研究者詳細 - 神谷　亨

論文 - 神谷　亨

分割表示757 件中 81 - 100 件目／全件表示 >>

Preface for ACPR 2023 Proceedings 査読有り国際誌

Liu C.L., Yagi Y., Kamiya T., Blumenstein M., Lu H., Yang W., Cho S.B.

Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics) 14408 LNCS v - vi 2023年01月

Remote Sensing Image Registration Based on Improved Geometric-Matching CNN 査読有り国際誌

Morishima F., Lu H., Kamiya T.

International Conference on Control, Automation and Systems 1745 - 1748 2023年01月

担当区分：最終著者,　責任著者記述言語：英語掲載種別：研究論文（国際会議プロシーディングス）

Environmental change detection is one of the uses of satellite images. This process is performed by subtracting image pairs obtained by different time series or sensors. Therefore, image registration is an important pre-processing step in detection of environmental changes. Currently, image registration methods based on deep learning are gaining attention. In general, higher satellite image resolution results in more accurate registration. However, the increase in image size leads to higher computational costs during training and estimation of deep learning models. Then, we propose a method that reduces the number of parameters of the model to lower the computational cost while maintaining the accuracy. This method makes it easier to handle high-resolution satellite images. The proposed method modified the GMCNN (Geometric-matching CNN) architecture by adding CSA (Cosine Similarity Attention) and SE (Squeeze-and-Excitation) layers to enhance the feature map, and point-wise convolution to reduce the number of parameters. The improved GMCNN decreases the grid MSE by 0.0037 compared to the conventional GMCNN. It also reduces the number of parameters by 49.6%.

DOI： 10.23919/ICCAS59377.2023.10316818

Scopus

その他リンク： https://www.scopus.com/inward/record.uri?partnerID=HzOxMe3b&scp=85179178272&origin=inward

Recognition of Specific Parts of Plastic Bottles Using Improved DeepLab v3+ 査読有り国際誌

Ideta D., Kamiya T.

International Conference on Control, Automation and Systems 1758 - 1761 2023年01月

担当区分：最終著者,　責任著者記述言語：英語掲載種別：研究論文（国際会議プロシーディングス）

In this paper, we focus on the manpower shortage in factories and conduct an experiment to try to automate the process. Factory automation has many advantages, such as reducing labor costs and improving production efficiency. Among them, we focused on the sorting of plastic bottles at a waste disposal plant. Currently, garbage is sent to the landfill without being sorted, and it is sorted by hand. Unlike cans, plastic bottles cannot be easily automated because they cannot be separated using magnets. Therefore, we propose an image processing technique to recognize where plastic bottles are located, and to use a robot arm to sort them, thereby realizing automation. In this paper, we focus on image processing technique based on deep learning. As basic research, we conducted an experiment to see how well the robot can identify a single plastic bottle in an image. We attempted to use semantic segmentation methods for detection, using DeepLabv3+ as the basic model, and improved it. scSE Block and fine-tuning were introduced to improve the accuracy significantly compared to previous studies.

DOI： 10.23919/ICCAS59377.2023.10316854

Scopus

その他リンク： https://www.scopus.com/inward/record.uri?partnerID=HzOxMe3b&scp=85179179143&origin=inward

Identification of Root Resorption on Panoramic X-ray Image Based on EfficientNet 査読有り国際誌

Sakata K., Kamiya T., Oda M., Morimoto Y.

International Conference on Control, Automation and Systems 1749 - 1752 2023年01月

担当区分：責任著者記述言語：英語掲載種別：研究論文（国際会議プロシーディングス）

Root resorption is a disease in which the cells of the root of the tooth are destroyed. This disease may be harmless, depending on the cause. When the root resorption has progressed, early detection and treatment is necessary. If root resorption is left untreated, the lesion can progress to pulp exposure, necrosis, sinus and periodontal defects, and perforation. Diagnosis of root resorption is made by clinical examination and a panoramic x-ray machine, however accurate diagnosis requires computed tomography (CT). CT allows full-plane examination of the area of interest and provides an accurate diagnosis. However, CT is less widely used than panoramic radiography in private dental office. Therefore, there is a need for a method to detect root resorption from panoramic radiographs with a high reliability comparable to that of CT images. In this paper, we propose an image analysis method to identify the presence or absence of root resorption from panoramic radiographs. In particular, we propose a method that uses EfficientNet to extract features with fine-tuning and learns them using L2-softmax loss as an error function. We applied the proposed method to panoramic X-ray images of 59 actual cases and obtained TPR: 71% and FPR: 43% respectively.

