Transfer Learning-based Computer-aided Diagnosis System for Predicting Grades of Diabetic Retinopathy

Diabetic retinopathy (DR) diagnosis through digital fundus images requires clinical experts to recognize the presence and importance of many intricate features. This task is very difficult for ophthalmologists and time-consuming. Therefore, many computer-aided diagnosis (CAD) systems were developed to automate this screening process of DR. In this paper, a CAD-DR system is proposed based on preprocessing and a pre-train transfer learning-based convolutional neural network (PCNN) to recognize the five stages of DR through retinal fundus images. To develop this CAD-DR system, a preprocessing step is performed in a perceptual-oriented color space to enhance the DR-related lesions and then a standard pre-train PCNN model is improved to get high classification results. The architecture of the PCNN model is based on three main phases. Firstly, the training process of the proposed PCNN is accomplished by using the expected gradient length (EGL) to decrease the image labeling efforts during the training of the CNN model. Secondly, the most informative patches and images were automatically selected using a few pieces of training labeled samples. Thirdly, the PCNN method generated useful masks for prognostication and identified regions of interest. Fourthly, the DR-related lesions involved in the classification task such as micro-aneurysms, hemorrhages, and exudates were detected and then used for recognition of DR. The PCNN model is pre-trained using a high-end graphical processor unit (GPU) on the publicly available Kaggle benchmark. The obtained results demonstrate that the CAD-DR system outperforms compared to other state-of-the-art in terms of sensitivity (SE), specificity (SP), and accuracy (ACC). On the test set of 30,000 images, the CAD-DR system achieved an average SE of 93.20%, SP of 96.10%, and ACC of 98%. This result indicates that the proposed CAD-DR system is appropriate for the screening of the severity-level of DR.     

An Optimal Deep Learning Based Computer-Aided Diagnosis System for Diabetic Retinopathy

Diabetic Retinopathy (DR) is a significant blinding disease that poses serious threat to human vision rapidly. Classification and severity grading of DR are difficult processes to accomplish. Traditionally, it depends on ophthalmoscopically-visible symptoms of growing severity, which is then ranked in a stepwise scale from no retinopathy to various levels of DR severity. This paper presents an ensemble of Orthogonal Learning Particle Swarm Optimization (OPSO) algorithm-based Convolutional Neural Network (CNN) Model EOPSO-CNN in order to perform DR detection and grading. The proposed EOPSO-CNN model involves three main processes such as preprocessing, feature extraction, and classification. The proposed model initially involves preprocessing stage which removes the presence of noise in the input image. Then, the watershed algorithm is applied to segment the preprocessed images. Followed by, feature extraction takes place by leveraging EOPSO-CNN model. Finally, the extracted feature vectors are provided to a Decision Tree (DT) classifier to classify the DR images. The study experiments were carried out using Messidor DR Dataset and the results showed an extraordinary performance by the proposed method over compared methods in a considerable way. The simulation outcome offered the maximum classification with accuracy, sensitivity, and specificity values being 98.47%, 96.43%, and 99.02% respectively.     

Strategy for Rapid Diabetic Retinopathy Exposure Based on Enhanced Feature Extraction Processing

In the modern world, one of the most severe eye infections brought on by diabetes is known as diabetic retinopathy (DR), which will result in retinal damage, and, thus, lead to blindness. Diabetic retinopathy (DR) can be well treated with early diagnosis. Retinal fundus images of humans are used to screen for lesions in the retina. However, detecting DR in the early stages is challenging due to the minimal symptoms. Furthermore, the occurrence of diseases linked to vascular anomalies brought on by DR aids in diagnosing the condition. Nevertheless, the resources required for manually identifying the lesions are high. Similarly, training for Convolutional Neural Networks (CNN) is more time-consuming. This proposed research aims to improve diabetic retinopathy diagnosis by developing an enhanced deep learning model (EDLM) for timely DR identification that is potentially more accurate than existing CNN-based models. The proposed model will detect various lesions from retinal images in the early stages. First, characteristics are retrieved from the retinal fundus picture and put into the EDLM for classification. For dimensionality reduction, EDLM is used. Additionally, the classification and feature extraction processes are optimized using the stochastic gradient descent (SGD) optimizer. The EDLM’s effectiveness is assessed on the KAGGLE dataset with 3459 retinal images, and results are compared over VGG16, VGG19, RESNET18, RESNET34, and RESNET50. Experimental results show that the EDLM achieves higher average sensitivity by 8.28% for VGG16, by 7.03% for VGG19, by 5.58% for ResNet18, by 4.26% for ResNet 34, and by 2.04% for ResNet 50, respectively.     

