研究者詳細 - 小田　昌宏

論文 - 小田　昌宏

分割表示 355 件中 1 - 20 件目／全件表示 >>

Computer-aided detection for esophageal achalasia (with video) 査読有り国際共著

Hironari Shiwaku, Masashi Misawa, Haruhiro Inoue, Kai Jiang, Masahiro Oda, Pietro Familiari, Guido Costamagna, Yuto Shimamura, Yuichiro Ikebuchi, Yugo Iwaya, Masaki Ominami, Bu'Hussain Hayee, Khek-Yu Ho, Jimmy B Y So, Hein Myat Thu Htet, Pradeep Bhandari, Kevin Grimes, Helmut Messmann, Bianca Maria Quarta Colosso, Roberta Maselli, Cesare Hassan, Alessandro Repici, Stavros N Stavropoulos, Norio Fukami, Robert Bechara, Michel Kahaleh, Amrita Sethi, Torsten Beyna, Horst Neuhaus, Philip W Y Chiu, Esperanza Grace Santi, Prateek Sharma, Nikolas Eleftheriadis, Hitomi Minami, Gregory Haber, Peter V Draganov, Stefan Seewald, Akio Shiwaku, Yoshiyuki Shiwaku, Kensaku Mori, Shin-Ei Kudo, Suguru Hasegawa

Digestive Endoscopy 2025年6月

Adaptive sensitivity-fisher regularization for heterogeneous transfer learning of vascular segmentation in laparoscopic videos

Zhao, XK; Hayashi, Y; Oda, M; Kitasaka, T; Misawa, K; Mori, K

INTERNATIONAL JOURNAL OF COMPUTER ASSISTED RADIOLOGY AND SURGERY 2025年6月

記述言語：英語出版者・発行元：International Journal of Computer Assisted Radiology and Surgery

Purpose: This study aims to enhance surgical safety by developing a method for vascular segmentation in laparoscopic surgery videos with limited visibility. We introduce an adaptive sensitivity-fisher regularization (ASFR) approach to adapt neural networks, initially trained on non-medical datasets, for vascular segmentation in laparoscopic videos. Methods: Our approach utilizes heterogeneous transfer learning by integrating fisher information and sensitivity analysis to mitigate catastrophic forgetting and overfitting caused by limited annotated data in laparoscopic videos. We calculate fisher information to identify and preserve critical model parameters while using sensitivity measures to guide adjustment for new task. Results: The fine-tuned models demonstrated high accuracy in vascular segmentation across various complex video sequences, including those with obscured vessels. For both invisible and visible vessels, our method achieved an average Dice score of 41.3. In addition to outperforming traditional transfer learning approaches, our method exhibited strong adaptability across multiple advanced video segmentation architectures. Conclusion: This study introduces a novel heterogeneous transfer learning approach, ASFR, which significantly enhances the precision of vascular segmentation in laparoscopic videos. ASFR effectively addresses critical challenges in surgical image analysis and paves the way for broader applications in laparoscopic surgery, promising improved patient outcomes and increased surgical efficiency.

DOI： 10.1007/s11548-025-03404-2

Web of Science

Scopus

PubMed

Semantic segmentation dataset authoring with simplified labels 査読有り Open Access

Leo Uramoto, Yuichiro Hayashi, Masahiro Oda, Takayuki Kitasaka, Kensaku Mori

International Journal of Computer Assisted Radiology and Surgery 2025年4月

Development of a YOLOv3-Based Model for Automated Detection of Thoracic Ossification of the Posterior Longitudinal Ligament and the Ligamentum Flavum on Plain Radiographs 査読有り Open Access

Sadayuki Ito, Hiroaki Nakashima, Naoki Segi, Jun Ouchida, Ippei Yamauchi, Takashi Hirai, Masahiro Oda, Kensaku Mori, Masashi Yamazaki, Toshitaka Yoshii, Shiro Imagama

Journal of Clinical Medicine 14 巻 ( 7 ) 2025年3月

Bidirectional teaching between lightweight multi-view networks for intestine segmentation from CT volume 査読有り Open Access

Qin An, Hirohisa Oda, Yuichiro Hayashi, Takayuki Kitasaka, Aitaro Takimoto, Akinari Hinoki, Hiroo Uchida, Kojiro Suzuki, Masahiro Oda, Kensaku Mori

Journal of Medical Imaging 12 巻 ( 2 ) 2025年3月

Enhanced self-supervised monocular depth estimation with self-attention and joint depth-pose loss for laparoscopic images 査読有り Open Access

