研究者詳細 - 小田　昌宏

論文 - 小田　昌宏

分割表示 345 件中 81 - 100 件目／全件表示 >>

Comprehensive Diagnostic Performance of Real-Time Characterization of Colorectal Lesions Using an Artificial Intelligence–Assisted System: A Prospective Study 査読有り

Minegishi Y., Kudo S.E., Miyata Y., Nemoto T., Mori K., Misawa M., Mori Y., Mochida K., Akimoto Y., Ishiyama M., Ogura Y., Abe M., Sato Y., Ogawa Y., Yasuharu M., Tanaka K., Ichimasa K., Nakamura H., Ogata N., Hisayuki T., Kudo T., Hayashi T., Wakamura K., Miyachi H., Baba T., Ishida F., Itoh H., Oda M.

Gastroenterology 163 巻 ( 1 ) 頁： 323 - 325.e3 2022年7月

Blood Vessel Segmentation from Low-Contrast and Wide-Field Optical Microscopic Images of Cranial Window by Attention-Gate-Based Network 査読有り

Wu Yunheng, Oda Masahiro, Hayashi Yuichiro, Takebe Takanori, Nagata Shogo, Wang Cheng, Mori Kensaku

2022 IEEE/CVF CONFERENCE ON COMPUTER VISION AND PATTERN RECOGNITION WORKSHOPS (CVPRW 2022) 2022-June 巻頁： 1863 - 1872 2022年6月

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）出版者・発行元：IEEE Computer Society Conference on Computer Vision and Pattern Recognition Workshops

The stereomicroscope, which is an optical microscope, is used to observe the organoids cultured in cranial windows. A cranial window is a light accessible observation window made on the brain of mice through craniotomy. Organoids research is often conducted on cranial windows. Hence, the observation of blood vessels in them is important for organoid research, like organoid vascularization. Therefore, achieving a simple, low-cost method that extracts blood vessel structures would significantly help researchers observe the blood vessels in cranial windows from microscopic images. However, wide-field optical microscopic images taken by stereomicroscope suffer from low contrast and dura mater occlusion, complicating the observation of the blood vessels in such images. To address such problems and assist researchers who are observing vascular structures, we propose a method that segments blood vessels in cranial windows from low-contrast and wide-field microscopic images. Our method is based on the Attention U-Net framework and clDice, which considers the connectivity of blood vessels. In addition, for low-contrast and partial occlusion problems, we used contrast enhancement and dehazing as preprocessing steps. Our method achieved a Dice score of 75.56%, a clDice score of 79.95%, and the Accuracy of 91.41% on our microscopic image dataset, suggesting that our method can extract blood vessels from low-contrast and wide-field microscopic images better than other methods.

DOI： 10.1109/CVPRW56347.2022.00203

BONE MARROW EDEMA SCORE IN HAND X-RAY FILM BY AI DEEP LEARNING ASSOCIATE WITH MRI BONE EDEMA IN RHEUMATOID ARTHRITIS.

Katayama K., Pan D., Oda M., Okubo T., Mori K.

ANNALS OF THE RHEUMATIC DISEASES 81 巻頁： 1773 - 1774 2022年6月

JACLS ALL-02 SR protocol reduced-intensity chemotherapy produces excellent outcomes in patients with low-risk childhood acute lymphoblastic leukemia

Takahashi Y., Ishida H., Imamura T., Tamefusa K., Suenobu S., Usami I., Yumura-Yagi K., Hasegawa D., Nishimura S., Suzuki N., Hashii Y., Deguchi T., Moriya-Saito A., Kosaka Y., Kato K., Kobayashi R., Kawasaki H., Hori H., Sato A., Kudo T., Nakahata T., Oda M., Hara J., Horibe K.

