Wyniki wyszukiwania - BazTech

1

Travel time picking of ambient noise cross-correlation using a deep neural network combining convolutional neural networks and Transformer

Jin Chunwei, Ye Fang, Zhang Han, Bao Xue

Acta Geophysica

|

2024

|

Vol. 72, no. 1

97--114

EN

The travel time of ambient noise cross-correlation is widely used in geophysics, but traditional methods for picking the travel time of correlation are either difficult to be applied to data with low signal-to-noise ratio (SNR), or make some assumptions which fail to be achieved in many realistic situations, or require a lot of complex calculations. Here, we present a neural network based on convolutional neural networks (CNN) and Transformer for the travel time picking of ambient noise crosscorrelation. CNNs expand the dimension of the vector of each time step for the input of Transformer. Transformer focuses the model’s attention on the key parts of the sequence. Model derives the travel time according to the attention. 102,000 cross-correlations are used to train the network. Compared with traditional methods, the approach is easy to use and has a better performance, especially for the low SNR data. Then, we test our model on another ambient noise cross-correlation dataset, which contains cross-correlations from different regions and at different scales. The model has good performance on the test dataset. It can be seen from the experiment that the travel time of the cross-correlation function of ambient noise with an average SNR as low as 9.3 can be picked. 97.2% of the picked travel times are accurate, and the positive and negative travel time of most cross-correlations are identical (90.2%). Our method can be applied to seismic instrument performance verification, seismic velocity imaging, source location and other applications for its good ability to pick travel time accurately.

2

Segmentation of cancer masses on breast ultrasound images using modified U-net

Khallassi Ihssane, El Yousfi Alaoui My Hachem, Jilbab Abdelilah

Informatyka, Automatyka, Pomiary w Gospodarce i Ochronie Środowiska

|

2023

|

T. 13, nr 3

11--15

EN

Breast cancer causes a huge number of women’s deaths every year. The accurate localization of a breast lesion is a crucial stage. The segmentation of breast ultrasound images participates in the improvement of the process of detection of breast anomalies. An automatic approach of segmentation of breast ultrasound images is presented in this paper, the proposed model is a modified u-net called Attention Residual U-net, designed to help radiologists in their clinical examination to determine adequately the limitation of breast tumors. Attention Residual U-net is a combination of existing models (Convolutional Neural Network U-net, the Attention Gate Mechanism and the Residual Neural Network). Public breast ultrasound images dataset of Baheya hospital in Egypt is used in this work. Dice coefficient, Jaccard index and Accuracy are used to evaluate the performance of the proposed model on the test set. Attention residual u-net can significantly give a dice coefficient = 90%, Jaccard index = 76% and Accuracy = 90%. The proposed model is compared with two other breast segmentation methods on the same dataset. The results show that the modified U-net model was able to achieve accurate segmentation of breast lesions in breast ultrasound images.

PL

Każdego roku rak piersi powoduje ogromną liczbę zgonów kobiet. Dokładna lokalizacja zmiany piersi jest kluczowym etapem. Segmentacja obrazów ultrasonograficznych piersi przyczynia się do poprawy procesu wykrywania nieprawidłowości piersi. W tym artykule przedstawiono automatyczne podejście do segmentacji obrazów ultrasonograficznych piersi, proponowany model to zmodyfikowany U-net, nazwany Attention Residual U-net, zaprojektowany w celu wspomagania radiologów podczas badania klinicznego, w celu odpowiedniego określenia zasięgu guzów piersiowych. Attention Residual U-net jest połączeniem istniejących modeli (konwolucyjną siecią neuronową U-net, Attention Gate Mechanism i Residual Neural Network). W tym badaniu wykorzystano publiczny zbiór danych obrazów ultrasonograficznych piersi szpitala Baheya w Egipcie. Do oceny wydajności zaproponowanego modelu na zbiorze testowym wykorzystano współczynnik Dice'a, indeks Jaccarda i dokładność. Attention Residual U-net może znacznie przyczynić się do uzyskania współczynnika Dice'a równego 90%, indeksu Jaccarda równego 76% i dokładności równiej 90%. Proponowany model został porównany z dwoma innymi metodami segmentacji piersi na tym samym zbiorze danych. Wyniki pokazują, że zmodyfikowany model U-net był w stanie osiągnąć dokładną segmentację zmian piersiowych na obrazach ultrasonograficznych piersi.

