Yukihiro Nomura
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Yukihiro Nomura has a total of 249 co-authors. Here are the top co-authors by number of shared publications:
Takeharu Yoshikawa
Department of Computational Diagnostic Radiology and Preventive Medicine, The University of Tokyo Hospital, Tokyo, Japan
32 shared publicationsNaoto Hayashi
Department of Computational Diagnostic Radiology and Preventive Medicine, The University of Tokyo Hospital, Bunkyo-ku, Tokyo, Japan
29 shared publications- Osamu Abe
Department of Radiology, Graduate School of Medicine, The University of Tokyo, Tokyo, Japan
29 shared publications - Shouhei Hanaoka
Department of Radiology, Graduate School of Medicine, The University of Tokyo, Tokyo, Japan
29 shared publications Soichiro Miki
The Univ. of Tokyo Hospital
25 shared publicationsShouhei Hanaoka
Department of Radiology, The University of Tokyo Hospital, Tokyo, Japan
24 shared publications
Latest Publications
- Open Access
Deep-learning body composition analysis of routine CT robust to scan range, slice order and contrast phase
Purpose: To develop and validate a fully automated pipeline that measures skeletal muscle and adipose tissue at L1 and L3 on routine CT with varied scan ranges, slice orders and contrast phases, rejects levels outside the scan and converts measures to non-contrast equivalents. Materials and methods: The pipeline (sarco) combines orientation detection, L1/L3 localisation with out-of-range rejection, U-Net segmentation, six-class contrast-phase estimation, phase-specific conversion to non-contrast-equivalent values and, as an auxiliary output, muscle and marrow age estimation. We evaluated it on five single-centre datasets (4986 series): agreement with radiologist-corrected labels using five-fold cross-fitting (2025 series), artificially truncated scans, contrast-enhanced series paired with non-contrast series of the same examination, and the prognostic value of fully automated measures in 2006 patients after curative cancer surgery. TotalSegmentator served as a comparator. Results: On held-out folds, L3 skeletal muscle area (SMA) agreed with the reference (intraclass correlation coefficient 0.983; bias +1.35 cm2). On truncated scans processed with cross-fitted localisation models, the chosen L3 slice was off by half a vertebra or more in 4.7-5.6% of volumes (TotalSegmentator 11.3-14.3%). Four-group phase accuracy was 0.949, exceeding TotalSegmentator in scans shorter than 400 mm (0.952-0.960 vs 0.643-0.668), and conversion reduced the root mean square error of L3 muscle attenuation against the same-examination non-contrast series from 7.96 to 2.81 HU. Combined age estimates had a mean absolute error of 6.5 years. Automatically measured low muscle mass increased the cross-validated Uno's C of the baseline model by 0.034 (95% CI 0.009-0.053; cancer-specific survival) and 0.031 (0.016-0.048; overall survival), whereas converted L3 muscle attenuation added 0.012 (-0.005 to 0.024) for overall survival. Conclusion: The pipeline provides robust, contrast-phase-aware body composition measures from routine CT with varied scan ranges and reproduces known prognostic associations without manual input.
Open AccessDeep-learning body composition analysis of routine CT robust to scan range, slice order and contrast phase
Purpose: To develop and validate a fully automated pipeline that measures skeletal muscle and adipose tissue at L1 and L3 on routine CT with varied scan ranges, slice orders and contrast phases, rejects levels outside the scan and converts measures to non-contrast equivalents. Materials and methods: The pipeline (sarco) combines orientation detection, L1/L3 localisation with out-of-range rejection, U-Net segmentation, six-class contrast-phase estimation, phase-specific conversion to non-contrast-equivalent values and, as an auxiliary output, muscle and marrow age estimation. We evaluated it on five single-centre datasets (4986 series): agreement with radiologist-corrected labels using five-fold cross-fitting (2025 series), artificially truncated scans, contrast-enhanced series paired with non-contrast series of the same examination, and the prognostic value of fully automated measures in 2006 patients after curative cancer surgery. TotalSegmentator served as a comparator. Results: On held-out folds, L3 skeletal muscle area (SMA) agreed with the reference (intraclass correlation coefficient 0.983; bias +1.35 cm2). On truncated scans processed with cross-fitted localisation models, the chosen L3 slice was off by half a vertebra or more in 4.7-5.6% of volumes (TotalSegmentator 11.3-14.3%). Four-group phase accuracy was 0.949, exceeding TotalSegmentator in scans shorter than 400 mm (0.952-0.960 vs 0.643-0.668), and conversion reduced the root mean square error of L3 muscle attenuation against the same-examination non-contrast series from 7.96 to 2.81 HU. Combined age estimates had a mean absolute error of 6.5 years. Automatically measured low muscle mass increased the cross-validated Uno's C of the baseline model by 0.034 (95% CI 0.009-0.053; cancer-specific survival) and 0.031 (0.016-0.048; overall survival), whereas converted L3 muscle attenuation added 0.012 (-0.005 to 0.024) for overall survival. Conclusion: The pipeline provides robust, contrast-phase-aware body composition measures from routine CT with varied scan ranges and reproduces known prognostic associations without manual input.