International Journal of Hematology 115 巻 ( 6 ) 頁： 890 - 897 2022年6月

出版者・発行元：International Journal of Hematology

Acute lymphoblastic leukemia (ALL) is the most common childhood cancer. As overall cure rates of childhood ALL have improved, reduction of overall treatment intensity while still ensuring excellent outcomes is imperative for low-risk patients. We report the outcomes of patients treated following the standard-risk protocol from the prospective Japan Association of Childhood Leukemia Study (JACLS) ALL-02 study, which was conducted between 2002 and 2008 for patients with newly diagnosed ALL aged 1–18 years. Of 1138 patients with B-cell precursor ALL, 388 (34.1%) were allocated to this protocol. Excellent outcomes were achieved despite the overall treatment intensity being lower than that of most contemporary protocols: 4 years event-free survival (EFS) was 92.3% and 4 years overall survival 98.2%. Patients with high hyperdiploidy (HHD) involving triple trisomy (trisomy of chromosomes 4, 10, and 17) or ETV6-RUNX1 had even better outcomes (4 years EFS 97.6% and 100%, respectively). Unique characteristics of this protocol include a selection of low-risk patients with a low initial WBC count and good early treatment response and reduction of cumulative doses of chemotherapeutic agents while maintaining dose density. In Japan, we are currently investigating the feasibility of this protocol while incorporating minimal residual disease into the patient stratification strategy.

DOI： 10.1007/s12185-022-03315-x

SR-CycleGAN: super-resolution of clinical CT to micro-CT level with multi-modality super-resolution loss. 査読有り

Zheng T, Oda H, Hayashi Y, Moriya T, Nakamura S, Mori M, Takabatake H, Natori H, Oda M, Mori K

Journal of Medical Imaging 9 巻 ( 2 ) 頁： 024003 2022年3月

A cascaded fully convolutional network framework for dilated pancreatic duct segmentation 査読有り国際共著

Shen Chen, Roth Holger R., Hayashi Yuichiro, Oda Masahiro, Miyamoto Tadaaki, Sato Gen, Mori Kensaku

International Journal of Computer Assisted Radiology and Surgery 17 巻 ( 2 ) 頁： 343 - 354 2022年2月

記述言語：英語掲載種別：研究論文（学術雑誌）出版者・発行元：International Journal of Computer Assisted Radiology and Surgery

Purpose: Pancreatic duct dilation can be considered an early sign of pancreatic ductal adenocarcinoma (PDAC). However, there is little existing research focused on dilated pancreatic duct segmentation as a potential screening tool for people without PDAC. Dilated pancreatic duct segmentation is difficult due to the lack of readily available labeled data and strong voxel imbalance between the pancreatic duct region and other regions. To overcome these challenges, we propose a two-step approach for dilated pancreatic duct segmentation from abdominal computed tomography (CT) volumes using fully convolutional networks (FCNs). Methods: Our framework segments the pancreatic duct in a cascaded manner. The pancreatic duct occupies a tiny portion of abdominal CT volumes. Therefore, to concentrate on the pancreas regions, we use a public pancreas dataset to train an FCN to generate an ROI covering the pancreas and use a 3D U-Net-like FCN for coarse pancreas segmentation. To further improve the dilated pancreatic duct segmentation, we deploy a skip connection on each corresponding resolution level and an attention mechanism in the bottleneck layer. Moreover, we introduce a combined loss function based on Dice loss and Focal loss. Random data augmentation is adopted throughout the experiments to improve the generalizability of the model. Results: We manually created a dilated pancreatic duct dataset with semi-automated annotation tools. Experimental results showed that our proposed framework is practical for dilated pancreatic duct segmentation. The average Dice score and sensitivity were 49.9% and 51.9%, respectively. These results show the potential of our approach as a clinical screening tool. Conclusions: We investigate an automated framework for dilated pancreatic duct segmentation. The cascade strategy effectively improved the segmentation performance of the pancreatic duct. Our modifications to the FCNs together with random data augmentation and the proposed combined loss function facilitate automated segmentation.

DOI： 10.1007/s11548-021-02530-x

Real-time Esophagus Achalasia Detection Method for Esophagoscopy Assistance 査読有り

Jiang Kai, Oda Masahiro, Shiwaku Hironari, Misawa Masashi, Mori Kensaku

MEDICAL IMAGING 2022: COMPUTER-AIDED DIAGNOSIS 12033 巻 2022年2月

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）出版者・発行元：Progress in Biomedical Optics and Imaging - Proceedings of SPIE