3

Implementation of an artificial intelligence-based ECG acquisition system for the detection of cardiac abnormalities

Benba Achraf, El Attaoui Fatima Zahra, Sandabad Sara

Informatyka, Automatyka, Pomiary w Gospodarce i Ochronie Środowiska

|

2023

|

T. 13, nr 1

22--25

EN

The electrocardiogram (ECG) is a common test that measures the electrical activity of the heart. On the ECG, several cardiac abnormalities can be seen, including arrhythmias, which are one of the major causes of cardiac mortality worldwide. The objective for the research community is accurate and automated cardiovascular analysis, especially given the maturity of artificial intelligence technology and its contribution to the health area. The goal of this effort is to create an acquisition system and use artificial intelligence to classify ECG readings. This system is designed in two parts: the first is the signal acquisition using the ECG Module AD8232; the obtained signal is a single derivation that has been amplified and filtered. The second section is the classification for heart illness identification; the suggested model is a deep convolutional neural network with 12 layers that was able to categorize five types of heartbeats from the MIT-BIH arrhythmia database. The results were encouraging, and the embedded system was built.

PL

Elektrokardiogram (EKG) to powszechny test, który mierzy aktywność elektryczną serca. W zapisie EKG można zauważyć kilka nieprawidłowości serca, w tym arytmie, które są jedną z głównych przyczyn śmiertelności sercowej na całym świecie. Celem społeczności naukowej jest dokładna i zautomatyzowana analiza układu sercowo-naczyniowego, zwłaszcza biorąc pod uwagę dojrzałość technologii sztucznej inteligencji i jej wkład w obszar zdrowia. Celem tych wysiłków jest stworzenie systemu akwizycji i wykorzystanie sztucznej inteligencji do klasyfikacji odczytów EKG. System ten składa się z dwóch części: pierwsza to akwizycja sygnału za pomocą modułu EKG AD8232; uzyskany sygnał jest pojedynczą pochodną, która została wzmocniona i przefiltrowana. Druga sekcja to klasyfikacja identyfikacji chorób serca; sugerowany model to głęboka konwolucyjna sieć neuronowa z 12 warstwami, która była w stanie sklasyfikować pięć typów uderzeń serca z bazy danych arytmii MIT-BIH. Wyniki były zachęcające i zbudowano system wbudowany.

4

Convolutional neural networks in the detection of astronomical objects from the Messier catalog

Beluch Witold, Śliwa Paweł

Computer Assisted Methods in Engineering and Science

|

2023

|

Vol. 30, no. 4

461--479

EN

This paper explores the application of convolutional neural networks in the field of amateur astronomy. The authors have employed the available astronomical datasets to develop a detector for identifying astronomical objects from the Messier catalog. A concept framework for creating such a detector for astronomical objects using artificial intelligence tools in the form of a detector based on convolutional neural networks is presented. Augmentation and pre-processing procedures have been used to extend the feature distribution in the training set. Examples confirming the effectiveness of the proposed detector of astronomical objects are presented.

5

A deep hybrid model for human-computer interaction using dynamic hand gesture recognition

Ramalingam Brindha, Angappan Geetha

Computer Assisted Methods in Engineering and Science

|

2023

|

Vol. 30, no. 3

263--276

EN

Dynamic hand gestures attract great interest and are utilized in different fields. Amongthese, man-machine interaction is an interesting area that makes use of the hand to providea natural way of interaction between them. A dynamic hand gesture recognition system isproposed in this paper, which helps to perform control operations in applications such asmusic players, video games, etc. The key motivation of this research is to provide a simple, touch-free system for effortless and faster human-computer interaction (HCI). As thisproposed model employs dynamic hand gestures, HCI is achieved by building a modelwith a convolutional neural network (CNN) and long short-term memory (LSTM) net-works. CNN helps in extracting important features from the images and LSTM helpsto extract the motion information between the frames. Various models are constructedby differing the LSTM and CNN layers. The proposed system is tested on an existing EgoGesture dataset that has several classes of gestures from which the dynamic gesturesare utilized. This dataset is used as it has more data with a complex background, actionsperformed with varying speeds, lighting conditions, etc. This proposed hand gesture recognition system attained an accuracy of 93%, which is better than other existing systemssubject to certain limitations.