- Open Access
- 8 October 2026
- Published byĀ MDPI AGĀ inĀ Diagnostics
- Vol.Ā 16 Ā (19) , 3244
Patient Posture-Adaptive Non-Contact Respiratory Rate Measurement Inside a Moving Ambulance Using RGB and Thermal Cameras
Background/Objectives: Monitoring and evaluating respiratory rate (RR) plays a critical role during emergency transport. Conventional RR measurement methods have primarily relied on electrocardiogram-based approaches, which are susceptible to large errors caused by patient body movement and impose a burden on patients due to their contact-based nature. This study aimed to develop a non-contact RR measurement method using RGB and thermal cameras that can be applied inside a moving ambulance. Methods: One of the key challenges in non-contact RR measurement inside an ambulance is accommodating changes in patient posture. We addressed this by patient posture-adaptive automatic image registration between RGB and thermal cameras and developing a nostril detection method robust to posture changes. Results: The image registration error of the proposed method was evaluated at various distances from the camera, and the error was confirmed to remain within 1 pixel. RR measurement experiments simulating various respiratory conditions were then conducted inside a moving ambulance. The average of mean absolute error (MAE) was 0.95, 0.59, and 2.80 breaths per minute (breaths/min) for bradypnea, normal, and tachypnea, respectively, in the supine position, and 1.08, 1.90, and 4.38 breaths/min, respectively, in the semi-sitting position. Evaluation of the proposed nostril detection method further confirmed improved RR estimation accuracy compared with the conventional method. Conclusions: This study achieved patient posture-adaptive automatic image registration and robust nostril detection and successfully conducted experiments under realistic and challenging conditions, representing a substantial step toward the real-world implementation of a non-contact RR measurement system in ambulances.
- 8 October 2026
- Published byĀ MDPI AGĀ inĀ Diagnostics
- Vol.Ā 16 Ā (19) , 3244
Open AccessPatient Posture-Adaptive Non-Contact Respiratory Rate Measurement Inside a Moving Ambulance Using RGB and Thermal Cameras
Background/Objectives: Monitoring and evaluating respiratory rate (RR) plays a critical role during emergency transport. Conventional RR measurement methods have primarily relied on electrocardiogram-based approaches, which are susceptible to large errors caused by patient body movement and impose a burden on patients due to their contact-based nature. This study aimed to develop a non-contact RR measurement method using RGB and thermal cameras that can be applied inside a moving ambulance. Methods: One of the key challenges in non-contact RR measurement inside an ambulance is accommodating changes in patient posture. We addressed this by patient posture-adaptive automatic image registration between RGB and thermal cameras and developing a nostril detection method robust to posture changes. Results: The image registration error of the proposed method was evaluated at various distances from the camera, and the error was confirmed to remain within 1 pixel. RR measurement experiments simulating various respiratory conditions were then conducted inside a moving ambulance. The average of mean absolute error (MAE) was 0.95, 0.59, and 2.80 breaths per minute (breaths/min) for bradypnea, normal, and tachypnea, respectively, in the supine position, and 1.08, 1.90, and 4.38 breaths/min, respectively, in the semi-sitting position. Evaluation of the proposed nostril detection method further confirmed improved RR estimation accuracy compared with the conventional method. Conclusions: This study achieved patient posture-adaptive automatic image registration and robust nostril detection and successfully conducted experiments under realistic and challenging conditions, representing a substantial step toward the real-world implementation of a non-contact RR measurement system in ambulances.