This paper presents an automated real-time esophagus achalasia (achalasia) detection method for esophagoscopy assistance. Achalasia is a well-recognized primary esophageal motor disorder of unknown etiology. To diagnose the achalasia, endoscopic evaluation of the esophagus and stomach is recommended to ensure that there is not a malignancy causing the disease or esophageal squamous cell carcinoma complicating achalasia. However, esophagoscopy is low sensitive in the early-stage of achalasia, only about half of patients with early-stage achalasia can be identified. Thus, a quantitative detection system of real-time esophagoscopy video is required for diagnosis assistance of achalasia. This paper presents to use of a convolutional neural network (CNN) to detect all achalasia frames in esophagoscopy videos. The features of achalasia cannot be easily distinguished. To better extract features from esophagoscopy frames, we introduce dense pooling connections and dilated convolutions in the CNN. We trained and evaluated our network with an original dataset that is extracted from several esophagoscopy videos of achalasia patients. Furthermore, we develop a real-time achalasia detection ComputerAided Diagnosis (CAD) system with the trained network. The CAD system can detect each frame from the input esophagoscopy videos with only 0.1 milliseconds delay. The real-time achalasia detection system achieved 0.872 accuracy, and 0.943 AUC score on our dataset.

DOI： 10.1117/12.2613289

Uncertainty meets 3D-spatial feature in colonoscopic polyp-size determination 査読有り

Itoh Hayato, Oda Masahiro, Jiang Kai, Mori Yuichi, Misawa Masashi, Kudo Shin-Ei, Imai Kenichiro, Ito Sayo, Hotta Kinichi, Mori Kensaku

Computer Methods in Biomechanics and Biomedical Engineering: Imaging & Visualization 2022年1月

記述言語：英語掲載種別：研究論文（学術雑誌）出版者・発行元：Computer Methods in Biomechanics and Biomedical Engineering: Imaging and Visualization

This paper presents a new 3D-spatial feature extraction from a 2D colonoscopic image for polyp-size estimation. The polyp-size estimation poses potential demands on colonoscopy, since an endoscopist’s subjective estimation is apt to result in uncertain polyp-size determination. This uncertain determination is derived from the lack of 3D-spatial information. Even though a previous work clarified that depth estimation mitigates uncertainty in binary polyp-size classification, precise estimation of 3D structure from a colonoscopic image(s) remains an unsolved challenge in medical image analysis. This work proposes a 3D-spatial feature that expresses a polyp’s precise 3D shape to mitigate uncertainty in polyp-size determination. First, we introduce an accurate depth estimation method to capture the 3D structure of a colon. Next, we integrate depth estimation and polyp localisation to extract a 3D polyp shape as a feature. Finally, we achieve polyp-size estimation by statistical learning of extracted features. The experimental results demonstrated the validity both of our depth estimation and 3D-spatial feature. Compared with deep RGB and RGB-D convolutional neural networks (CNNs), a shallow CNN with the proposed 3D-spatial feature achieved a more accurate polyp-size estimation with a mean absolute error of 1.36 mm, whereas the one of the deep CNN is 3.11 mm.

DOI： 10.1080/21681163.2021.2004445

Impact of the clinical use of artificial intelligence-assisted neoplasia detection for colonoscopy: a large-scale prospective propensity score-matched study (with video) 査読有り

Ishiyama Misaki, Kudo Shin-ei, Misawa Masashi, Mori Yuichi, Maeda Yasuhara, Ichimasa Katsuro, Kudo Toyoki, Hayashi Takemasa, Wakamura Kunihiko, Miyachi Hideyuki, Ishida Fumio, Itoh Hayato, Oda Masahiro, Mori Kensaku

Gastrointestinal Endoscopy 95 巻 ( 1 ) 頁： 155 - 163 2022年1月

記述言語：英語掲載種別：研究論文（学術雑誌）出版者・発行元：Gastrointestinal Endoscopy

Background and Aims: Recently, the use of computer-aided detection (CADe) for colonoscopy has been investigated to improve the adenoma detection rate (ADR). We aimed to assess the efficacy of a regulatory-approved CADe in a large-scale study with high numbers of patients and endoscopists. Methods: This was a propensity score–matched prospective study that took place at a university hospital between July 2020 and December 2020. We recruited patients aged ≥20 years who were scheduled for colonoscopy. Patients with polyposis, inflammatory bowel disease, or incomplete colonoscopy were excluded. We used a regulatory-approved CADe system and conducted a propensity score matching–based comparison of the ADR between patients examined with and without CADe as the primary outcome. Results: During the study period, 2261 patients underwent colonoscopy with the CADe system or routine colonoscopy, and 172 patients were excluded in accordance with the exclusion criteria. Thirty endoscopists (9 nonexperts and 21 experts) were involved in this study. Propensity score matching was conducted using 5 factors, resulting in 1836 patients included in the analysis (918 patients in each group). The ADR was significantly higher in the CADe group than in the control group (26.4% vs 19.9%, respectively; relative risk, 1.32; 95% confidence interval, 1.12-1.57); however, there was no significant increase in the advanced neoplasia detection rate (3.7% vs 2.9%, respectively). Conclusions: The use of the CADe system for colonoscopy significantly increased the ADR in a large-scale prospective study including 30 endoscopists (Clinical trial registration number: UMIN000040677.)