Classification of Fundus Images Based on Deep Learning for Detecting Eye Diseases

Various techniques to diagnose eye diseases such as diabetic retinopathy (DR), glaucoma (GLC), and age-related macular degeneration (AMD), are possible through deep learning algorithms. A few recent studies have examined a couple of major diseases and compared them with data from healthy subjects. However, multiple major eye diseases, such as DR, GLC, and AMD, could not be detected simultaneously by computer-aided systems to date. There were just high-performance-outcome researches on a pair of healthy and eye-diseased group, besides of four categories of fundus image classification. To have a better knowledge of multi-categorical classification of fundus photographs, we used optimal residual deep neural networks and effective image preprocessing techniques, such as shrinking the region of interest, iso-luminance plane contrast-limited adaptive histogram equalization, and data augmentation. Applying these to the classification of three eye diseases from currently available public datasets, we achieved peak and average accuracies of 91.16% and 85.79%, respectively. The specificities for images from the eyes of healthy, GLC, AMD, and DR patients were 90.06%, 99.63%, 99.82%, and 91.90%, respectively. The better specificity performances may alert patient in an early stage of eye diseases to prevent vision loss. This study presents a possible occurrence of a multi-categorical deep neural network technique that can be deemed as a successful pilot study of classification for the three most-common eye diseases and can be used for future assistive devices in computer-aided clinical applications.     

Early Detection of Diabetic Retinopathy Using Machine Intelligence through Deep Transfer and Representational Learning

Diabetic retinopathy (DR) is a retinal disease that causes irreversible blindness. DR occurs due to the high blood sugar level of the patient, and it is clumsy to be detected at an early stage as no early symptoms appear at the initial level. To prevent blindness, early detection and regular treatment are needed. Automated detection based on machine intelligence may assist the ophthalmologist in examining the patients’ condition more accurately and efficiently. The purpose of this study is to produce an automated screening system for recognition and grading of diabetic retinopathy using machine learning through deep transfer and representational learning. The artificial intelligence technique used is transfer learning on the deep neural network, Inception-v4. Two configuration variants of transfer learning are applied on Inception-v4: Fine-tune mode and fixed feature extractor mode. Both configuration modes have achieved decent accuracy values, but the fine-tuning method outperforms the fixed feature extractor configuration mode. Fine-tune configuration mode has gained 96.6% accuracy in early detection of DR and 97.7% accuracy in grading the disease and has outperformed the state of the art methods in the relevant literature.     

Automatic diagnosis of diabetic retinopathy with the aid of adaptive average filtering with optimized deep convolutional neural network

The most effective treatment for diabetic retinopathy (DR) is the early detection through regular screening, which is critical for a better prognosis. Automatic screening of the images would assist the physicians in diagnosing the condition of patients easily and accurately. This condition searches out for special importance of image processing technology in the way of processing the retinal fundus images. Accordingly, this article plans to develop an automatic DR detection model with the aid of three main stages like (a) image preprocessing, (b) blood vessel segmentation, and (c) classification. The preprocessing phase includes two steps: conversion of RGB to Lab, and contrast enhancement. The Histogram equalization process is done using the contrast enhancement of an image. To the next of preprocessing, the segmentation phase starts with a valuable procedure. It includes (a), thresholding the contrast-enhanced and filtered images, (b) thresholding the keypoints of contrast-enhanced and filtered images, and (c) adding both thresholded binary images. Here, the filtering process is performed by proposed adaptive average filtering, where the filter coefficients are tuned or optimized by an improved meta-heuristic algorithm called fitness probability-based CSO (FP-CSO). Finally, the classification part uses Deep CNN, where the improvement is exploited on the convolutional layer, which is optimized by the same improved FP-CSO. Since the conventional CSO depends on a fitness probability in the improved algorithm, the proposed algorithm termed as FP-CSO. Finally, valuable comparative and performance analysis has confirmed the effectiveness of the proposed model.     