- Open Access
- 27 September 2026
- Published byĀ Springer NatureĀ inĀ International Journal of Computer Assisted Radiology and Surgery
- p. 1-10
Performance changes in automated lesion detection under federated learning with sequential institution addition
Purpose: Federated learning (FL) enables multiple institutions to collaboratively train machine learning models while keeping data local and has attracted attention in medical image processing, including computer-aided detection (CAD). In FL, performance is expected to improve through retraining as additional institutions participate. The purpose of this study was to investigate how CAD software performance changes as the number of participating institutions is sequentially increased within an FL framework. Methods: We used two types of CAD software for cerebral aneurysm detection in magnetic resonance (MR) angiography images and brain metastasis detection in contrast-enhanced T1-weighted MR images. Datasets from different institutions or scanner vendors were treated as independent FL clients and incorporated sequentially. Training strategies included from-scratch training, fine-tuning (FT) of selected layers, and full fine-tuning (FFT) of all parameters. Performance was assessed using the competition performance metric on test sets from the initial institutions as well as from all participating institutions. Results: For both CAD software types, sequential institution addition combined with FT generally showed higher median performance changes than from-scratch training. FT showed performance comparable to that of FFT while requiring substantially fewer trainable parameters. Performance improvements generally accumulated with sequential institution addition, whereas simultaneous addition resulted in less consistent improvements. Conclusions: Sequential institution addition under FL may improve CAD software performance when combined with appropriate FT strategies. FT that updates only a subset of layers may achieve performance changes comparable to those of FFT while requiring substantially fewer trainable parameters across different lesion detection tasks.
- 27 September 2026
- Published byĀ Springer NatureĀ inĀ International Journal of Computer Assisted Radiology and Surgery
- p. 1-10
Open AccessPerformance changes in automated lesion detection under federated learning with sequential institution addition
Purpose: Federated learning (FL) enables multiple institutions to collaboratively train machine learning models while keeping data local and has attracted attention in medical image processing, including computer-aided detection (CAD). In FL, performance is expected to improve through retraining as additional institutions participate. The purpose of this study was to investigate how CAD software performance changes as the number of participating institutions is sequentially increased within an FL framework. Methods: We used two types of CAD software for cerebral aneurysm detection in magnetic resonance (MR) angiography images and brain metastasis detection in contrast-enhanced T1-weighted MR images. Datasets from different institutions or scanner vendors were treated as independent FL clients and incorporated sequentially. Training strategies included from-scratch training, fine-tuning (FT) of selected layers, and full fine-tuning (FFT) of all parameters. Performance was assessed using the competition performance metric on test sets from the initial institutions as well as from all participating institutions. Results: For both CAD software types, sequential institution addition combined with FT generally showed higher median performance changes than from-scratch training. FT showed performance comparable to that of FFT while requiring substantially fewer trainable parameters. Performance improvements generally accumulated with sequential institution addition, whereas simultaneous addition resulted in less consistent improvements. Conclusions: Sequential institution addition under FL may improve CAD software performance when combined with appropriate FT strategies. FT that updates only a subset of layers may achieve performance changes comparable to those of FFT while requiring substantially fewer trainable parameters across different lesion detection tasks.
- Open Access
- 30 June 2026
- Published byĀ Springer NatureĀ inĀ Journal of Imaging Informatics in Medicine
Automatic Phase and Sequence Identification in Gd-EOB-DTPAāEnhanced Liver MRI Using Deep Convolutional and Sequential Learning
The purpose of this study is to develop and validate a deep learning model for automatic identification of acquisition sequences (including T1-weighted sequences with various dynamic contrast-enhanced phases and other auxiliary sequences) in gadolinium ethoxybenzyl diethylenetriamine pentaacetic acid (Gd-EOB-DTPA)āenhanced liver MRI, enabling automated examination-level data curation. This retrospective study included internal Gd-EOB-DTPAāenhanced liver MRI examinations acquired at our institution between June 2018 and May 2020, with independent external test datasets from three additional institutions. Each examination comprised multiple dynamic contrast-enhanced phases and auxiliary sequences, resulting in 13 predefined label categories. A deep learning pipeline was constructed using a convolutional neural network (ConvNeXt) for feature extraction, followed by sequential models (gated recurrent unit [GRU]-based or transformer-based) to model temporal relationships across series. Models were trained using series-level 3D image volumes without reliance on textual Digital Imaging and Communications in Medicine metadata. Model selection was performed on a validation set. Performance was evaluated on internal and external test sets using examination-level accuracy, complete-correct rate, and category-level accuracy. Sequential modeling substantially improved performance compared with a convolution-only baseline. The ConvNeXt + GRU model achieved the highest examination-level accuracy and was selected for final evaluation. Dynamic contrast-enhanced phases were identified with high accuracy across datasets. Reduced performance on the external test set was mainly observed for auxiliary sequences with high inter-institutional variability, particularly in T2-weighted imaging. The proposed framework enables accurate automatic identification of dynamic phases and auxiliary sequences in Gd-EOB-DTPAāenhanced liver MRI, supporting robust examination-level data organization in multicenter studies.