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

International Journal of Computer Assisted Radiology and Surgery 2025年2月

Multi-dimensional consistency learning between 2D Swin U-Net and 3D U-Net for intestine segmentation from CT volume 査読有り Open Access

Qin An, Hirohisa Oda, Yuichiro Hayashi, Takayuki Kitasaka, Hiroo Uchida, Akinari Hinoki, Kojiro Suzuki, Aitaro Takimoto, Masahiro Oda, Kensaku Mori

International Journal of Computer Assisted Radiology and Surgery 2025年2月

Classification of fundus autofluorescence images based on macular function in retinitis pigmentosa using convolutional neural networks Open Access

Kominami, T; Ueno, S; Ota, J; Inooka, T; Oda, M; Mori, K; Nishiguchi, KM

JAPANESE JOURNAL OF OPHTHALMOLOGY 2025年2月

記述言語：英語出版者・発行元：Japanese Journal of Ophthalmology

Purpose: To determine whether convolutional neural networks (CNN) can classify the severity of central vision loss using fundus autofluorescence (FAF) images and color fundus images of retinitis pigmentosa (RP), and to evaluate the utility of those images for severity classification. Study design: Retrospective observational study. Methods: Medical charts of patients with RP who visited Nagoya University Hospital were reviewed. Eyes with atypical RP or previous surgery were excluded. The mild group was comprised of patients with a mean deviation value of > − 10 decibels, and the severe group of < − 20 decibels, in the Humphrey field analyzer 10-2 program. CNN models were created by transfer learning of VGG16 pretrained with ImageNet to classify patients as either mild or severe, using FAF images or color fundus images. Results: Overall, 165 patients were included in this study; 80 patients were classified into the severe and 85 into the mild group. The test data comprised 40 patients in each group, and the images of the remaining patients were used as training data, with data augmentation by rotation and flipping. The highest accuracies of the CNN models when using color fundus and FAF images were 63.75% and 87.50%, respectively. Conclusion: Using FAF images may enable the accurate assessment of central vision function in RP. FAF images may include more parameters than color fundus images that can evaluate central visual function.

DOI： 10.1007/s10384-025-01163-w

Artificial intelligence support improves diagnosis accuracy in anterior segment eye diseases 招待有り査読有り

Hiroki Maehara, Yuta Ueno, Takefumi Yamaguchi, Yoshiyuki Kitaguchi, Dai Miyazaki, Ryohei Nejima, Takenori Inomata, Naoko Kato, Tai-ichiro Chikama, Jun Ominato, Tatsuya Yunoki, Kinya Tsubota, Masahiro Oda, Manabu Suzutani, Tetsuju Sekiryu, Tetsuro Oshika

Scientific Reports 15 巻 2025年2月

Double-mix pseudo-label framework: enhancing semi-supervised segmentation on category-imbalanced CT volumes Open Access

Zhang, LY; Hayashi, Y; Oda, M; Mori, K

INTERNATIONAL JOURNAL OF COMPUTER ASSISTED RADIOLOGY AND SURGERY 2025年2月

記述言語：英語出版者・発行元：International Journal of Computer Assisted Radiology and Surgery

Purpose: Deep-learning-based supervised CT segmentation relies on fully and densely labeled data, the labeling process of which is time-consuming. In this study, our proposed method aims to improve segmentation performance on CT volumes with limited annotated data by considering category-wise difficulties and distribution. Methods: We propose a novel confidence-difficulty weight (CDifW) allocation method that considers confidence levels, balancing the training across different categories, influencing the loss function and volume-mixing process for pseudo-label generation. Additionally, we introduce a novel Double-Mix Pseudo-label Framework (DMPF), which strategically selects categories for image blending based on the distribution of voxel-counts per category and the weight of segmentation difficulty. DMPF is designed to enhance the segmentation performance of categories that are challenging to segment. Result: Our approach was tested on two commonly used datasets: a Congenital Heart Disease (CHD) dataset and a Beyond-the-Cranial-Vault (BTCV) Abdomen dataset. Compared to the SOTA methods, our approach achieved an improvement of 5.1% and 7.0% in Dice score for the segmentation of difficult-to-segment categories on 5% of the labeled data in CHD and 40% of the labeled data in BTCV, respectively. Conclusion: Our method improves segmentation performance in difficult categories within CT volumes by category-wise weights and weight-based mixture augmentation. Our method was validated across multiple datasets and is significant for advancing semi-supervised segmentation tasks in health care. The code is available at https://github.com/MoriLabNU/Double-Mix.