6

How many neurons are needed to make a short-term prediction of the Bitcoin exchange rate?

Frontczak Michał, Hachaj Tomasz

Technical Transactions

|

2022

|

Vol. 119, iss. 1

art. no. e2022011

EN

The goal of our work was to select a neural network architecture that would give the best prediction of the Bitcoin exchange rate using historical data. Our work fits into the very important topic of predicting the value of the cryptocurrency exchange rate, and makes use of recent data which, as a result of the high Bitcoin exchange rate dynamics of the last year, differs significantly from those of previous years. We propose and test a number of neural network-based architectures and conduct a discussion of the results. Unlike previous state of-the-art works, we conducted a comprehensive comparison of three different neural network-based models: MLP (multilayer perceptron), LSTM (long short-term memory) and CNN (convolutional neural network). We tested them for a wide range of parameters. The results we present are, to the best of our knowledge, the most up to date when it comes to the application of artificial intelligence methods for the prediction of cryptocurrency exchange rates. The best-performing architectures were used for a website that gives real-time predictions of the Bitcoin exchange rate. The website is available at http://stpbtc-ii.up.krakow.pl/. Source codes of our research are available to download in order to make our experiment reproducible.

PL

Celem naszej pracy było stworzenie architektury sieci neuronowej, która przy wykorzystaniu danych historycznych pozwalałaby na dokładną predykcję kursu Bitcoin. Nasza praca wpisuje się w bardzo ważny temat przewidywania wartości kursu kryptowaluty. Niemniej istotny jest fakt, że w naszej pracy wykorzystujemy najnowsze dane, które z powodu dużej dynamiki kursu Bitcoin w ostatnim roku znacznie różnią się od danych z lat wcześniejszych. Proponujemy i testujemy kilka architektur opartych na sieciach neuronowych oraz przeprowadzamy dyskusję wyników. W odróżnieniu od poprzednich prac, przeprowadzamy wszechstronne porównanie trzech różnych modeli opartych na sieciach neuronowych: MLP (multilayer perceptron), LSTM (long short-term memory) i CNN (convolutional neural network). Przetestowaliśmy je dla szerokiego zakresu parametrów. Przedstawione przez nas wyniki są, według naszej wiedzy, najbardziej aktualnymi, jeśli chodzi o zastosowanie metod sztucznej inteligencji do przewidywania kursów kryptowalut. Najlepiej działająca architektura została wykorzystana na stronie internetowej, która w czasie rzeczywistym prognozuje kurs Bitcoina. Strona ta jest dostępna pod adresem http://stpbtc-ii.up.krakow.pl/. Kody źródłowe naszych badań są dostępne do pobrania w celu umożliwienia odtworzenia naszego eksperymentu.

7

2D inversion of magnetotelluric data using deep learning technology

Liao Xiaolong, Shi Zeyu, Zhang Zhihou, Yan Qixiang, Liu Pengfei

Acta Geophysica

|

2022

|

Vol. 70, no 3

1047--1060

EN

The inverse problem of magnetotelluric data is extremely difficult due to its nonlinear and ill-posed nature. The existing gradient-descent approaches for this task surface from the problems of falling into local minima and relying on reliable initial models, while statistical-based methods are computationally expensive. Inspired by the excellent nonlinear mapping ability of deep learning, in this study, we present a novel magnetotelluric inversion method based on fully convolutional networks. This approach directly builds an end-to-end mapping from apparent resistivity and phase data to resistivity anomaly model. The implementation of the proposed method contains two stages: training and testing. During the training stage, the weight sharing mechanism of fully convolutional network is considered, and only the single anomalous body model samples are used for training, which greatly shortens the modeling time and reduces the difficulty of network training. After that, the unknown combinatorial anomaly model can be reconstructed from the magnetotelluric data using the trained network. The proposed method is tested in both synthetic and field data. The results show that the deep learning-based inversion method proposed in this paper is computationally efficient and has high imaging accuracy.