- 30 June 2026
- Published byĀ Springer NatureĀ inĀ Journal of Imaging Informatics in Medicine
Open AccessAutomatic Phase and Sequence Identification in Gd-EOB-DTPAāEnhanced Liver MRI Using Deep Convolutional and Sequential Learning
The purpose of this study is to develop and validate a deep learning model for automatic identification of acquisition sequences (including T1-weighted sequences with various dynamic contrast-enhanced phases and other auxiliary sequences) in gadolinium ethoxybenzyl diethylenetriamine pentaacetic acid (Gd-EOB-DTPA)āenhanced liver MRI, enabling automated examination-level data curation. This retrospective study included internal Gd-EOB-DTPAāenhanced liver MRI examinations acquired at our institution between June 2018 and May 2020, with independent external test datasets from three additional institutions. Each examination comprised multiple dynamic contrast-enhanced phases and auxiliary sequences, resulting in 13 predefined label categories. A deep learning pipeline was constructed using a convolutional neural network (ConvNeXt) for feature extraction, followed by sequential models (gated recurrent unit [GRU]-based or transformer-based) to model temporal relationships across series. Models were trained using series-level 3D image volumes without reliance on textual Digital Imaging and Communications in Medicine metadata. Model selection was performed on a validation set. Performance was evaluated on internal and external test sets using examination-level accuracy, complete-correct rate, and category-level accuracy. Sequential modeling substantially improved performance compared with a convolution-only baseline. The ConvNeXt + GRU model achieved the highest examination-level accuracy and was selected for final evaluation. Dynamic contrast-enhanced phases were identified with high accuracy across datasets. Reduced performance on the external test set was mainly observed for auxiliary sequences with high inter-institutional variability, particularly in T2-weighted imaging. The proposed framework enables accurate automatic identification of dynamic phases and auxiliary sequences in Gd-EOB-DTPAāenhanced liver MRI, supporting robust examination-level data organization in multicenter studies.
- Open Access
- 19 June 2026
- Published byĀ MDPI AGĀ inĀ Applied Sciences
- Vol.Ā 16 Ā (12) , 6217
Research and Application of Digital Tongue Diagnosis Technology in Tongue Image Characteristics of Different Ethnic Groups
Background: Tongue diagnosis is a fundamental diagnostic method in traditional medicine. Studies restricted to single ethnic groups may introduce bias and limit the clinical applicability of digital tongue diagnosis across diverse populations. Objectives: This study examined differences in tongue image features between Japanese and Brazilian (Caucasian ancestry) participants using digital tongue diagnosis technology and explored potential influencing factors. Methods: Tongue images were collected from 143 Japanese and 116 Brazilian participants attending traditional medicine clinics in Japan and Brazil. An independently developed tongue image analysis system (TIAS) was employed to extract shape, texture (gray level co-occurrence matrix), color (L*a*b color space), and deep-learning derived features (crack, prickle, tooth-mark, peel, greasy coating, stasis). Statistical analyses and machine learning models with SHAP explainability were used to compare features and identify key classification parameters. Results: Significant inter-group differences were observed in tongue shape, texture parameters (particularly at the root and tip), color parameters (especially middle-a-mean, middle-b-mean, tip-a-mean, and tip-b-mean), and deep features. The Japanese group showed a markedly higher prevalence of greasy coating (72.03% vs. 41.38%, p < 0.001) and stasis. Machine learning analysis revealed that the b value in the middle region of the tongue (middle-b-mean) contributed most strongly to the classification of greasy coating. Conclusions: The digital tongue image analysis system enables accurate and objective quantification of tongue features. Pronounced ethnic differences exist, particularly in the distribution of greasy coating. The middle-b-mean has the greatest impact on greasy coating classification. These findings underscore the importance of considering ethnic background when developing digital tongue diagnosis systems.