DOI： 10.23919/ICCAS59377.2023.10316940

Scopus

その他リンク： https://www.scopus.com/inward/record.uri?partnerID=HzOxMe3b&scp=85179183951&origin=inward

Identification of Nodular Shadows from CT Images Using Improved CoAtNet Incorporated Clinical Recording 査読有り国際誌

Nishitaki Y., Kamiya T., Kido S.

International Conference on Control, Automation and Systems 1727 - 1732 2023年01月

担当区分：責任著者記述言語：英語掲載種別：研究論文（国際会議プロシーディングス）

When lung cancer is suspected, a CT scan of the chest is widely used as a method of precise examination. However, the number of CT images obtained in a single examination is enormous, placing a heavy burden on the physician who reads the images. In addition, there is concern about the possibility of undetected lesions due to differences in the skills and experience of the reader. Therefore, computer-aided diagnosis systems have been introduced to reduce their workload and undetected lesions. When physicians make a diagnosis, they consider not only CT images but also information about the patient. Therefore, attempts are being made to improve the accuracy of diagnosis by mimicking this process using artificial intelligence. In this paper, we propose a model for identifying nodular shadows using deep learning, aiming to improve the accuracy of diagnosis by introducing medical record information in addition to image information. Normal tissue is divided into three classes: branched vessels, thin vessels, and round vessels, and a total of four classes are classified, including abnormal tissue. Experimental results show that the accuracy of nodular shade discrimination is improved when medical record information is added.

DOI： 10.23919/ICCAS59377.2023.10316787

Scopus

その他リンク： https://www.scopus.com/inward/record.uri?partnerID=HzOxMe3b&scp=85179182295&origin=inward

Facial Symmetry Analysis for Cleft Lip Patients from 4D Point Cloud Data 査読有り国際誌

Kihara N., Kimura-Nomoto N., Okawachi T., Li G（天津工大）., Nakamura N., Kamiya T.

International Conference on Control, Automation and Systems 1502 - 1505 2023年01月

担当区分：最終著者,　責任著者記述言語：英語掲載種別：研究論文（国際会議プロシーディングス）

Cleft lip is one of the most common birth defects. Several operations are performed to form a natural lip. A problem with these operations is that the criteria for the facial symmetry are unclear. Based on this background, we propose a method for evaluating the facial symmetry using 4D point cloud data. We evaluate the facial symmetry in two ways. One is based on the temporal changes in the face landmarks. Corresponding points are searched using two local geometric descriptors. The results are approximated to the 3D lines and these slopes are compared to obtain the left-right difference of the movement. The other is based on the center of gravity. The lip part is extracted from the point cloud and divided into subregions. The centroid coordinates are then calculated for each subregion and compared to obtain the left-right difference of the facial structure. The experiment is performed using artificially generated 4D data. As an experimental result, it is shown that our method can find point correspondences with smaller error than comparative methods.

DOI： 10.23919/ICCAS59377.2023.10316967

Scopus

その他リンク： https://www.scopus.com/inward/record.uri?partnerID=HzOxMe3b&scp=85179178953&origin=inward

Extraction of Lung Tumor Regions from Thoracic CT Images Using An Improved U-Net 査読有り国際誌

Takahashi R., Kamiya T., Terasawa T., Aoki T.

International Conference on Control, Automation and Systems 1489 - 1493 2023年01月

担当区分：責任著者記述言語：英語掲載種別：研究論文（国際会議プロシーディングス）

Lung cancer currently kills the largest number of people in the world, with lung cancer accounting for the highest number of deaths among men and women. Therefore, early detection and treatment are important issues to reduce the number of lung cancer deaths. Genetic testing can confirm the presence or absence of driver gene mutations involved in cancer cell growth and other factors. In the process of identifying the presence or absence of driver gene mutations, lung cancer regions are extracted manually in cooperation with a radiologist. One of the concerns is that manual lung cancer extraction places a heavy burden on the physician who reads the images. Therefore, we propose an automatic extraction of lung tumor regions from chest CT images using deep learning, with the goal of developing a CAD system that reduces the workload of the physician and prevents lesions from being overlooked. We constructed a new model based on U-Net with the addition of CBAM and MultiRes Block. Experimental results on the model using CT images of the chest showed a 4.2% improvement in accuracy for Dice and a 4.6% improvement for IoU compared to the original U-Net model.