DOI： 10.1016/j.gie.2021.07.022

Enhancing Model Generalization for Substantia Nigra Segmentation Using a Test-time Normalization-Based Method

Hu T., Itoh H., Oda M., Hayashi Y., Lu Z., Saiki S., Hattori N., Kamagata K., Aoki S., Kumamaru K.K., Akashi T., Mori K.

Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics) 13437 LNCS 巻頁： 736 - 744 2022年

掲載種別：研究論文（国際会議プロシーディングス）出版者・発行元：Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)

Automatic segmentation of substantia nigra (SN), which is Parkinson’s disease-related tissue, is an important step toward accurate computer-aided diagnosis systems. Conventional methods for SN segmentation depend heavily on limited magnetic resonance imaging (MRI) modalities such as neuromelanin and quantitative susceptibility mapping, which require longer imaging times and are rare in public datasets. To enable a multi-modal investigation for SN anatomic alterations based on medical bigdata researches, the need for automated SN segmentation arises from commonly investigated T2-weighted MRIs. To improve the performance of the automated SN segmentation from a T2-weighted MRI and enhance the model generalization for cross-center researches, this paper proposes a novel test-time normalization (TTN) method to increase the geometric and intensity similarity between the query data and the model’s trained data. Our proposed method requires no additional training procedure or extra annotation for the unseen data. Our results showed that our proposed TTN achieved a mean Dice score of 71.08% in comparison with the baseline model’s 69.87% score with in-house dataset. Additionally, improved SN segmentation performance was observed from the unseen and unlabeled datasets.

DOI： 10.1007/978-3-031-16449-1_70

その他リンク： https://dblp.uni-trier.de/db/conf/miccai/miccai2022-7.html#HuIOHLSHKAKAM22

Spatially variant biases considered self-supervised depth estimation based on laparoscopic videos 査読有り

Li Wenda, Hayashi Yuichiro, Oda Masahiro, Kitasaka Takayuki, Misawa Kazunari, Mori Kensaku

Computer Methods in Biomechanics and Biomedical Engineering: Imaging & Visualization 2021年12月

記述言語：英語掲載種別：研究論文（学術雑誌）出版者・発行元：Computer Methods in Biomechanics and Biomedical Engineering: Imaging and Visualization

Depth estimation is an essential tool in obtaining depth information for robotic surgery and augmented reality technology in the current laparoscopic surgery robot system. Since there is a lack of ground-truth for depth values and laparoscope motions during operation, depth estimation networks have difficulties in predicting depth maps from laparoscopic images under the supervised strategy. It is challenging to generate the correct depth maps for the different environments from abdominal images. To tackle these problems, we propose a novel monocular self-supervised depth estimation network with sparse nest architecture. We design a non-local block to capture broader and deeper context features that can further enhance the scene-variant generalisation capacity of the network for the differences in datasets. Moreover, we introduce an improved multi-mask feature in the loss function to tackle the classical occlusion problem based on the time-series information from stereo videos. We also use heteroscedastic aleatoric uncertainty to reduce the effect of noisy data for depth estimation. We compared our proposed method with other existing methods for different scenes in datasets. The experimental results show that the proposed model outperformed the state-of-the-art models qualitatively and quantitatively.

DOI： 10.1080/21681163.2021.2015723

Depth estimation from single-shot monocular endoscope image using image domain adaptation and edge-aware depth estimation 査読有り

Oda Masahiro, Itoh Hayato, Tanaka Kiyohito, Takabatake Hirotsugu, Mori Masaki, Natori Hiroshi, Mori Kensaku

Computer Methods in Biomechanics and Biomedical Engineering: Imaging & Visualization 2021年12月

担当区分：筆頭著者記述言語：英語掲載種別：研究論文（学術雑誌）出版者・発行元：Computer Methods in Biomechanics and Biomedical Engineering: Imaging and Visualization