- 19 June 2026
- Published byĀ MDPI AGĀ inĀ Applied Sciences
- Vol.Ā 16 Ā (12) , 6217
Open AccessResearch and Application of Digital Tongue Diagnosis Technology in Tongue Image Characteristics of Different Ethnic Groups
Background: Tongue diagnosis is a fundamental diagnostic method in traditional medicine. Studies restricted to single ethnic groups may introduce bias and limit the clinical applicability of digital tongue diagnosis across diverse populations. Objectives: This study examined differences in tongue image features between Japanese and Brazilian (Caucasian ancestry) participants using digital tongue diagnosis technology and explored potential influencing factors. Methods: Tongue images were collected from 143 Japanese and 116 Brazilian participants attending traditional medicine clinics in Japan and Brazil. An independently developed tongue image analysis system (TIAS) was employed to extract shape, texture (gray level co-occurrence matrix), color (L*a*b color space), and deep-learning derived features (crack, prickle, tooth-mark, peel, greasy coating, stasis). Statistical analyses and machine learning models with SHAP explainability were used to compare features and identify key classification parameters. Results: Significant inter-group differences were observed in tongue shape, texture parameters (particularly at the root and tip), color parameters (especially middle-a-mean, middle-b-mean, tip-a-mean, and tip-b-mean), and deep features. The Japanese group showed a markedly higher prevalence of greasy coating (72.03% vs. 41.38%, p < 0.001) and stasis. Machine learning analysis revealed that the b value in the middle region of the tongue (middle-b-mean) contributed most strongly to the classification of greasy coating. Conclusions: The digital tongue image analysis system enables accurate and objective quantification of tongue features. Pronounced ethnic differences exist, particularly in the distribution of greasy coating. The middle-b-mean has the greatest impact on greasy coating classification. These findings underscore the importance of considering ethnic background when developing digital tongue diagnosis systems.
- Open Access
Age-related nonlinear trajectories of abdominal organ volumes on CT: a longitudinal study
Objectives: To characterize longitudinal age-related changes in abdominal organ volumes using CT volumetry and to model nonlinear trajectories across multiple organs. Materials & Methods: This retrospective single-center study included adults who underwent whole-body screening low-dose CT between 2006 and 2017. Subjects with at least eight examinations during a follow-up period of at least 78 months were included. After applying exclusion criteria, 700 participants with 6,739 CT series were analyzed. Non-contrast CT images were processed using automated organ segmentation, and volumes of the liver, pancreas, spleen, and kidneys were quantified. Longitudinal changes were modeled using generalized additive mixed models with sex-specific smooth functions of age and subject-level random effects. Age-dependent rates of change were estimated from model derivatives. Results: A total of 700 participants (mean age, 56.9 ± 9.8 years, 29.6% women) were evaluated. Liver, pancreas, and kidney volumes showed mild increases or plateaued at approximately 40ā60 years of age, depending on the organ, and were followed by gradual declines with advancing age, whereas splenic volume showed a progressive decrease across the age range. These patterns showed nonlinear age dependence. The transition from positive to negative change rates tended to occur earlier in women than in men for several organs, particularly the liver and kidneys. Conclusion: Longitudinal CT analysis demonstrated nonlinear age-related changes in abdominal organ volumes, with organ-specific trajectories and sex-related differences in the timing and magnitude of volume changes. Question: How do abdominal organ volumes change longitudinally with age, and can their trajectories be characterized for each organ? Findings: Longitudinal CT analysis demonstrated nonlinear, organ-specific volume trajectories, with transitions from stability to decline around 40ā60 years and earlier transitions in women than men. Clinical Relevance: Longitudinal reference patterns of abdominal organ volumes on CT improve the interpretation of age-related changes and support more accurate differentiation between physiological variation and disease-related volume alterations.
Open AccessAge-related nonlinear trajectories of abdominal organ volumes on CT: a longitudinal study
Objectives: To characterize longitudinal age-related changes in abdominal organ volumes using CT volumetry and to model nonlinear trajectories across multiple organs. Materials & Methods: This retrospective single-center study included adults who underwent whole-body screening low-dose CT between 2006 and 2017. Subjects with at least eight examinations during a follow-up period of at least 78 months were included. After applying exclusion criteria, 700 participants with 6,739 CT series were analyzed. Non-contrast CT images were processed using automated organ segmentation, and volumes of the liver, pancreas, spleen, and kidneys were quantified. Longitudinal changes were modeled using generalized additive mixed models with sex-specific smooth functions of age and subject-level random effects. Age-dependent rates of change were estimated from model derivatives. Results: A total of 700 participants (mean age, 56.9 ± 9.8 years, 29.6% women) were evaluated. Liver, pancreas, and kidney volumes showed mild increases or plateaued at approximately 40ā60 years of age, depending on the organ, and were followed by gradual declines with advancing age, whereas splenic volume showed a progressive decrease across the age range. These patterns showed nonlinear age dependence. The transition from positive to negative change rates tended to occur earlier in women than in men for several organs, particularly the liver and kidneys. Conclusion: Longitudinal CT analysis demonstrated nonlinear age-related changes in abdominal organ volumes, with organ-specific trajectories and sex-related differences in the timing and magnitude of volume changes. Question: How do abdominal organ volumes change longitudinally with age, and can their trajectories be characterized for each organ? Findings: Longitudinal CT analysis demonstrated nonlinear, organ-specific volume trajectories, with transitions from stability to decline around 40ā60 years and earlier transitions in women than men. Clinical Relevance: Longitudinal reference patterns of abdominal organ volumes on CT improve the interpretation of age-related changes and support more accurate differentiation between physiological variation and disease-related volume alterations.