DOI： 10.23919/ICCAS59377.2023.10316783

Scopus

その他リンク： https://www.scopus.com/inward/record.uri?partnerID=HzOxMe3b&scp=85179183296&origin=inward

Classification of Histological Types of Primary Lung Cancer from CT Images Using Clinical Information 査読有り国際誌

Honda N., Kamiya T., Kido S.

International Conference on Control, Automation and Systems 1753 - 1757 2023年01月

担当区分：責任著者記述言語：英語掲載種別：研究論文（国際会議プロシーディングス）

Identification of primary lung cancer is very important because it influences the course of treatment, especially for small cell carcinomas, which metastasize rapidly and must be detected at an early stage. In addition to imaging, clinical information is often used in CAD (computer aided diagnosis) systems. In addition to images, clinical information is often used in CAD systems, especially information on smoking history, which is considered to be important in the diagnosis of lung cancer. In this paper, we propose a method to identify primary lung cancer by adding clinical information from medical records in addition to images in order to improve the accuracy of diagnosis. We use tumor images surrounded by rectangular regions from CT images in an open dataset as input images and train the method by deep learning. We evaluate the proposed method by discriminating tumors from unknown data. In our experiments, we found that the accuracy was improved by about 5% when clinical information was added to 655 images, which included four classes of cancer: adenocarcinoma, small cell carcinoma, squamous cell carcinoma, and large cell carcinoma.

DOI： 10.23919/ICCAS59377.2023.10316865

Scopus

その他リンク： https://www.scopus.com/inward/record.uri?partnerID=HzOxMe3b&scp=85179178408&origin=inward

Classification of Driver Gene Mutations from 3D-CT Images Based on Radiomics Features 査読有り国際誌

Watanabe S., Kamiya T., Terasawa T., Aoki T.

International Conference on Control, Automation and Systems 1733 - 1736 2023年01月

担当区分：責任著者記述言語：英語掲載種別：研究論文（国際会議プロシーディングス）

Cancer is a leading cause of death worldwide, accounting for nearly 10 million deaths in 2020. Lung cancer is the most common cause of death from this cancer. Diagnosis is made mainly by biopsy, which checks for mutations in the driver genes of lung cancer. If there is a mutation, molecularly targeted drugs with higher therapeutic efficacy can be used. However, it is difficult for physicians to make a decision and places a heavy burden on patients. To solve this problem, a computer aided diagnosis (CAD) system is needed to identify the presence or absence of driver gene mutations from CT images. In this paper, Radiomics is used to extract features from 3D lung cancer in CT images and analyze them using machine learning to classify the presence or absence of mutations. Because the features obtained is huge, dimensionality reduction is performed using Null Importance. Furthermore, the accuracy is improved by adding gender as clinical information. The proposed method was applied to a dataset consisting of 175 cases. As a result, we obtained an AUC (Area Under the Curve) of 0.985, accuracy of 92.0%, true positive rate of 85.7%, and false positive rate of 2.20%.

DOI： 10.23919/ICCAS59377.2023.10316810

Scopus

その他リンク： https://www.scopus.com/inward/record.uri?partnerID=HzOxMe3b&scp=85179182718&origin=inward

Automatic Segmentation of Finger Bone Regions from CR Image Using the Improved HRNet 査読有り国際誌

Hiraoka T., Kamiya T., Aoki T.

International Conference on Control, Automation and Systems 1498 - 1501 2023年01月

担当区分：責任著者記述言語：英語掲載種別：研究論文（国際会議プロシーディングス）

In Japan, it is serious problems such as increasing nursing care of handicapped people requiring and the aging of caregivers which is caused in a super-aging society. One of the main causes of being classified as persons requiring nursing care include joint diseases such as rheumatoid arthritis and bone fractures due to osteoporosis. Early detection and treatment of these diseases are considered important. Rheumatoid arthritis and osteoporosis are generally diagnosed by simple X-ray examination. However, there are problems with radiographic diagnosis by physicians, such as lack of objectivity and reproducibility of diagnosis, and increased workload on the radiologists. To solve these problems, a Computer-Aided Diagnosis (CAD) system is being developed. Because the CAD system may use the results of quantitative computer analysis, it is expected to improve the reproducibility and accuracy of diagnosis and reduce the burden on physicians. Therefore, this paper proposes a segmentation method of phalanx region from CR images to develop a CAD system. The proposed method is based on HRNet + JPU + U-Net. The proposed method was applied to 101 cases of X-ray images, and mIoU=0.897 was obtained. Experiments for segmentation from images confirmed the usefulness of the proposed method by improving the extraction accuracy of the boundary of the phalanx region.

