Faculty Profiles - ODA, Masahiro

Papers - ODA, Masahiro

Division display 21 - 40 of about 372 ／ All the affair displays >>

Enhanced self-supervised monocular depth estimation with self-attention and joint depth-pose loss for laparoscopic images Reviewed

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

International Journal of Computer Assisted Radiology and Surgery 2025.2

Multi-dimensional consistency learning between 2D Swin U-Net and 3D U-Net for intestine segmentation from CT volume Reviewed 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 2025.2

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

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

JAPANESE JOURNAL OF OPHTHALMOLOGY 2025.2

Language：English Publisher：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

Web of Science

Scopus

PubMed

Artificial intelligence support improves diagnosis accuracy in anterior segment eye diseases Invited Reviewed Open Access

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 Vol. 15 2025.2

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

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

INTERNATIONAL JOURNAL OF COMPUTER ASSISTED RADIOLOGY AND SURGERY 2025.2

Language：English Publisher：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

Web of Science

Scopus

PubMed

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

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 2025.2

The importance of clinical experience in AI-assisted corneal diagnosis: verification using intentional AI misleading Reviewed Open Access

Scientific Reports Vol. 15 ( 1462 ) 2025.1

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 Vol. 13406 2025

Publisher：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

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) Vol. 15371 LNCS page： 3 - 13 2025

Publisher：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

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) Vol. 15371 LNCS page： 108 - 118 2025

Publisher：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

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 Vol. 13408 2025

Publisher：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 Vol. 13408 2025

Publisher：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 Vol. 13408 2025

Publisher：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 Vol. 13410 2025

Publisher：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 Vol. 13406 2025

Publisher：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

Uncertainty-Aware Semi-Supervised Learning for Enhanced Multi-Organ Segmentation in CT Volumes

Dong, G; Oda, M; Hayashi, Y; Misawa, K; Mori, K

MEDICAL IMAGING 2025: IMAGE PROCESSING Vol. 13406 2025

Publisher：Progress in Biomedical Optics and Imaging Proceedings of SPIE

We propose an uncertainty-aware semi-supervised learning (SSL) strategy for multi-organ segmentation in CT volumes. CT volume segmentation is crucial for enhancing diagnostic accuracy, facilitating quantitative analysis, and enabling automation and assisted diagnosis. Previous SSL strategies have shown great performance on CT volumes by ensuring consistency between different outputs of the model. However, these methods do not address the noise and uncertainty within individual outputs, which can impact the results. Our method integrates uncertainty estimation into the SSL framework, dynamically generating masks based on the uncertainty of individual predictions. This approach not only maintains consistency between multiple outputs but also mitigates the impact of noise and uncertainty within each prediction. Under the 10% labeled data setting, comparative experiments demonstrated that our method improved segmentation performance across effectively segmented organs. Specifically, compared to the baseline, the Dice scores increased by 2.50% for the vein, 1.99% for the liver, 5.16% for the spleen, and 6.09% for the stomach, indicating the potential benefits of incorporating uncertainty estimation into the SSL framework.

DOI： 10.1117/12.3047155

Web of Science

Scopus

WPDM: Enhancing Subarachnoid Hemorrhage CT Detection with Worley-Perlin Noise Diffusion on Imbalanced Data

Lu, ZY; Hu, T; Oda, M; Hayashi, Y; Watadani, T; Mori, K

MEDICAL IMAGING 2025: COMPUTER-AIDED DIAGNOSIS Vol. 13407 2025

Publisher：Progress in Biomedical Optics and Imaging Proceedings of SPIE

We propose a diffusion model based on Worley-Perlin noise, which provides higher robustness for image synthesis problems with striated textures, e.g. subarachnoid hemorrhage (SAH) CT slices. Unlike Gaussian noise, which uniformly perturbs images and causes the generated samples to be overly homogeneous, or Simplex noise, which excessively disrupts images and leads to distortion, our method accounts for the morphological structure of textured images. It ensures effective perturbation while avoiding image distortion and generates high-quality samples for downstream tasks. To validate the effectiveness of our proposed generative model in improving SAH CT detection performance on imbalanced datasets, we incorporated the generated SAH samples into the training process and compared the results with those of models trained on the original dataset. The experimental results demonstrated that our synthesized data can help improve the classification accuracy in an imbalanced data situation, with the F1-score improving by nearly 6.4%.

DOI： 10.1117/12.3047558

Web of Science

Scopus

Enforcing Geometric Constraints of Surface Normal and Pose for Self-supervised Monocular Depth Estimation on Laparoscopic Images.

Wenda Li 0004, Yuichiro Hayashi, Masahiro Oda 0001, Takayuki Kitasaka, Kazunari Misawa, Kensaku Mori

MICCAI (9) Vol. 15968 LNCS page： 235 - 245 2025

Preliminary study of substantia nigra analysis by tensorial feature extraction Reviewed

Hayato Itoh, Masahiro Oda, Shinji Saiki, Koji Kamagata, Wataru Sako, Kei-ichi Ishikawa, Nobutaka Hattori, Shigeki Aoki, Kensaku Mori

International Journal of Computer Assisted Radiology and Surgery Vol. 19 ( 11 ) page： 2133 - 2142 2024.11

【放射線科と人工知能】脊椎・脊髄疾患の画像診断におけるAI利用の現状

伊藤定之, 中島宏彰, 世木直喜, 大内田隼, 小田昌宏, 山内一平, 大出幸史, 長谷康弘, 岡田裕也, 鏡味佑志朗, 森健策, 今釜史郎

臨床放射線 Vol. 69 ( 6 ) page： 787 - 792 2024.11

Language：Japanese Publisher：金原出版(株)

＜文献概要＞脊椎・脊髄疾患は,高齢化社会においてますます注目される分野であり,特に脊髄腫瘍や脊髄圧迫症などの重大な病態が含まれる。画像診断はこれらの疾患の検出と診断に不可欠であり,近年,人工知能(AI)技術が医療において急速に進展し,画像診断の効率と精度を向上させる手段として期待されている。本稿では,脊椎・脊髄疾患におけるAIの現状とその応用について,特に画像診断分野における具体的な利用例や技術的な進展を中心に紹介する。

researchmap

　 PREV - NEXT 　

<ODA, Masahiro>

Papers - ODA, Masahiro