Detection of the Pine Wilt Disease Tree Candidates for Drone Remote Sensing Using Artificial Intelligence Techniques

Pine wilt disease (PWD) has recently caused substantial pine tree losses in Republic of Korea. PWD is considered a severe problem due to the importance of pine trees to Korean people, so this problem must be handled appropriately. Previously, we examined the history of PWD and found that it had already spread to some regions of Republic of Korea; these became our study area. Early detection of PWD is required. We used drone remote sensing techniques to detect trees with similar symptoms to trees infected with PWD. Drone remote sensing was employed because it yields high-quality images and can easily reach the locations of pine trees. To differentiate healthy pine trees from those with PWD, we produced a land cover (LC) map from drone images collected from the villages of Anbi and Wonchang by classifying them using two classifier methods, i.e., artificial neural network (ANN) and support vector machine (SVM). Furthermore, compared the accuracy of two types of Global Positioning System (GPS) data, collected using drone and hand-held devices, for identifying the locations of trees with PWD. We then divided the drone images into six LC classes for each study area and found that the SVM was more accurate than the ANN at classifying trees with PWD. In Anbi, the SVM had an overall accuracy of 94.13%, which is 6.7% higher than the overall accuracy of the ANN, which was 87.43%. We obtained similar results in Wonchang, for which the accuracy of the SVM and ANN was 86.59% and 79.33%, respectively. In terms of the GPS data, we used two type of hand-held GPS device. GPS device 1 is corrected by referring to the benchmarks sited on both locations, while the GPS device 2 is uncorrected device which used the default setting of the GPS only. The data collected from hand-held GPS device 1 was better than those collected using hand-held GPS device 2 in Wonchang. However, in Anbi, we obtained better results from GPS device 2 than from GPS device 1. In Anbi, the error in the data from GPS device 1 was 7.08 m, while that of the GPS device 2 data was 0.14 m. In conclusion, both classifiers can distinguish between healthy trees and those with PWD based on LC data. LC data can also be used for other types of classification. There were some differences between the hand-held and drone GPS datasets from both areas.     

Portable Food-Freshness Prediction Platform Based on Colorimetric Barcode Combinatorics and Deep Convolutional Neural Networks

Artificial scent screening systems (known as electronic noses, E-noses) have been researched extensively. A portable, automatic, and accurate, real-time E-nose requires both robust cross-reactive sensing and fingerprint pattern recognition. Few E-noses have been commercialized because they suffer from either sensing or pattern-recognition issues. Here, cross-reactive colorimetric barcode combinatorics and deep convolutional neural networks (DCNNs) are combined to form a system for monitoring meat freshness that concurrently provides scent fingerprint and fingerprint recognition. The barcodes—comprising 20 different types of porous nanocomposites of chitosan, dye, and cellulose acetate—form scent fingerprints that are identifiable by DCNN. A fully supervised DCNN trained using 3475 labeled barcode images predicts meat freshness with an overall accuracy of 98.5%. Incorporating DCNN into a smartphone application forms a simple platform for rapid barcode scanning and identification of food freshness in real time. The system is fast, accurate, and non-destructive, enabling consumers and all stakeholders in the food supply chain to monitor food freshness.     

A Feature Selection Strategy to Optimize Retinal Vasculature Segmentation

Diabetic retinopathy (DR) is a complication of diabetes mellitus that appears in the retina. Clinitians use retina images to detect DR pathological signs related to the occlusion of tiny blood vessels. Such occlusion brings a degenerative cycle between the breaking off and the new generation of thinner and weaker blood vessels. This research aims to develop a suitable retinal vasculature segmentation method for improving retinal screening procedures by means of computer-aided diagnosis systems. The blood vessel segmentation methodology relies on an effective feature selection based on Sequential Forward Selection, using the error rate of a decision tree classifier in the evaluation function. Subsequently, the classification process is performed by three alternative approaches: artificial neural networks, decision trees and support vector machines. The proposed methodology is validated on three publicly accessible datasets and a private one provided by Hospital Sant Joan of Reus. In all cases we obtain an average accuracy above 96% with a sensitivity of 72% in the blood vessel segmentation process. Compared with the state-of-the-art, our approach achieves the same performance as other methods that need more computational power. Our method significantly reduces the number of features used in the segmentation process from 20 to 5 dimensions. The implementation of the three classifiers confirmed that the five selected features have a good effectiveness, independently of the classification algorithm.     