DOI： 10.1007/s11548-024-03281-1

A deep learning-based prediction model for prognosis of cervical spine injury: a Japanese multicenter survey 査読有り

Sadayuki Ito, Hiroaki Nakashima, Naoki Segi, Noriaki Yokogawa, Takeshi Sasagawa, Toru Funayama, Fumihiko Eto, Akihiro Yamaji, Kota Watanabe, Satoshi Nori, Kazuki Takeda, Takeo Furuya, Atsushi Yunde, Hideaki Nakajima, Tomohiro Yamada, Tomohiko Hasegawa, Yoshinori Terashima, Ryosuke Hirota, Hidenori Suzuki, Yasuaki Imajo, Shota Ikegami, Masashi Uehara, Hitoshi Tonomura, Munehiro Sakata, Ko Hashimoto, Yoshito Onoda, Kenichi Kawaguchi, Yohei Haruta, Nobuyuki Suzuki, Kenji Kato, Hiroshi Uei, Hirokatsu Sawada, Kazuo Nakanishi, Kosuke Misaki, Hidetomi Terai, Koji Tamai, Akiyoshi Kuroda, Gen Inoue, Kenichiro Kakutani, Yuji Kakiuchi, Katsuhito Kiyasu, Hiroyuki Tominaga, Hiroto Tokumoto, Yoichi Iizuka, Eiji Takasawa, Koji Akeda, Norihiko Takegami, Haruki Funao, Yasushi Oshima, Takashi Kaito, Daisuke Sakai, Toshitaka Yoshii, Tetsuro Ohba, Bungo Otsuki, Shoji Seki, Masashi Miyazaki, Masayuki Ishihara, Masahiro Oda, Seiji Okada, Shiro Imagama, Satoshi Kato

European Spine Journal 2025年2月

The importance of clinical experience in AI-assisted corneal diagnosis: verification using intentional AI misleading 査読有り

Scientific Reports 15 巻 ( 1462 ) 2025年1月

Content-consistent adaptation network for substantia nigra segmentation in cross-modalities and -datasets scenarios without retraining

Hu T., Itoh H., Oda M., Saiki S., Hattori N., Kamagata K., Sako W., Ishikawa K.I., Aoki S., Mori K.

Progress in Biomedical Optics and Imaging - Proceedings of SPIE 13406 巻 2025年

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

Fully automated segmentation of substantia nigra (SN) is crucial for developing the interpretable computer-aided diagnosis system for Parkinson's disease (PD). In real-world applications, the segmentation models may be trained with high quality medical images, while the testing dataset may be scanned with an older machine and is of relatively lower quality. To address this domain shift problem, conventional domain adaptation methods propose various strategies to adapt the segmentation model to the testing data. These conventional adaptation methods typically include a retraining stage in the target domain, either for the segmentation model or a translation model. This process makes the predictions in real multi-modalities and -datasets scenarios highly time-consuming. To enable rapid SN segmentation in cross-modalities and -datasets tasks, this paper proposes a content-consistent adaptation network using only source domain data to synthesize training data with the consistent contents of the target domain. The proposed method requires no retraining during the testing stage, thus enabling rapid SN segmentation across modalities and datasets. The results indicated the competitive performances of our proposed method in both the cross-modalities and -datasets tasks.

DOI： 10.1117/12.3047729

Scopus

CLSMI2T3: 3D CLSM Vasculature Volume Reconstruction from A Single 2D Slice by Off-Focal Plane Signal Using Synthetic Data

Wu Y., Menten M.J., Kreitner L., Kawamura S., Oda M., Hayashi Y., Takebe T., Rueckert D., Mori K.

Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics) 15371 LNCS 巻頁： 108 - 118 2025年

出版者・発行元：Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)

In this paper, we propose a novel framework for reconstructing 3D Confocal Laser Scanning Microscopy (CLSM) vasculature volumes from single 2D slices using synthetic and real CLSM data. CLSM is an imaging technique commonly used for in-vivo 3D imaging. However, the utility of 3D CLSM imaging is limited due to long acquisition times and phototoxicity caused by prolonged light exposure. We propose a novel solution to reconstruct a 3D volume from a 2D slice by off-focal plane signal using deep learning (DL). However, the limited availability of real 3D data has to be overcome to train such a 3D reconstruction model. Therefore, we propose a novel framework to effectively address the lack of 3D data for training a robust and accurate 3D reconstruction model by simulating and synthesizing CLSM volumes. Our framework consists of (1) a 3D CLSM volume simulation and synthesizing for training data and (2) a 3D CLSM volume reconstruction part. Experiments show that we can effectively reconstruct 3D CLSM vasculature volumes from single slices, even only using synthetic data. Furthermore, we will also release our CLSM Image of Vascular Organoid for REConstruction (CLSMI-VO-REC) dataset and code at https://github.com/MoriLabNU/3D_CLSM_Volume_Reconstruction.

