of best performance with hybrid-granularity model for the specific query. Finally, we analyze and verify the advantage on performance of the proposed model by comparison with the original model in experiment. The proposed solution is used in some practical systems, which shows that this solution can guarantee the correctness of query results while improving the responsive efficiency of the query significantly.

Log data is of great value for operation and maintenance of systems and networks. With log data, system behaviors can be monitored, traced, analyzed, to detect unusual circumstances and identify warnings, for the purpose of taking timely measure. However, nowadays many general systems making queries and retrieval over log data is traditionally based on the full amount of raw data. When the amount of log data increase substantially, it's difficult to execute queries based on the raw data of extremely large scales to meet response time requirements. As log data will not be updated after generated, this paper proposes that data and queries can be preprocessed to form data preprocessing results of different granularities. Queries submitted by users can take advantage of corresponding preprocessing results, to improve the query response time. This paper proposes a model for queries over log data of various granularities. The main work includes the follows (1) develop a query model based on hybrid granularity, which enables a query to execute on various granularities of data sets after preprocessing, (2) analyze and prove the completeness and correctness of the proposed query model based on hybrid granularity, (3) describe the query action based on hybrid granularity model formally and present the algorithm framework of the query transformation, (4) analyze and demonstrate the advantages of efficiency of the proposed query model compared to the original data query model. The proposed solution is used in some practical systems. The results show that this solution can guarantee the correctness of the query results while it is able to improve the responsive efficiency of the query significantly. Preprocessing, (2) analyze and prove the completeness and correctness of the proposed query model based on hybrid granularity, (3) describe the query action based on hybrid-granularity model formally and present the algorithm framework of the query transformation, (4) analyze and demonstrate the advantages in efficiency of the proposed query model compared to the original data query model. The proposed solution is used in some practical systems. The results show that this solution can guarantee the correctness of query results while it is able to improve the responsive efficiency of the query significantly.