MFMasahiro Fujiwara

Masahiro Fujiwara

*This is an algorithmically generated author record based on publicly available information.

Current Organizations / Institutes

The organizations/institutes from the scholar's latest article

Previous Organizations / Institutes

The organizations / institutes from the scholar's other articles in the past ten years


Masahiro Fujiwara has a total of 38 co-authors. Here are the top co-authors by number of shared publications:


Latest Publications

  • Open Access

    Comparison of active learning algorithms in classifying head computed tomography reports using bidirectional encoder representations from transformers

    by Tomohiro Wataya, Azusa Miura, Takahisa Sakisuka, Masahiro Fujiwara, Hisashi Tanaka, Yu Hiraoka, Junya Sato, Miyuki Tomiyama, Daiki Nishigaki, Kosuke Kita, et al.

    Purpose: Systems equipped with natural language (NLP) processing can reduce missed radiological findings by physicians, but the annotation costs are burden in the development. This study aimed to compare the effects of active learning (AL) algorithms in NLP for estimating the significance of head computed tomography (CT) reports using bidirectional encoder representations from transformers (BERT). Methods: A total of 3728 head CT reports annotated with five categories of importance were used and UTH-BERT was adopted as the pre-trained BERT model. We assumed that 64% (2385 reports) of the data were initially in the unlabeled data pool (UDP), while the labeled data set (LD) used to train the model was empty. Twenty-five reports were repeatedly selected from the UDP and added to the LD, based on seven metrices: random sampling (RS: control), four uncertainty sampling (US) methods (least confidence (LC), margin sampling (MS), ratio of confidence (RC), and entropy sampling (ES)), and two distance-based sampling (DS) methods (cosine distance (CD) and Euclidian distance (ED)). The transition of accuracy of the model was evaluated using the test dataset. Results: The accuracy of the models with US was significantly higher than RS when reports in LD were < 1800, whereas DS methods were significantly lower than RS. Among the US methods, MS and RC were even better than the others. With the US methods, the required labeled data decreased by 15.4–40.5%, and most efficient in RC. In addition, in the US methods, data for minor categories tended to be added to LD earlier than RS and DS. Conclusions: In the classification task for the importance of head CT reports, US methods, especially RC and MS can lead to the effective fine-tuning of BERT models and reduce the imbalance of categories. AL can contribute to other studies on larger datasets by providing effective annotation.


  • Open Access

    Comparison of natural language processing algorithms in assessing the importance of head computed tomography reports written in Japanese

    by Tomohiro Wataya, Azusa Miura, Takahisa Sakisuka, Masahiro Fujiwara, Hisashi Tanaka, Yu Hiraoka, Junya Sato, Miyuki Tomiyama, Daiki Nishigaki, Kosuke Kita, et al.

    Purpose: To propose a five-point scale for radiology report importance called Report Importance Category (RIC) and to compare the performance of natural language processing (NLP) algorithms in assessing RIC using head computed tomography (CT) reports written in Japanese. Materials and methods: 3728 Japanese head CT reports performed at Osaka University Hospital in 2020 were included. RIC (category 0: no findings, category 1: minor findings, category 2: routine follow-up, category 3: careful follow-up, and category 4: examination or therapy) was established based not only on patient severity but also on the novelty of the information. The manual assessment of RIC for the reports was performed under the consensus of two out of four neuroradiologists. The performance of four NLP models for classifying RIC was compared using fivefold cross-validation: logistic regression, bidirectional long–short-term memory (BiLSTM), general bidirectional encoder representations of transformers (general BERT), and domain-specific BERT (BERT for medical domain). Results: The proportion of each RIC in the whole data set was 15.0%, 26.7%, 44.2%, 7.7%, and 6.4%, respectively. Domain-specific BERT showed the highest accuracy (0.8434 ± 0.0063) in assessing RIC and significantly higher AUC in categories 1 (0.9813 ± 0.0011), 2 (0.9492 ± 0.0045), 3 (0.9637 ± 0.0050), and 4 (0.9548 ± 0.0074) than the other models (p < .05). Analysis using layer-integrated gradients showed that the domain-specific BERT model could detect important words, such as disease names in reports. Conclusions: Domain-specific BERT has superiority over the other models in assessing our newly proposed criteria called RIC of head CT radiology reports. The accumulation of similar and further studies of has a potential to contribute to medical safety by preventing missed important findings by clinicians.