We propose a depth estimation method from a single-shot monocular endoscopic image using Lambertian surface translation by domain adaptation and depth estimation using multi-scale edge loss. We employ a two-step estimation process including Lambertian surface translation from unpaired data and depth estimation. The texture and specular reflection on the surface of an organ reduce the accuracy of depth estimations. We apply Lambertian surface translation to an endoscopic image to remove these texture and reflections. Then, we estimate the depth by using a fully convolutional network (FCN). During the training of the FCN, improvement of the object edge similarity between an estimated image and a ground truth depth image is important for getting better results. We introduced a muti-scale edge loss function to improve the accuracy of depth estimation. We quantitatively evaluated the proposed method using real colonoscopic images. The estimated depth values were proportional to the real depth values. Furthermore, we applied the estimated depth images to automated anatomical location identification of colonoscopic images using a convolutional neural network. The identification accuracy of the network improved from 69.2% to 74.1% by using the estimated depth images.

DOI： 10.1080/21681163.2021.2012835

Attention-Guided Pancreatic Duct Segmentation from Abdominal CT Volumes 査読有り国際共著

Shen C., Roth H.R., Yuichiro H., Oda M., Miyamoto T., Sato G., Mori K.

Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics) 12969 LNCS 巻頁： 46 - 55 2021年11月

記述言語：英語掲載種別：研究論文（学術雑誌）出版者・発行元：Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)

Pancreatic duct dilation indicates a high risk of pancreatic ductal adenocarcinoma (PDAC), the deadliest cancer with a poor prognosis. Segmentation of dilated pancreatic duct from CT taken from patients without PDAC shows the potential to assist the early detection of PDAC. Most current researches include pancreatic duct segmentation as one additional class for patients who have already detected PDAC. However, the dilated pancreatic duct for people who have not yet developed PDAC is typically much smaller, making the segmentation difficult. Deep learning-based segmentation on tiny components is challenging because of the large imbalance between the target object and irrelevant regions. In this work, we explore an attention-guided approach for dilated pancreatic duct segmentation as a screening tool for pre-PDAC patients, enhancing the pancreas regions’ concentration and ignoring the unnecessary features. We employ a multi-scale aggregation to combine the information at different scales to improve the segmentation performance further. Our proposed multi-scale pancreatic attention-guided approach achieved a Dice score of 54.16% on dilated pancreatic duct dataset, which shows a significant improvement over prior techniques.

DOI： 10.1007/978-3-030-90874-4_5

Multi-task Federated Learning for Heterogeneous Pancreas Segmentation 査読有り国際共著

Shen C., Wang P., Roth H.R., Yang D., Xu D., Oda M., Wang W., Fuh C.S., Chen P.T., Liu K.L., Liao W.C., Mori K.

Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics) 12969 LNCS 巻頁： 101 - 110 2021年11月

記述言語：英語掲載種別：研究論文（学術雑誌）出版者・発行元：Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)

Federated learning (FL) for medical image segmentation becomes more challenging in multi-task settings where clients might have different categories of labels represented in their data. For example, one client might have patient data with “healthy” pancreases only while datasets from other clients may contain cases with pancreatic tumors. The vanilla federated averaging algorithm makes it possible to obtain more generalizable deep learning-based segmentation models representing the training data from multiple institutions without centralizing datasets. However, it might be sub-optimal for the aforementioned multi-task scenarios. In this paper, we investigate heterogeneous optimization methods that show improvements for the automated segmentation of pancreas and pancreatic tumors in abdominal CT images with FL settings.

DOI： 10.1007/978-3-030-90874-4_10

Intestine Segmentation with Small Computational Cost for Diagnosis Assistance of Ileus and Intestinal Obstruction 査読有り

Oda H., Hayashi Y., Kitasaka T., Takimoto A., Hinoki A., Uchida H., Suzuki K., Oda M., Mori K.

Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics) 12969 LNCS 巻頁： 3 - 12 2021年11月

記述言語：英語掲載種別：研究論文（学術雑誌）出版者・発行元：Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)

This paper proposes an intestine segmentation method from CT volumes for the intestinal obstruction and the ileus diagnosis assistance. The previous method was built based on the 3D U-Net, whose computational cost was high. Nevertheless, there was no confirmation that the 3-dimensional network contributed to the segmentation performance. In this paper, we propose a method utilizing the 2D U-Net on behalf of the previous methods’ 3D U-Net. Experimental results using 110 CT volumes showed that both the proposed (2D U-Net) and previous (3D U-Net) methods achieved similar scores for the segmentation accuracy and system usefulness. In addition, the proposed method’s inference was 0.5 min on average, around 8-times faster than the 3D U-Net’s. Although subsequent processes still require more than 20 min for each case, utilizing lightweight networks is essential for practical use in the future, especially for emergency diagnosis.