8

Vehicle classification using the convolution neural network approach

Trivedi Janak, Devi Mandalapu Sarada, Dhara Dave

Zeszyty Naukowe. Transport / Politechnika Śląska

|

2021

|

z. 112

201--209

EN

We present vehicle detection classification using the Convolution Neural Network (CNN) of the deep learning approach. The automatic vehicle classification for traffic surveillance video systems is challenging for the Intelligent Transportation System (ITS) to build a smart city. In this article, three different vehicles: bike, car and truck classification are considered for around 3,000 bikes, 6,000 cars, and 2,000 images of trucks. CNN can automatically absorb and extract different vehicle dataset’s different features without a manual selection of features. The accuracy of CNN is measured in terms of the confidence values of the detected object. The highest confidence value is about 0.99 in the case of the bike category vehicle classification. The automatic vehicle classification supports building an electronic toll collection system and identifying emergency vehicles in the traffic.

9

Multi-model hybrid ensemble weighted adaptive approach with decision level fusion for personalized affect recognition based on visual cues

Jadhav Nagesh, Sugandhi Rekha

Bulletin of the Polish Academy of Sciences. Technical Sciences

|

2021

|

Vol. 69, nr 6

art. no. e138819

EN

In the domain of affective computing different emotional expressions play an important role. To convey the emotional state of human emotions, facial expressions or visual cues are used as an important and primary cue. The facial expressions convey humans affective state more convincingly than any other cues. With the advancement in the deep learning techniques, the convolutional neural network (CNN) can be used to automatically extract the features from the visual cues; however variable sized and biased datasets are a vital challenge to be dealt with as far as implementation of deep models is concerned. Also, the dataset used for training the model plays a significant role in the retrieved results. In this paper, we have proposed a multi-model hybrid ensemble weighted adaptive approach with decision level fusion for personalized affect recognition based on the visual cues. We have used a CNN and pre-trained ResNet-50 model for the transfer learning. VGGFace model’s weights are used to initialize weights of ResNet50 for fine-tuning the model. The proposed system shows significant improvement in test accuracy in affective state recognition compared to the singleton CNN model developed from scratch or transfer learned model. The proposed methodology is validated on The Karolinska Directed Emotional Faces (KDEF) dataset with 77.85% accuracy. The obtained results are promising compared to the existing state of the art methods.

10

Grid Search of Convolutional Neural Network model in the case of load forecasting

Tran Thanh Ngoc

Archives of Electrical Engineering

|

2021

|

Vol. 70, nr 1

25--36

EN

The Convolutional Neural Network (CNN) model is one of the most effective models for load forecasting with hyperparameters which can be used not only to determine the CNN structure and but also to train the CNN model. This paper proposes a frame work for Grid Search hyperparameters of the CNN model. In a training process, the optimalmodels will specify conditions that satisfy requirement for minimum of accuracy scoresof Root Mean Square Error (RMSE), Mean Absolute Percentage Error (MAPE) and Mean Absolute Error (MAE). In the testing process, these optimal models will be used to evaluate the results along with all other ones. The results indicated that the optimal models have accuracy scores near the minimum values. Load demand data of Queensland (Australia) and Ho Chi Minh City (Vietnam) were utilized to verify the accuracy and reliability of the Grid Search framework.

11

SACNN: Spatial Adversarial Convolutional Neural Network for Textile Defect Detection

Hou Wei, Tao Xian, Ma Wenzhi, Xu De

Fibres & Textiles in Eastern Europe

|

2020

|

Vol. 28, no. 6 (144)

127--133

EN

Constructing textile defect detection systems is significant for quality control in industrial production, but it is costly and laborious to label sufficient detailed samples. This paper proposes a model called ‘spatial adversarial convolutional neural network’ which tries to solve the problem above by only using the image-level label. It consists of two parts: a feature extractor and feature competition. Firstly, a string of convolutional blocks is used as a feature extractor. After feature extraction, a maximum greedy feature competition is taken among features in the feature layer. The feature competition mechanism can lead the network to converge to the defect location. To evaluate this mechanism, experiments were carried on two datasets. As the training time increases, the model can spontaneously focus on the actual defective location, and is robust towards an unbalanced sample. The classification accuracy of the two datasets can reach more than 98%, and is comparable with the method of labelling samples in detail. Detection results show that defect location from the model is more compact and accurate than in the Grad-CAM method. Experiments show that our model has potential usage in defect detection in an industrial environment.