- Open Access
- 2 April 2026
- Published byĀ Springer NatureĀ inĀ Japanese Journal of Radiology
- Vol.Ā 44 Ā (8) , 1-8
AI achieves board-level performance on the Japan diagnostic radiology board examination through direct image interpretation
Purpose: To evaluate text-only versus vision-enabled performance of late-2025 large language models (LLMs) on the Japan Diagnostic Radiology Board Examination (JDRBE) and compare model performance with newly board-certified radiologists. Materials and methods: Image-based questions from the JDRBE 2021 and 2023ā2025 were collected, and ground truth answers were determined by expert consensus. Four commercial multimodal LLMs were evaluated: Gemini 2.5 Pro (March 2025, baseline), Gemini 3 Pro, GPT-5.1, and Claude Opus 4.5 (all released in November 2025). Each question was answered with image input (āvisionā) and without images (ātext-onlyā). For the JDRBE 2025, subjective legitimacy of responses was independently rated by two radiologists using a five-point Likert scale, and low-rated responses were further analyzed by error type. Additional analyses on the JDRBE 2025 subset included image-shuffling and multi-run variability assessment (five runs). Model accuracies were also compared with those of five newly board-certified radiologists who passed the JDRBE 2025. Results: Gemini 3 Pro achieved the highest accuracy among all models, scoring 85.3% (279/327) in the vision condition and significantly outperforming its text-only accuracy (74.3%, P < 0.001). Gemini 2.5 Pro and Claude Opus 4.5 also improved with image input, whereas GPT-5.1 did not. For the JDRBE 2025, Gemini 3 Pro in the vision condition received the highest legitimacy ratings, and its accuracy (88%) was above the range observed in a reference group of five newly board-certified radiologists (65%ā83%), but hallucination was still the most common error type. Image-shuffling analysis using the 2025 subset showed no performance gain in all models, supporting reliance on visual input. Multi-run variability analysis showed high agreement across runs. Conclusion: Among late-2025 commercial LLMs, Gemini 3 Pro demonstrated board-level performance on the JDRBE through direct medical image interpretation. Secondary The performance of vision-enabled large language models on the Japan Diagnostic Radiology Board Examination was evaluated. Among the models released in November 2025, Gemini 3 Pro demonstrated significant capabilities in direct medical image interpretation, achieving accuracy above that of a reference group of five newly board-certified radiologists.
- 2 April 2026
- Published byĀ Springer NatureĀ inĀ Japanese Journal of Radiology
- Vol.Ā 44 Ā (8) , 1-8
Open AccessAI achieves board-level performance on the Japan diagnostic radiology board examination through direct image interpretation
Purpose: To evaluate text-only versus vision-enabled performance of late-2025 large language models (LLMs) on the Japan Diagnostic Radiology Board Examination (JDRBE) and compare model performance with newly board-certified radiologists. Materials and methods: Image-based questions from the JDRBE 2021 and 2023ā2025 were collected, and ground truth answers were determined by expert consensus. Four commercial multimodal LLMs were evaluated: Gemini 2.5 Pro (March 2025, baseline), Gemini 3 Pro, GPT-5.1, and Claude Opus 4.5 (all released in November 2025). Each question was answered with image input (āvisionā) and without images (ātext-onlyā). For the JDRBE 2025, subjective legitimacy of responses was independently rated by two radiologists using a five-point Likert scale, and low-rated responses were further analyzed by error type. Additional analyses on the JDRBE 2025 subset included image-shuffling and multi-run variability assessment (five runs). Model accuracies were also compared with those of five newly board-certified radiologists who passed the JDRBE 2025. Results: Gemini 3 Pro achieved the highest accuracy among all models, scoring 85.3% (279/327) in the vision condition and significantly outperforming its text-only accuracy (74.3%, P < 0.001). Gemini 2.5 Pro and Claude Opus 4.5 also improved with image input, whereas GPT-5.1 did not. For the JDRBE 2025, Gemini 3 Pro in the vision condition received the highest legitimacy ratings, and its accuracy (88%) was above the range observed in a reference group of five newly board-certified radiologists (65%ā83%), but hallucination was still the most common error type. Image-shuffling analysis using the 2025 subset showed no performance gain in all models, supporting reliance on visual input. Multi-run variability analysis showed high agreement across runs. Conclusion: Among late-2025 commercial LLMs, Gemini 3 Pro demonstrated board-level performance on the JDRBE through direct medical image interpretation. Secondary The performance of vision-enabled large language models on the Japan Diagnostic Radiology Board Examination was evaluated. Among the models released in November 2025, Gemini 3 Pro demonstrated significant capabilities in direct medical image interpretation, achieving accuracy above that of a reference group of five newly board-certified radiologists.