DOI： 10.1007/978-3-030-90874-4_1

COVID-19 Infection Segmentation from Chest CT Images Based on Scale Uncertainty 査読有り

Oda M., Zheng T., Hayashi Y., Otake Y., Hashimoto M., Akashi T., Aoki S., Mori K.

Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics) 12969 LNCS 巻頁： 88 - 97 2021年11月

担当区分：筆頭著者,　責任著者記述言語：英語掲載種別：研究論文（学術雑誌）出版者・発行元：Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)

This paper proposes a segmentation method of infection regions in the lung from CT volumes of COVID-19 patients. COVID-19 spread worldwide, causing many infected patients and deaths. CT image-based diagnosis of COVID-19 can provide quick and accurate diagnosis results. An automated segmentation method of infection regions in the lung provides a quantitative criterion for diagnosis. Previous methods employ whole 2D image or 3D volume-based processes. Infection regions have a considerable variation in their sizes. Such processes easily miss small infection regions. Patch-based process is effective for segmenting small targets. However, selecting the appropriate patch size is difficult in infection region segmentation. We utilize the scale uncertainty among various receptive field sizes of a segmentation FCN to obtain infection regions. The receptive field sizes can be defined as the patch size and the resolution of volumes where patches are clipped from. This paper proposes an infection segmentation network (ISNet) that performs patch-based segmentation and a scale uncertainty-aware prediction aggregation method that refines the segmentation result. We design ISNet to segment infection regions that have various intensity values. ISNet has multiple encoding paths to process patch volumes normalized by multiple intensity ranges. We collect prediction results generated by ISNets having various receptive field sizes. Scale uncertainty among the prediction results is extracted by the prediction aggregation method. We use an aggregation FCN to generate a refined segmentation result considering scale uncertainty among the predictions. In our experiments using 199 chest CT volumes of COVID-19 cases, the prediction aggregation method improved the dice similarity score from 47.6% to 62.1%.

DOI： 10.1007/978-3-030-90874-4_9

Aorta-aware GAN for non-contrast to artery contrasted CT translation and its application to abdominal aortic aneurysm detection. 査読有り

Hu T, Oda M, Hayashi Y, Lu Z, Kumamaru KK, Akashi T, Aoki S, Mori K

International Journal of Computer Assisted Radiology and Surgery 17 巻 ( 1 ) 頁： 97 - 105 2021年10月

Graph Cuts Loss to Boost Model Accuracy and Generalizability for Medical Image Segmentation 査読有り

Zheng Zhou, Oda Masahiro, Mori Kensaku

2021 IEEE/CVF INTERNATIONAL CONFERENCE ON COMPUTER VISION WORKSHOPS (ICCVW 2021) 2021-October 巻頁： 3297 - 3306 2021年10月

記述言語：英語掲載種別：研究論文（学術雑誌）出版者・発行元：Proceedings of the IEEE International Conference on Computer Vision

Segmentation accuracy and generalization ability are essential for deep learning models, especially in medical image segmentation. We present a novel, robust yet straightforward loss function to boost model accuracy and generalizability for medical image segmentation. We reformulate the graph cuts cost function to a loss function for supervised learning. The graph cuts loss innately focuses on a dual penalty to optimize the regional properties and boundary regularization. We benchmark the proposed loss on six public retinal vessel segmentation datasets with a comprehensive intra-dataset and cross-dataset evaluation. Results reveal that the proposed loss is more generalizable, narrowing the performance gap between different architectures. Besides, models trained with our loss show higher segmentation accuracy and better generalization ability than those trained with other mainstream losses. Moreover, we extend our loss to other segmentation tasks, e.g., left atrium and liver tumor segmentation. The proposed loss still achieves comparable performance to the state-of-the-art, demonstrating its potential for any N-D segmentation problem. The code is available at https://github.com/zzh_enggit/graphcutsloss.