PL

Konstruowanie systemów wykrywania wad tekstyliów ma duże znaczenie dla kontroli jakości w produkcji przemysłowej, ale etykietowanie wystarczająco szczegółowych próbek jest kosztowne i pracochłonne. W artykule zaproponowano model zwany „przestrzenną przeciwstawną splotową siecią neuronową”, który próbuje rozwiązać powyższy problem jedynie przy użyciu etykiety na poziomie obrazu. Składa się z dwóch części: ekstraktora fabuły i konkursu fabularnego. Po pierwsze, ciąg bloków splotowych jest używany jako ekstraktor cech. Po wyodrębnieniu cech dochodzi do maksymalnej zachłannej konkurencji między funkcjami w warstwie cech. Mechanizm współzawodnictwa cech może doprowadzić do konwergencji sieci do lokalizacji defektu. Aby ocenić ten mechanizm, przeprowadzono eksperymenty na dwóch zbiorach danych. Wraz ze wzrostem czasu szkolenia model może spontanicznie skupić się na rzeczywistej wadliwej lokalizacji i jest odporny na niezrównoważoną próbkę. Dokładność klasyfikacji obu zbiorów danych może sięgać ponad 98% i jest porównywalna ze szczegółową metodą znakowania próbek. Wyniki detekcji pokazują, że lokalizacja defektu z modelu jest bardziej zwarta i dokładna niż w metodzie Grad-CAM. Eksperymenty pokazują, że zaprezentowany model ma potencjalne zastosowanie do wykrywania defektów w środowisku przemysłowym.

12

Spectral entropy and deep convolutional neural network for ECG beat classification

Asgharzadeh-Bonab Akbar, Amirani Mehdi Chehel, Mehri Alaeddin

Biocybernetics and Biomedical Engineering

|

2020

|

Vol. 40, no. 2

691--700

EN

Sudden cardiac death is the result of abnormal heart conditions. Therefore, early detection of such abnormal conditions is vital to identify heart problems. Hence, in this paper, we aim to present a new computer-aided diagnosis (CAD) method based on time-frequency analysis of electrocardiogram (ECG) signals and deep neural networks for arrhythmia detection. Time-frequency transforms have the capability of providing spectral information at different times, which is very useful for analyzing non-stationary signals. On the other side, entropy is an attractive measurement from ECG signals which can distinguish different types of them. In this paper, time-frequency spectral entropy is proposed to extract the efficient features from ECG signals. All computed entropies cannot provide separability among different classes, two-directional two-dimensional principal component analysis (2D2PCA) can be used to reduce the dimension of the extracted features. Finally, the convolutional neural network (CNN) classifies the time-frequency features to diagnose the ECG beat signals and detect arrhythmias. The results show that the spectral entropy can provide good separation between different among ECG beats and the proposed method outperforms the recently introduced method for analyzing ECG signals.

13

Multi-path convolutional neural network in fundus segmentation of blood vessels

Tian Chun, Fang Tao, Fan Yingle, Wu Wei

Biocybernetics and Biomedical Engineering

|

2020

|

Vol. 40, no. 2

583--595

EN

There is a close correlation between retinal vascular status and physical diseases such as eye lesions. Retinal fundus images are an important basis for diagnosing diseases such as diabetes, glaucoma, hypertension, coronary heart disease, etc. Because the thickness of the retinal blood vessels is different, the minimum diameter is only one or two pixels wide, so obtaining accurate measurement results becomes critical and challenging. In this paper, we propose a new method of retinal blood vessel segmentation that is based on a multi-path convolutional neural network, which can be used for computer-based clinical medical image analysis. First, a low-frequency image characterizing the overall characteristics of the retinal blood vessel image and a high-frequency image characterizing the local detailed features are respectively obtained by using a Gaussian low-pass filter and a Gaussian high-pass filter. Then a feature extraction path is constructed for the characteristics of the low- and high-frequency images, respectively. Finally, according to the response results of the low-frequency feature extraction path and the high-frequency feature extraction path, the whole blood vessel perception and local feature information fusion coding are realized, and the final blood vessel segmentation map is obtained. The performance of this method is evaluated and tested by DRIVE and CHASE_DB1. In the experimental results of the DRIVE database, the evaluation indexes accuracy (Acc), sensitivity (SE), and specificity (SP) are 0.9580, 0.8639, and 0.9665, respectively, and the evaluation indexes Acc, SE, and SP of the CHASE_DB1 database are 0.9601, 0.8778, and 0.9680, respectively. In addition, the method proposed in this paper could effectively suppress noise, ensure continuity after blood vessel segmentation, and provide a feasible new idea for intelligent visual perception of medical images.