DOI： 10.1007/978-3-031-77786-8_11

Scopus

A Deployable Microscopic Image Segmentation Look-Up Table Based on A Dilated CNN

Wu Y., Pan J., Kawamura S., Oda M., Hayashi Y., Takebe T., Rueckert D., Mori K.

Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics) 15371 LNCS 巻頁： 3 - 13 2025年

出版者・発行元：Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)

Microscopic image segmentation is indispensable biomedical research. However, accurately segmenting using traditional automatic segmentation methods remains challenging due to fluorescence presenting high intra- and inter-image variability, hindering biomedical analysis. Deep learning (DL)-based methods show remarkable potential in microscopic image segmentation. However, DL models typically require GPU resources, specialized programming skills, and a non-trivial environment setup, making it difficult for medical researchers to use them currently. Therefore, we propose a turnkey solution for microscopic image segmentation by Look-up Table (LUT) which is a data structure used for storing input values and pre-calculated results. Our solution can inherit the segmentation accuracy of DL-based methods and eliminates the need for extensive computational resources, specialized programming, and a complex implementation environment. Specifically, (1) By transferring all potential inputs and their DL model-computed outputs into a LUT, segmentation is achieved through direct output look-up for given pixel values. (2) Additionally, we develop a dilated CNN with adaptive edge padding, enhancing feature learning while meeting LUT building requirements. (3) Moreover, we publicly release a confocal microscopic image dataset and develop a LUT-based plugin (SEG-LUT plugin) for Fiji ImageJ. The experimental results show our framework outperforms other automatic segmentation methods in the Fiji ImageJ. Our solution offers a pathway for easily utilizing DL-based methods for microscopic image segmentation. Our codes, plugin, and dataset are available at https://github.com/MoriLabNU/Microscopic_Image_Segment_LUT.

DOI： 10.1007/978-3-031-77786-8_1

Scopus

Automated Skill Evaluation Method for Surgical Images using Optical Flow and Attention Mechanism

Feng Y., Hayashi Y., Oda M., Kitasaka T., Yasui A., Shirota C., Uchida H., Mori K.

Progress in Biomedical Optics and Imaging Proceedings of SPIE 13408 巻 2025年

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

This paper describes an automated skill evaluation method for surgical images based on optical flow and self-attention mechanisms. This method uses a segment of a surgical video as input to a deep learning network and outputs scores. Previous methods primarily analyzed the kinematic data of surgical robots to assess surgical skills; however, kinematic data is often unavailable in non-robotic surgeries. Therefore, we propose a regression model that uses surgical video as input data to evaluate surgical skills through optical flow and self-attention mechanisms. We use optical flow to extract motion information from surgical images and employ the ViViT (Video Vision Transformer) model to extract features from both image information and motion information. Finally, we concatenate the two feature vectors and output six scores corresponding to the GRS (Global Rating Scale) evaluation criteria. To validate the evaluation performance of the model, we calculated the RMSE (Root Mean Squared Error) between the model's output scores of surgical videos and the ground truth scores in the surgical skill assessment dataset JIGSAWS. A low RMSE of 0.79 demonstrated that our method can reliably estimate scores, highlighting its potential for surgical skill assessment.

DOI： 10.1117/12.3049068

Scopus

Enhanced Sequential Picking Strategy with Hyperparameter Optimization for Efficient Surgical Phase Recognition

Yao G., Hayashi Y., Oda M., Misawa K., Mori K.

Progress in Biomedical Optics and Imaging Proceedings of SPIE 13408 巻 2025年

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

Accurate surgical phase recognition from surgical videos is critical for surgical training and evaluation. However, traditional models like Transformers face limitations in memory and computational efficiency, particularly as input sequence length increases. We introduce an enhanced Sequential Picking Strategy with hyperparameter optimization within the SPT-SVNet model, which integrates spatial and temporal embeddings. Through hyperparameter search, we apply different sequential picking strategies by optimizing these hyperparameters, enabling the model to selectively extract crucial features from temporal embeddings and reduce computational demands to varying degrees. Optimizing these hyperparameters led to significant performance improvements, reducing memory usage by 8.6% and training time by 16.2%, while maintaining or even enhancing model accuracy. We evaluated our approach using the SPT-SVNet model. Evaluations on the Cholec80 dataset demonstrated remarkable results, with our method achieving an accuracy of 90.5% and a precision of 92.4%, surpassing existing models.