- Open Access
- 3 March 2026
- Published byĀ MDPI AGĀ inĀ Instruments
- Vol.Ā 10 Ā (1) , 15
Smartphone-Based Quantitative Measurement of Capillary Refill Time
Capillary refill time (CRT) is widely used in pediatric and emergency medicine as an indicator of peripheral circulation. CRT is defined as the time required for the skin to return to its original color after external compression is applied and then released. In current clinical practice, however, CRT assessment remains qualitative and relies heavily on the magnitude and consistency of compression applied by the measurer, as well as on subjective visual color perception, which together result in limited measurement reliability. To improve measurement reliability, several quantitative CRT measurement devices have been developed. Nevertheless, these devices are dedicated specifically to CRT measurement, which limits their versatility and complicates clinical implementation. In this study, we developed a simple and quantitative CRT measurement method using a smartphone. Based on skin color changes captured by the rear camera, we proposed a method to assess the adequacy of the applied compression force and implemented an application to calculate CRT. In addition, we investigated an algorithm to reduce the influence of pulse waves observed in the post-release waveform, enabling more stable CRT estimation. Furthermore, a dedicated smartphone case was designed to immobilize the finger during measurement, thereby improving measurement reliability. The feasibility of the proposed method was evaluated by examining agreement with a previously developed CRT measurement device and by assessing intraexaminer reliability, confirming its effectiveness.
- 3 March 2026
- Published byĀ MDPI AGĀ inĀ Instruments
- Vol.Ā 10 Ā (1) , 15
Open AccessSmartphone-Based Quantitative Measurement of Capillary Refill Time
Capillary refill time (CRT) is widely used in pediatric and emergency medicine as an indicator of peripheral circulation. CRT is defined as the time required for the skin to return to its original color after external compression is applied and then released. In current clinical practice, however, CRT assessment remains qualitative and relies heavily on the magnitude and consistency of compression applied by the measurer, as well as on subjective visual color perception, which together result in limited measurement reliability. To improve measurement reliability, several quantitative CRT measurement devices have been developed. Nevertheless, these devices are dedicated specifically to CRT measurement, which limits their versatility and complicates clinical implementation. In this study, we developed a simple and quantitative CRT measurement method using a smartphone. Based on skin color changes captured by the rear camera, we proposed a method to assess the adequacy of the applied compression force and implemented an application to calculate CRT. In addition, we investigated an algorithm to reduce the influence of pulse waves observed in the post-release waveform, enabling more stable CRT estimation. Furthermore, a dedicated smartphone case was designed to immobilize the finger during measurement, thereby improving measurement reliability. The feasibility of the proposed method was evaluated by examining agreement with a previously developed CRT measurement device and by assessing intraexaminer reliability, confirming its effectiveness.