14

Mediastinal lymph node malignancy detection in computed tomography images using fully convolutional network

Tekchandani Hitesh, Verma Shrish, Londhe Narendra D.

Biocybernetics and Biomedical Engineering

|

2020

|

Vol. 40, no. 1

187--199

EN

Differential diagnosis of malignant and benign mediastinal lymph nodes (LNs) through invasive pathological tests is a complex and painful procedure because of sophisticated anatomical locations of LNs in the chest. The image based automatic machine learning techniques have been attempted in the past for malignancy detection. But these conventional methods suffer from complex selection of hand-crafted features and trade-off between performance parameters due to them. Today deep learning approaches are out-performing conventional machine learning techniques and able to overcome these issues. However, the existing convolutional neural network (CNN) based models also are prone to overfitting because of fully connected (FC) layers. Therefore, in this paper authors have proposed a fully convolutional network (FCN) based deep learning model for lymph nodes malignancy detection in computed tomography (CT) images. Moreover, the proposed FCN has been customized with batch normalization and advanced activation function Leaky ReLU to accelerate the training and to overcome the problem of dying ReLU, respectively. The performance of the proposed FCN has been also tuned to its best for smaller data size using data augmentation methods. The generalization of the proposed model is tested using the network parameter variation. To understand the reliability of the proposed model, it has also been compared with state-of-art related deep learning networks. The proposed FCN model has achieved an average accuracy, sensitivity, specificity, and area under curve as 90.28%, 90.63%, 89.95%, and 0.90, respectively. Our results also confirms the successful usabilility of augmentation methods for working on smaller datasets and deep learning approaches.

15

Detection of lung cancer on chest CT images using minimum redundancy maximum relevance feature selection method with convolutional neural networks

Togaçar Mesut, Ergen Burhan, Cömert Zafer

Biocybernetics and Biomedical Engineering

|

2020

|

Vol. 40, no. 1

23--39

EN

Lung cancer is a disease caused by the involuntary increase of cells in the lung tissue. Early detection of cancerous cells is of vital importance in the lungs providing oxygen to the human body and excretion of carbon dioxide in the body as a result of vital activities. In this study, the detection of lung cancers is realized using LeNet, AlexNet and VGG-16 deep learning models. The experiments were carried out on an open dataset composed of Computed Tomography (CT) images. In the experiment, convolutional neural networks (CNNs) were used for feature extraction and classification purposes. In order to increase the success rate of the classification, the image augmentation techniques, such as cutting, zooming, horizontal turning and filling, were applied to the dataset during the training of the models. Because of the outstanding success of AlexNet model, the features obtained from the last fully-connected layer of the model were separately applied as the input to linear regression (LR), linear discriminant analysis (LDA), decision tree (DT), support vector ma-chine (SVM), k -nearest neighbor (kNN) and softmax classifiers. A combination of AlexNet model and k NN classifier achieved the most efficient classification accuracy as 98.74 %. Then, the minimum redundancy maximum relevance (mRMR) feature selection method was applied to the deep feature set to choose the most efficient features. Consequently, the success rate was yielded as 99.51 % by reclassifying the dataset with the selected features and k NN model. The proposed model is consistent diagnosis model for lung cancer detection using chest CT images.