DOI： 10.1109/ICCVW54120.2021.00369

Super-Resolution by Latent Space Exploration: Training with Poorly-Aligned Clinical and Micro CT Image Dataset 査読有り

Zheng T., Oda H., Hayashi Y., Nakamura S., Oda M., Mori K.

Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics) 12965 LNCS 巻頁： 24 - 33 2021年9月

記述言語：英語掲載種別：研究論文（学術雑誌）出版者・発行元：Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)

This paper proposes a super-resolution (SR) method, for performing SR on a poorly-aligned dataset. Super-resolution methods commonly needs aligned low-resolution (LR) and high-resolution (HR) images for training. For obtaining paired LR and HR images in medical imaging, we need to align low and high-resolution data using image registration technology. However, since the hardness of aligning LR and HR images, the aligned LR-HR dataset is always low quality. Conventional SR methods always fail to train using poorly-aligned datasets since these methods need high-quality LR-HR datasets. To tackle this problem, we propose a two-step framework for SR using poorly-aligned datasets. In the first step, we decompose image representation into two parts: one is a content code that captures the image content; the other is a style code that captures the image style and anatomy difference between LR / HR images. To perform SR of a given LR image, we input the content code and a latent variable simultaneously into the SR network to obtain an SR result. In the second step, using the trained SR network and an LR image, we search for a content code, and a style code for generating the most proper SR image. This is conducted by searching for the best content code and the best style code by latent space exploration. We conducted experiments using a poorly-aligned clinical-micro CT lung specimen dataset. Experimental results illustrated the proposed method outperformed conventional SR methods by increasing SSIM from 0.309 to 0.312, and have much more convincing perceptual quality than conventional SR methods.

DOI： 10.1007/978-3-030-87592-3_3

Performance improvement of weakly supervised fully convolutional networks by skip connections for brain structure segmentation 査読有り

Sugino Takaaki, Roth Holger R., Oda Masahiro, Kin Taichi, Saito Nobuhito, Nakajima Yoshikazu, Mori Kensaku

Medical Physics 48 巻 ( 11 ) 頁： 7215 - 7227 2021年9月

記述言語：英語掲載種別：研究論文（学術雑誌）出版者・発行元：Medical Physics

Purpose: For the planning and navigation of neurosurgery, we have developed a fully convolutional network (FCN)-based method for brain structure segmentation on magnetic resonance (MR) images. The capability of an FCN depends on the quality of the training data (i.e., raw data and annotation data) and network architectures. The improvement of annotation quality is a significant concern because it requires much labor for labeling organ regions. To address this problem, we focus on skip connection architectures and reveal which skip connections are effective for training FCNs using sparsely annotated brain images. Methods: We tested 2D FCN architectures with four different types of skip connections. The first was a U-Net architecture with horizontal skip connections that transfer feature maps at the same scale from the encoder to the decoder. The second was a U-Net++ architecture with dense convolution layers and dense horizontal skip connections. The third was a full-resolution residual network (FRRN) architecture with vertical skip connections that pass feature maps between each downsampled scale path and the full-resolution scale path. The last one was a hybrid architecture with a combination of horizontal and vertical skip connections. We validated the effect of skip connections on medical image segmentation from sparse annotation based on these four FCN architectures, which were trained under the same conditions. Results: For multiclass segmentation of the cerebrum, cerebellum, brainstem, and blood vessels from sparsely annotated MR images, we performed a comparative evaluation of segmentation performance among the above four FCN approaches: U-Net, U-Net++, FRRN, and hybrid architectures. The experimental results show that the horizontal skip connections in the U-Net architectures were effective for the segmentation of larger sized objects, whereas the vertical skip connections in the FRRN architecture improved the segmentation of smaller sized objects. The hybrid architecture with both horizontal and vertical skip connections achieved the best results of the four FCN architectures. We then performed an ablation study to explore which skip connections in the FRRN architecture contributed to the improved segmentation of blood vessels. In the ablation study, we compared the segmentation performance between architectures with a horizontal path (HP), an HP and vertical up paths (HP+VUPs), an HP and vertical down paths (HP+VDPs), and an HP and vertical up and down paths (FRRN). We found that the vertical up paths were effective in improving the segmentation of smaller sized objects. Conclusions: This paper investigated which skip connection architectures were effective for multiclass brain segmentation from sparse annotation. Consequently, using vertical skip connections with horizontal skip connections allowed FCNs to improve segmentation performance.

DOI： 10.1002/mp.15192

　前のページ - 次のページ　

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18