航天光学遥感器信噪比的人工神经网络评价

提出基于人工神经网络进行航天光学遥感器信噪比评价的方法,首先对航天遥感图像进行分析,从图像中将与景物结构和噪声有关的特征向量分别提取出来,作为ANN的输入。网络通过对大量信噪比已知的图像样本训练后,可完成对航天光学遥感器传输下来的任意一幅地面景物图像进行系统的信噪比测试,从而避免了采用特定景物目标进行测量中的诸多弊端,测量平均误差低于10%。     

Optimal Deep Learning Based Inception Model for Cervical Cancer Diagnosis

Prevention of cervical cancer becomes essential and is carried out by the use of Pap smear images. Pap smear test analysis is laborious and tiresome work performed visually using a cytopathologist. Therefore, automated cervical cancer diagnosis using automated methods are necessary. This paper designs an optimal deep learning based Inception model for cervical cancer diagnosis (ODLIM-CCD) using pap smear images. The proposed ODLIM-CCD technique incorporates median filtering (MF) based pre-processing to discard the noise and Otsu model based segmentation process. Besides, deep convolutional neural network (DCNN) based Inception with Residual Network (ResNet) v2 model is utilized for deriving the feature vectors. Moreover, swallow swarm optimization (SSO) based hyperparameter tuning process is carried out for the optimal selection of hyperparameters. Finally, recurrent neural network (RNN) based classification process is done to determine the presence of cervical cancer or not. In order to showcase the improved diagnostic performance of the ODLIM-CCD technique, a series of simulations occur on benchmark test images and the outcomes highlighted the improved performance over the recent approaches with a superior accuracy of 0.9661.     

基于独立分量分析特征提取的复合神经网络故障诊断法

首先利用基于固定点迭代的快速算法(FASTICA)提取不同机械状态模式(包括正常、齿轮故障及机座松动)特征．随后以此训练某一典型神经网络(如径向基网络或自组织映射网络)，以实现模式的最终分类。借助独立分量分析(ICA)及基于残余互信息(RMI)的二次特征抽取策略，隐藏于多通道振动观测中的高阶特征得以有效提取，进而实现机械状态模式的准确识别。对照分类实验结果表明，基于无导师学习的自组织映射(ICA-SOM)分类方法不仅具有较好的故障模式分类能力，且实现简单直观．在机器健康状况监测中有较大的应用潜力。     

Neural network identification of aerial objects from radar images

The type of neural network for the effective identification of objects in radar images is selected. Estimates of the recognition of aerial objects in radar images are obtained. Algorithms for processing radar information for high-grade functioning of a neural network classifier are obtained. __________ Translated from Izmeritel’naya Tekhnika, No. 2, pp. 53–59, February, 2006.     

An effective deep learning model for grading abnormalities in retinal fundus images using variational auto-encoders

Diabetic retinopathy (DR) and Diabetic Macular Edema (DME) are severe diseases that affect the eyes due to damage in blood vessels. Computer-aided automated grading will help clinicians conduct disease diagnoses at ease. Experiments of automated image processing with deep learning techniques using CNN produce promising results, especially in the medical imaging domain. However, the disease grading tasks in retinal images using CNN struggle to retain high-quality information at the output. A novel deep learning model based on variational auto-encoder to grade DR and DME abnormalities in retinal images is proposed. The objective of the proposed model is to extract the most relevant retinal image features efficiently. It focuses on addressing less relevant candidate region generation and translational invariance present in images. The experiments are conducted in IDRID dataset and evaluated using accuracy, U-kappa, sensitivity, specificity and precision metrics. The results outperform compared with other state-of-art techniques.     

Artificial neural networks for pattern recognition

This tutorial article deals with the basics of artificial neural networks (ANN) and their applications in pattern recognition. ANN can be viewed as computing models inspired by the structure and function of the biological neural network. These models are expected to deal with problem solving in a manner different from conventional computing. A distinction is made between pattern and data to emphasize the need for developing pattern processing systems to address pattern recognition tasks. After introducing the basic principles of ANN, some fundamental networks are examined in detail for their ability to solve simple pattern recognition tasks. These fundamental networks together with the principles of ANN will lead to the development of architectures for complex pattern recognition tasks. A few popular architectures are described to illustrate the need to develop an architecture specific to a given pattern recognition problem. Finally several issues that still need to be addressed to solve practical problems using ANN approach are discussed. This paper is mostly a consolidation of work reported by several researchers in the literature, some of which is cited in the references. The author has borrowed several ideas and illustrations from the references quoted in this paper.     