16

CNN-based superresolution reconstruction of 3D MR images using thick-slice scans

Jurek Jakub, Kocinski Marek, Materka Andrzej, Elgalal Marcin, Majos Agata

Biocybernetics and Biomedical Engineering

|

2020

|

Vol. 40, no. 1

111--125

EN

Due to inherent physical and hardware limitations, 3D MR images are often acquired in the form of orthogonal thick slices, resulting in highly anisotropic voxels. This causes the partial volume effect, which introduces blurring of image details, appearance of staircase artifacts and significantly decreases the diagnostic value of images. To restore high resolution isotropic volumes, we propose to use a convolutional neural network (CNN) driven by patches taken from three orthogonal thick-slice images. To assess the validity and efficiency of this postprocessing approach, we used 1x1x1 mm3-voxel brain images of different modalities, available via the well known BrainWeb database. They served as a high resolution reference and were numerically preprocessed to create input images of different slice thickness and anatomical orientation, for CNN training, validation and testing. The visual quality of reconstructed images was indeed superior, compared to images obtained by fusion of interpolated thick-slice images, or to images reconstructed with the CNN using a single input MR scan. The significant increase of objectively computed figures of merit, e.g. the Structural Similarity Image Metric, was also noticed. Keeping in mind that any single value of such quality metrics represents a number of psychophysical effects, we applied the CNN trained on brain images for superresolution reconstruction of synthetic and acquired blood vessel tree images. We then used the restored superresolution volumes for estimation of vessel radii. It was demonstrated that vessel radius values derived from superresolution images of simulated vessel trees are significantly more accurate than those obtained from a standard fusion of interpolated thick-slice orthogonal scans. Superiority of the CNN-based superresolution images was also observed for scanner-acquired MR scans according to the evaluated parameters. These three experiments show the efficiency of CNN-based image reconstruction for qualitative and quantitative improvement of its diagnostic quality, as well as illustrates the practical usefulness of transfer learning - networks trained on example images of one kind can be used to restore superresolution images of physically different objects.

17

A novel automation-assisted cervical cancer reading method based on convolutional neural network

Xiang Yao, Sun Wanxin, Pan Changli, Yan Meng, Yin Zhihua, Liang Yixiong

Biocybernetics and Biomedical Engineering

|

2020

|

Vol. 40, no. 2

611--623

EN

While automation-assisted reading system can improve efficiency, their performance often relies on the success of accurate cell segmentation and hand-craft feature extrac- tion. This paper presents an efficient and totally segmentation-free method for automat-ed cervical cell screening that utilizes modern object detector to directly detect cervical cells or clumps, without the design of specific hand-crafted feature. Specifically, we use the state-of-the-art CNN-based object detection methods, YOLOv3, as our baseline model. In order to improve the classification performance of hard examples which are four highly similar categories, we cascade an additional task-specific classifier. We also investigate the presence of unreliable annotations and coped with them by smoothing the distribu- tion of noisy labels. We comprehensively evaluate our methods on our test set which is consisted of 1014 annotated cervical cell images with size of 4000 3000 and complex cellular situation corresponding to 10 categories. Our model achieves 97.5% sensitivity (Sens) and 67.8% specificity (Spec) on cervical cell image-level screening. Moreover, we obtain a best mean average precision (mAP) of 63.4% on cervical cell-level diagnosis, and improve the average precision (AP) of hard examples which are the most valuable but most difficult to distinguish. Our automation-assisted cervical cell reading system not only achieves cervical cell image-level classification but also provides more detailed location and category reference information of abnormal cells. The results indicate feasible performance of our method, together with the efficiency and robustness, provid- ing a new idea for future development of computer-assisted reading systems in clinical cervical screening.

18

A new method of diagnosing athlete's anterior cruciate ligament health status using surface electromyography and deep convolutional neural network

Hatamzadeh Mehran, Hassannejad Reza, Sharifnezhad Ali

Biocybernetics and Biomedical Engineering

|

2020

|

Vol. 40, no. 1

65--76

EN

Anterior cruciate ligament (ACL) injury is one of the most common injuries in high-demand sports. Due to long-term treatment of this injury, diagnosing recovery of ACL becomes important, particularly six months postoperatively. The purpose of this research is to provide a cost-effective and intelligent method to diagnose ACL's health status. For this purpose, 11 healthy and 27 ACL-injured subjects have been selected. In the proposed method, the athlete performs a single-leg landing protocol and surface electromyographic signals (EMG) are taken from eight lower limb muscles. Then, time–frequency distributions of EMG signals in each landing are calculated as an image, using pseudo Wigner–Ville distribution (PWVD), which are the inputs of a deep convolutional neural network (DCNN). By time–frequency analysis, it has been made clear that any change in ACL's health status causes changes in the extent of energy spread in PWVD, distribution volume, frequency content, damping rate and the peak value of EMG signals. In this research, a new relationship between ACL's health status and lower limb muscles activity is introduced through monitoring of PWVD images. The result indicates that the designed expert system is able to diagnose ACL's health status with 95.8% accuracy. In this non-invasive method, PWVD images of EMG signals are chosen as the inputs of DCNN, instead of MRI images, which, in addition to their high accuracy in diagnosing, are safer and much cheaper. The presented method can play an important role in assessing the recovery process, six months postoperatively and after that.