DOI： 10.23919/ICCAS59377.2023.10317032

Scopus

その他リンク： https://www.scopus.com/inward/record.uri?partnerID=HzOxMe3b&scp=85179183712&origin=inward

An Image Registration Technique for Brain MR Images Using Linear Transform by 3DCNN 査読有り国際誌

Baba S., Kamiya T.

International Conference on Control, Automation and Systems 1762 - 1765 2023年01月

担当区分：最終著者,　責任著者記述言語：英語掲載種別：研究論文（国際会議プロシーディングス）

In this paper, we focus on brain atrophy in Alzheimer's disease and propose a 3D image registration method of brain MR images for the purpose of calculating the atrophy rate by temporal subtraction technique. The nonlinear 3D image registration model of Stergios et al. is recombined into a linear image registration model, and image registration is performed with isotropic affine and rigid body transformations. In addition, the accuracy is improved by the Instance-specific Optimization, which performs the optimization process for each case. For the training of the model, we used pseudo-data obtained by arbitrarily linear transformation of the ADNI brain MR images. Synthetic data and actual ADNI time series data (AD, MCI) were used for the test. The experimental results show that the addition of the optimization process improves the normalized cross-correlation (NCC) by more than 0.1. The loss function of the model is compared between Grid loss and MSE, and it is confirmed that MSE is effective for NCC. Furthermore, the output images confirm the usefulness of the image registration of the skull in the linear transformation.

DOI： 10.23919/ICCAS59377.2023.10316755

Scopus

その他リンク： https://www.scopus.com/inward/record.uri?partnerID=HzOxMe3b&scp=85179178497&origin=inward

An Denoising Method Low-Dose CT Image Using Image Restoration Convolutional Neural Network 査読有り国際誌

Sadamatsu Y., Murakami S., Guangxu L（天津工大）., Kamiya T.

International Conference on Control, Automation and Systems 1737 - 1740 2023年01月

担当区分：最終著者,　責任著者記述言語：英語掲載種別：研究論文（国際会議プロシーディングス）

Radiation is widely used in medicine to diagnose and treat disease. CT (Computed Tomography) scans allow early detection of externally invisible diseases and appropriate treatment. However, radiation exposure from the examination may result in a future risk of cancer. Efforts are therefore being made to reduce radiation exposure. During the examination, noise is generated in the image when the dose is reduced. Noise reduces the visibility of the image and may cause lesions to be missed. In this study, we focus on Convolutional Neural Networks (CNNs), a type of deep learning model that has recorded high accuracy in image processing tasks. The proportion of frequency components in an image has more low-frequency components and fewer high-frequency components. Since image features are treated equally across channels, important information such as noise and edges are easily lost. To solve this problem, we propose CNN with channel attention module. In addition, we employ MAE as the loss function to enable effective learning. Using whole body slice CT images of pigs, we evaluate the image quality by Peak Signal-to-Noise Ratio (PSNR) and show that the proposed method is effective.

DOI： 10.23919/ICCAS59377.2023.10317050

Scopus

その他リンク： https://www.scopus.com/inward/record.uri?partnerID=HzOxMe3b&scp=85179178606&origin=inward

A Method for Denoising from Low-Dose CT Images Based on Noise2Noise 査読有り国際誌

Sawada S., Murakami S., Guangxu L., Kamiya T.

International Conference on Control, Automation and Systems 1494 - 1497 2023年01月

担当区分：最終著者,　責任著者記述言語：英語掲載種別：研究論文（国際会議プロシーディングス）

Japan has the highest number of CT scanners per million people among developed countries, and the annual medical radiation dose is about six times higher than the world average. Screening of high-risk populations by means of low-dose CT has been verified to be able to reduce lung cancer mortality. However, the low-dose CT image noise should be accurately estimated in order to achieve good image quality as normal dose CT. In this paper, we apply a denoising method using convolutional neural networks without use of the clean images during learning phase. The CNN model is based on SRResNet, which has achieved high performance in super-resolution tasks, and has two attention mechanisms, Spatial Attention and Channel Attention, in the Residual Block. Experiments using whole body slice CT images of pigs showed that the proposed method is useful by comparison with normal dose CT images and by evaluation of image quality using peak signal-to-noise ratio (PSNR).

DOI： 10.23919/ICCAS59377.2023.10317004

Scopus

その他リンク： https://www.scopus.com/inward/record.uri?partnerID=HzOxMe3b&scp=85179178164&origin=inward

A Detection Method for Nodular Shadows from Temporal Subtraction Images Using A Machine Learning Technique Incorporated Radiomics Features 査読有り国際誌

Baba N., Kamiya T., Terasawa T., Aoki T., Kido S.