PET image reconstruction using ANN

The aim of this study is to improve the positron emission tomography (PET) image quality for medical diagnosis. The statistical reconstructions on the maximum a posteriori (MAP) algorithm often results in a blurring effect, which fails to determine the toughness class in the reconstructed image. The development of new reconstruction algorithms for PET is an active field of research. In this article, artificial neural network (ANN) is proposed for replicating the output image, which is generated from the acquired projection data with the corresponding angles using the PET images. This article proposes the advantage of arranging the neural network to stock up the information of the continuous capacity. This reduces the storage space and recuperates as much sequence of the continuous quantity as possible. The performance of image quality parameters using ANN is better when compared with MAP, FBP‐NN (filtered back projection with nearest neighbor interpolation). Thus ANN provides 63% better peak signal to noise ratio (PSNR) when compared with FBP‐NN and 47% better when compared to MAP. Thus, ANN is better than FBP and MAP algorithm, by providing better PSNR. © 2014 Wiley Periodicals, Inc. Int J Imaging Syst Technol, 24, 249–255, 2014     

基于卷积神经网络和Transformer网络的鸟声识别

针对传统鸟声识别算法中特征提取方式单一、分类识别准确率低等问题,提出一种结合卷积神经网络和Transformer网络的鸟声识别方法。该方法综合考虑网络局部特征学习和全局上下文依赖性构造,从原始鸟声音频信号中提取短时傅里叶变换(Short Time Fourier Transform,STFT)语谱图特征,将其输入到卷积神经网络(ConvolutionalNeural Network,CNN)中提取局部频谱特征信息,同时提取鸟声信号的对数梅尔特征及一阶差分、二阶差分特征用于合成梅尔频率倒谱系数(Mel Frequency Cepstrum Coefficient,MFCC)混合特征向量,将其输入到Transformer网络中获取全局序列特征信息,最后融合所提取的特征可得到更丰富的鸟声特征参数,通过Softmax分类器得到鸟声识别结果。在Birdsdata和xeno-canto鸟声数据集上进行实验,平均识别准确率分别达到了97.81%和89.47%。实验结果表明该方法相较于其他现有的鸟声识别模型具有更高的识别准确率。     

Digital Vision Based Concrete Compressive Strength Evaluating Model Using Deep Convolutional Neural Network

Compressive strength of concrete is a significant factor to assess building structure health and safety. Therefore, various methods have been developed to evaluate the compressive strength of concrete structures. However, previous methods have several challenges in costly, time-consuming, and unsafety. To address these drawbacks, this paper proposed a digital vision based concrete compressive strength evaluating model using deep convolutional neural network (DCNN). The proposed model presented an alternative approach to evaluating the concrete strength and contributed to improving efficiency and accuracy. The model was developed with 4,000 digital images and 61,996 images extracted from video recordings collected from concrete samples. The experimental results indicated a root mean square error (RMSE) value of 3.56 (MPa), demonstrating a strong feasibility that the proposed model can be utilized to predict the concrete strength with digital images of their surfaces and advantages to overcome the previous limitations. This experiment contributed to provide the basis that could be extended to future research with image analysis technique and artificial neural network in the diagnosis of concrete building structures.     

基于GST-ECNN的运动想象脑电信号识别方法

在对脑电信号的解码研究中,存在着现有时频分析方法对高频信号处理能力有限,多通道信号信息冗余,常用卷积神经网络分类器ReLU激活函数受学习速率的影响较大,对不同层采用相同的正则化很难获得满意结果等问题。为此,提出了一种基于广义S变换特征提取和增强卷积神经网络分类相结合的方法,同时提出一种结合Relief算法和向前选择搜索策略的包裹式方法进行通道选择。结果表明,提出的方法利用较少的信号通道,具有更强的特征提取和分类的能力,在第Ⅳ届BCI的数据集I上取得最高98.44±1.5%的分类准确率,高于其他现有算法。该方法良好的分类性能不仅减少了计算消耗,也有效提高了分类准确率,对脑电信号特征提取和分类具有一定的参考意义。     

卷积神经网络在无纺布缺陷分类检测中的应用

目的针对传统无纺布缺陷分类检测中人工依赖性强、效率低等问题,提出一种能够满足工厂要求的卷积神经网络分类检测方法。方法首先建立包括脏点、褶皱、断裂、缺纱和无缺陷等5种共计7万张无纺布图像样本库,其次构造一个具有不同神经元个数的卷积层和池化层的神经网络,然后采用反向传播算法逐层更新权值,通过梯度下降法最小化损失函数,最后利用Softmax分类器实现无纺布的缺陷分类检测。结果构建了12层的卷积神经网络,通过2万张样本进行测试实验,无缺陷样本准确率可以达到100%,缺陷样本分类准确率均在95%以上,检测时间在35 ms以内。结论该方法能够满足工业生产线中对于无纺布缺陷实时分类检测的要求。