DOI： 10.1117/12.3047731

Scopus

Estimation of hidden blood vessel locations in laparoscopic videos by combining preoperative and intraoperative information

Enomoto K., Hayashi Y., Misawa K., Kitasaka T., Oda M., Mori K.

Progress in Biomedical Optics and Imaging Proceedings of SPIE 13408 巻 2025年

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

This paper presents a method for estimating hidden blood vessel locations in laparoscopic videos by combining preoperative and intraoperative information. In many laparoscopic surgeries, surgeons need to cut blood vessels. However, because the blood vessels are often hidden by adipose tissues and other anatomical structures, direct observation may not be possible. A computer-aided surgery system that reveals the locations of these concealed blood vessels could significantly assist surgeons during laparoscopic surgery. Therefore, we develop a method for estimating the hidden blood vessels in laparoscopic videos. The proposed method estimates the location of the hidden blood vessels by utilizing both the preoperative information and the intraoperative laparoscopic videos. We apply fully convolutional networks (FCNs) to each frame of the laparoscopic videos to estimate these locations. Our FCN model utilizes attention mechanisms to incorporate the preoperative information. The preoperative information for the first frame is derived from preoperative CT volumes. In the following frames, the preoperative information is deformed using the optical flow among successive two frames. We tested the proposed method on seven videos recorded during robot-assisted laparoscopic gastrectomies for gastric cancer. Experimental results demonstrated that the proposed method successfully estimated the locations of hidden blood vessels in laparoscopic video. The results also highlight the benefits of integrating preoperative and intraoperative data.

DOI： 10.1117/12.3047525

Scopus

Prior knowledge-based blood vessel reconstruction and thrombus visualization using confocal laser scanning microscopic images

Wang, C; Hayashi, Y; Oda, M; Kawamura, S; Takebe, T; Mori, K

MEDICAL IMAGING 2025: CLINICAL AND BIOMEDICAL IMAGING 13410 巻 2025年

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

This paper describes an algorithm for the reconstruction of 3D blood vessel (BV) structure and thrombus. Biologists use a confocal laser scanning microscope (CLSM) to observe the brain of a mouse. However, the CLSM images obtained are in 2D, which limits the understanding of the 3D structure. In a previous method, the 3D BV structure was reconstructed from the centerline of the BV extracted from CLSM images. However, the reconstructed 3D BV structure is not realistic due to the influence of image noise. Therefore, we apply a smooth operation to the centerline of BV to make the reconstructed 3D BV structure more realistic. Meanwhile, we segment the thrombus from the CLSM images and visualize it within the reconstructed 3D BV. The 3D BV structure reconstruction method consists of three steps: (1) BV region segmentation, (2) centerline calculation and 3D mapping, and (3) 3D BV structure reconstruction. The 3D thrombus is reconstructed based on the reconstructed BV shape. Experimental results showed that the BVs obtained by the proposed method were smoother and more realistic compared to the previous method. In addition, the proposed method displayed the reconstructed thrombus in 3D space together with the BV, allowing 3D viewing from any viewpoint.

DOI： 10.1117/12.3048339

Web of Science

Scopus

SemiOrth: A Novel Orthogonal Dual Network Architecture for Enhanced Semi-Supervised Medical Image Segmentation

Zhang, LY; Oda, M; Mori, K

MEDICAL IMAGING 2025: IMAGE PROCESSING 13406 巻 2025年

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

Medical image segmentation is crucial for Computer-Aided Diagnosis in healthcare. However, the confidentiality of medical data and the difficulty of creating segmentation annotations require training in high-precision models with limited annotated data. This study introduces a novel co-model training framework that uses an orthogonality loss function to enhance the heterogeneity between two models, enabling them to focus on different target features more effectively. Additionally, we employ a hybrid approach to refine pseudo-labels by evaluating predictions from the same image under varying augmentations, improving their accuracy. Tested on the Automated Cardiac Diagnosis Challenge dataset, our methodology achieved competitive performance with only 8% labeled data, underscoring its efficacy in optimizing medical image segmentation with minimal annotations.

DOI： 10.1117/12.3047175

Web of Science

Scopus

　前のページ - 次のページ　

<小田 昌宏>

論文 - 小田 昌宏

<小田　昌宏>

論文 - 小田　昌宏