International Conference on Control, Automation and Systems 1741 - 1744 2023年01月

担当区分：最終著者,　責任著者記述言語：英語掲載種別：研究論文（国際会議プロシーディングス）

Lung cancer progresses rapidly, and early detection and treatment are important. Computed tomography (CT) equipment is mainly used for the examination. However, the number of CT images is huge and places a heavy burden on physicians. Therefore, a computer aided diagnosis (CAD) system is necessary to overcome these problem. One of the CAD is temporal subtraction technique. The technique is that performs a subtraction operation using the current and past images of the same patient to remove normal structures and emphasize newly appearing areas that have changed over time. However, because two images taken at different dates are used, artifacts due to misalignment often occur. Therefore, it is necessary to classify partial images with high false-positive rates and enhanced temporal changes into lesions and normal tissue. In this paper, the temporal subtraction image generation method is used to extract candidate areas of abnormal shadows as initial shadows. The proposed method was applied to 22 cases of chest CT images, and the results showed that the proposed method achieved a discrimination accuracy of Accuracy=74.54, TPR=72.65, FPR=22.59, and AUC=79.16 in XGBoost, which is a kind of machine learning method.

DOI： 10.23919/ICCAS59377.2023.10316767

Scopus

その他リンク： https://www.scopus.com/inward/record.uri?partnerID=HzOxMe3b&scp=85179180349&origin=inward

機械学習による胸部CT画像からのドライバー遺伝子情報変異有無の識別法

吉福，神谷，寺澤，青木

バイオメディカル・ファジィ・システム学会年次大会講演論文集 2022年12月

Recognition of sidewalk environment based on WideSegPlus 査読有り国際誌

Sakai, Lu, Yujie Li, Kamiya

Proceedings Volume 12508, International Symposium on Artificial Intelligence and Robotics 2022 ( SPIE ) 2022年12月

Grasping position estimation method using depth image for thin objects 査読有り国際誌

Yoshihara, Koga, Lu, Kamiya

Proceedings Volume 12508, International Symposium on Artificial Intelligence and Robotics 2022 ( SPIE ) 2022年12月

Underwater image super-resolution using improved SRCNN 査読有り国際誌

Horimachi, Lu, Zheng, Kamiya, Nakatoh, Serikawa

Proceedings Volume 12508, International Symposium on Artificial Intelligence and Robotics 2022 ( SPIE ) 2022年12月

Underwater video networking and targe tracking 査読有り国際誌

Ota, Lu, Jianru Li, Yuchao Zheng, Kamiya, Nakatoh, Serikawa

Proceedings Volume 12508, International Symposium on Artificial Intelligence and Robotics 2022 2022年12月

Automatic Classification of Respiratory Sound Considering Hierarchical Structure 査読有り国際誌

Tabata M., Lu H., Kamiya T., Mabu S., Kido S.

International Conference on Control, Automation and Systems 2022-November 537 - 541 2022年11月

担当区分：最終著者,　責任著者記述言語：英語掲載種別：研究論文（国際会議プロシーディングス）

Respiratory diseases are one of the leading causes of death worldwide. Approximately 8 million people die annually from respiratory diseases. Diagnosis is made primarily by auscultation using a stethoscope. The lack of quantitative criteria makes diagnosis difficult in the field where physicians are in short supply. To solve this problem, a computer aided diagnosis (CAD) system that quantitatively analyzes and classifies respiratory sounds and outputs them as a 'second opinion' is needed. In this paper, HPSS (Harmonious/Percussive Sound Separation) is used to separate abnormal respiratory sound features. Images are generated from the spectral envelopes obtained by linear prediction coefficients (LPC) for each of the three types of respiratory sound data before separation. The CNN (convolutional neural networks) framework based on hierarchical structure of the correct labels is introduced. The proposed method was applied to the dataset used in the International Conference on Biomedical and Health Informatics (ICBHI) 2017 Challenge. As a result, we obtained a sensitivity of 63.5%, specificity of 85.1%, average score of 74.3%, harmonic score of 72.7%, area under the curve of 87.8%, and false negative rate of24.5%, respectively.

DOI： 10.23919/ICCAS55662.2022.10003771

Scopus

その他リンク： https://www.scopus.com/inward/record.uri?partnerID=HzOxMe3b&scp=85146537757&origin=inward

　前のページ - 次のページ　

<神谷 亨>

論文 - 神谷 亨

<神谷　亨>

論文 - 神谷　亨