  • High-speed and Low-Latency Ocular Parallax Rendering Improves Binocular Fusion in Stereoscopic Vision

    by Yuri Mikawa, Masahiro Fujiwara, Yasutoshi Makino, Hiroyuki Shinoda

    Most of the current virtual/augmented reality (VR/AR) displays use binocular parallax to present 3D images. However, they do not consider ocular parallax, which refers to the slight movement of the viewpoint with eye rotation, such as saccade. A commercial head-mounted display (HMD) has a large latency in realizing ocular parallax rendering. In our study, a high-speed (1,000 fps) and low-latency (average 4.072 ms) ocular-parallax-rendering device was assembled, and its effect was examined wherein a reasonable approximation algorithm for viewpoint was applied. A user study experiment employing a random dot stereogram (RDS) showed improvements in binocular fusion, the causes of which are comprehensively discussed.

  • Spontaneous closure of non-cavernous sinus dural arteriovenous fistulas: A case series and systematic review of the literature

    by Nobuo Kashiwagi, Koichi Miyazaki, Hiroto Takahashi, Kiyoshi Tsuji, Masahiro Fujiwara, Atsuko Arisawa, Hajime Nakamura, Haruhiko Kishima, Kazunari Ishii, Noriyuki Tomiyama


  • Open Access

    Improved visualization of the chorda tympani nerve using ultra‐high-resolution computed tomography

    by Masahiro Fujiwara, Yoshiyuki Watanabe, Nobuo Kashiwagi, Yumi Ohta, Takashi Sato, Megumi Nishigaki, Noriyuki Tomiyama

    Recognition of the anatomical course of the chorda tympani nerve (CTN) is important for preventing iatrogenic injuries during middle-ear surgery. This study aims to compare visualization of the CTN using two computed tomography (CT) methods: conventional high-resolution CT (C-HRCT) and ultra‐high-resolution CT (U-HRCT). We performed a retrospective visual assessment of 59 CTNs in normal temporal bones of 54 consecutive patients who underwent both C-HRCT and U-HRCT. After dividing CTN into three anatomical segments (posterior canaliculus, tympanic segment, and anterior canaliculus), two neuroradiologists scored the visualizations on a four-point scale. On C-HRCT, the visual scores of the posterior canaliculus, tympanic segment, and anterior canaliculus were 3.5 ± 0.7, 1.6 ± 0.6, and 3.1 ± 0.7, respectively. The respective values were significantly higher in all segments on U-HRCT: 3.9 ± 0.2, 2.4 ± 0.6, 3.5 ± 0.6 (p < 0.01). Although the difference in scores between methods was greatest for the tympanic segment, the visual score on U-HRCT was lower for the tympanic segment than for the anterior and posterior segments (p < 0.01). Ultra‐high-resolution CT provides superior visualization of the CTN, especially the tympanic segment.


  • Improvement of the diagnostic accuracy for intracranial haemorrhage using deep learning–based computer-assisted detection

    by Yoshiyuki Watanabe, Takahiro Tanaka, Atsushi Nishida, Hiroto Takahashi, Masahiro Fujiwara, Takuya Fujiwara, Atsuko Arisawa, Hiroki Yano, Noriyuki Tomiyama, Hajime Nakamura, et al.

    Purpose To elucidate the effect of deep learning–based computer-assisted detection (CAD) on the performance of different-level physicians in detecting intracranial haemorrhage using CT. Methods A total of 40 head CT datasets (normal, 16; haemorrhagic, 24) were evaluated by 15 physicians (5 board-certificated radiologists, 5 radiology residents, and 5 medical interns). The physicians attended 2 reading sessions without and with CAD. All physicians annotated the haemorrhagic regions with a degree of confidence, and the reading time was recorded in each case. Our CAD system was developed using 433 patients’ head CT images (normal, 203; haemorrhagic, 230), and haemorrhage rates were displayed as corresponding probability heat maps using U-Net and a machine learning–based false-positive removal method. Sensitivity, specificity, accuracy, and figure of merit (FOM) were calculated based on the annotations and confidence levels. Results In patient-based evaluation, the mean accuracy of all physicians significantly increased from 83.7 to 89.7% (p < 0.001) after using CAD. Additionally, accuracies of board-certificated radiologists, radiology residents, and interns were 92.5, 82.5, and 76.0% without CAD and 97.5, 90.5, and 81.0% with CAD, respectively. The mean FOM of all physicians increased from 0.78 to 0.82 (p = 0.004) after using CAD. The reading time was significantly lower when CAD (43 s) was used than when it was not (68 s, p < 0.001) for all physicians. Conclusion The CAD system developed using deep learning significantly improved the diagnostic performance and reduced the reading time among all physicians in detecting intracranial haemorrhage.