19

A fully automatic computer-aided diagnosis system for hepatocellular carcinoma using convolutional neural networks

Li Jing, Wu Yurun, Shen Nanyan, Zhang Jiawen, Chen Enlong, Sun Jie, Deng Zongqian, Zhang Yuchen

Biocybernetics and Biomedical Engineering

|

2020

|

Vol. 40, no. 1

238--248

EN

The cancer of liver, which is the leading cause of cancer death, is commonly diagnosed by comparing the changes of gray level of liver tissue in the different phases of the patient's CT images. To aid the doctor in reducing misdiagnosis or missed diagnosis, a fully automatic computer-aided diagnosis (CAD) system is proposed to diagnose hepatocellular carcinoma (HCC) using convolutional neural network (CNN) classifier. The automatic segmentation and classification are two core technologies of the proposed CAD system, which are both realized based on CNN. The segmentation of liver and tumor is implemented by a fully convolutional networks (FCN) based on a fine tuning VGG-16 model with two additional 'skip structures' using a weighted loss function which helps to solve the problem of inaccurate tumor segmentation caused by the inevitably unbalanced training data. HCC classification is implemented by a 9-layer CNN classifier, whose input is a 4-channel image data constructed by combining the segmentation result of FCN with the original CT image. A total of 165 venous phase CT images including 46 diffuse tumors, 43 nodular tumors, and 76 massive tumors are used to evaluate the performance of the proposed CAD system. The classification accuracy of CNN classifier for diffuse, nodular and massive tumors are 98.4%, 99.7% and 98.7% respectively, which are significantly improved in contrast with the traditional feature-based ANN and SVM classifiers. The proposed CAD system, which is unaffected by the difference of preprocessing method and feature type, is proved satisfactory and feasible by the test set.

20

Сross platform tools for modeling and recognition of the fingerspelling alphabet of gesture language

Kondratiuk Serhii S., Krak Iurii V., Wójcik Waldemar

Informatyka, Automatyka, Pomiary w Gospodarce i Ochronie Środowiska

|

2019

|

T. 9, nr 2

24--27

EN

A solution for the problems of the finger spelling alphabet of gesture language modelling and recognition based on cross-platform technologies is proposed. Modelling and recognition performance can be flexible and adjusted, based on the hardware it operates or based on the availability of an internet connection. The proposed approach tunes the complexity of the 3D hand model based on the CPU type, amount of available memory and internet connection speed. Sign recognition is also performed using cross-platform technologies and the tradeoff in model size and performance can be adjusted. the methods of convolutional neural networks are used as tools for gestures of alphabet recognition. For the gesture recognition experiment, a dataset of 50,000 images was collected, with 50 different hands recorded, with almost 1,000 images per each person. The experimental researches demonstrated the effectiveness of proposed approaches.

PL

Zaproponowano rozwiązanie problemów z alfabetem daktylograficznym w modelowaniu języka gestów i rozpoznawaniu znaków w oparciu o technologie wieloplatformowe. Wydajność modelowania i rozpoznawania może być elastyczna i dostosowana, w zależności od wykorzystywanego sprzętu lub dostępności łącza internetowego. Proponowane podejście dostosowuje złożoność modelu 3D dłoni w zależności od typu procesora, ilości dostępnej pamięci i szybkości połączenia internetowego. Rozpoznawanie znaków odbywa się również z wykorzystaniem technologii międzyplatformowych, a kompromis w zakresie wielkości modelu i wydajności może być dostosowany. Jako narzędzia do rozpoznawania gestów alfabetu wykorzystywane są metody konwolucyjnych sieci neuronowych. Na potrzeby eksperymentu rozpoznawania gestów zebrano zbiór danych obejmujący 50 000 obrazów, przy czym zarejestrowano 50 różnych rąk, a na każdą osobę przypadało prawie 1000 obrazów. Badania eksperymentalne wykazały skuteczność proponowanego podejścia.