- Open Access
- 23 December 2025
- Published byĀ JMIR Publications Inc.Ā inĀ JMIR Medical Education
- Vol.Ā 11, e78214
Trends in the Japanese National Medical Licensing Examination: Cross-Sectional Study
Background: The Japanese National Medical Licensing Examination (NMLE) is mandatory for all medical graduates seeking to become licensed physicians in Japan. Given the cultural emphasis on summative assessment, the NMLE has had a significant impact on Japanese medical education. Although the NMLE Content Guidelines have been revised approximately every five years over the last 2 decades, objective literature analyzing how the examination itself has evolved is absent. Objective: To provide a holistic view of the trends of the actual examination over time, this study used a combined rule-based and data-driven approach. We primarily focused on classifying the items according to the perspectives outlined in the NMLE Content Guidelines, complementing this approach with a natural language processing technique called topic modeling to identify latent topics. Methods: We collected publicly available NMLE data for 2001-2024. Six examination iterations (2880 items) were manually classified from 3 perspectives (level, content, and taxonomy) based on pre-established rules derived from the guidelines. Temporal trends within each classification were evaluated using the Cochran-Armitage test. Additionally, we conducted topic modeling for all 24 examination iterations (11,540 items) using the bidirectional encoder representations from transformersābased topic modeling framework. Temporal trends were traced using linear regression models of topic frequencies to identify topics growing in prominence. Results: In the level classification, the proportion of items addressing common or emergent diseases increased from 60% (115/193) to 76% (111/147; P<.001). In the content classification, the proportion of items assessing knowledge of pathophysiology decreased from 52% (237/459) to 33% (98/293; P<.001), whereas the proportion assessing practical knowledge of primary emergency care increased from 21% (95/459) to 29% (84/293; P<.001). In the taxonomy classification, the proportion of items that could be answered solely through simple recall of knowledge decreased from 51% (279/550) to 30% (118/400; P<.001), while the proportion assessing advanced analytical skills, such as interpreting and evaluating the meaning of each answer choice according to the given context, increased from 4% (21/550) to 19% (75/400; P<.001). Topic modeling identified 25 distinct topics, of which 10 exhibited an increasing trend. Nonāorgan-specific topics with notable increases included ācomprehensive clinical items,ā āaccountability in medical practice and patientsā rights,ā ācare, daily living support, and community health care,ā and āinfection control and safety management in basic clinical procedures.ā Conclusions: This study identified significant shifts in the Japanese NMLE over the past 2 decades, suggesting that Japanese undergraduate medical education is evolving to place greater importance on practical problem-solving abilities than on rote memorization. This study also identified latent topics that showed increased prominence, possibly reflecting underlying social conditions.
- 23 December 2025
- Published byĀ JMIR Publications Inc.Ā inĀ JMIR Medical Education
- Vol.Ā 11, e78214
Open AccessTrends in the Japanese National Medical Licensing Examination: Cross-Sectional Study
Background: The Japanese National Medical Licensing Examination (NMLE) is mandatory for all medical graduates seeking to become licensed physicians in Japan. Given the cultural emphasis on summative assessment, the NMLE has had a significant impact on Japanese medical education. Although the NMLE Content Guidelines have been revised approximately every five years over the last 2 decades, objective literature analyzing how the examination itself has evolved is absent. Objective: To provide a holistic view of the trends of the actual examination over time, this study used a combined rule-based and data-driven approach. We primarily focused on classifying the items according to the perspectives outlined in the NMLE Content Guidelines, complementing this approach with a natural language processing technique called topic modeling to identify latent topics. Methods: We collected publicly available NMLE data for 2001-2024. Six examination iterations (2880 items) were manually classified from 3 perspectives (level, content, and taxonomy) based on pre-established rules derived from the guidelines. Temporal trends within each classification were evaluated using the Cochran-Armitage test. Additionally, we conducted topic modeling for all 24 examination iterations (11,540 items) using the bidirectional encoder representations from transformersābased topic modeling framework. Temporal trends were traced using linear regression models of topic frequencies to identify topics growing in prominence. Results: In the level classification, the proportion of items addressing common or emergent diseases increased from 60% (115/193) to 76% (111/147; P<.001). In the content classification, the proportion of items assessing knowledge of pathophysiology decreased from 52% (237/459) to 33% (98/293; P<.001), whereas the proportion assessing practical knowledge of primary emergency care increased from 21% (95/459) to 29% (84/293; P<.001). In the taxonomy classification, the proportion of items that could be answered solely through simple recall of knowledge decreased from 51% (279/550) to 30% (118/400; P<.001), while the proportion assessing advanced analytical skills, such as interpreting and evaluating the meaning of each answer choice according to the given context, increased from 4% (21/550) to 19% (75/400; P<.001). Topic modeling identified 25 distinct topics, of which 10 exhibited an increasing trend. Nonāorgan-specific topics with notable increases included ācomprehensive clinical items,ā āaccountability in medical practice and patientsā rights,ā ācare, daily living support, and community health care,ā and āinfection control and safety management in basic clinical procedures.ā Conclusions: This study identified significant shifts in the Japanese NMLE over the past 2 decades, suggesting that Japanese undergraduate medical education is evolving to place greater importance on practical problem-solving abilities than on rote memorization. This study also identified latent topics that showed increased prominence, possibly reflecting underlying social conditions.
- Open Access
- 17 September 2025
- Published byĀ ElsevierĀ inĀ Computers in Biology and Medicine
- Vol.Ā 197 Ā (Pt B) , 111064
Classification of peripheral pulmonary lesions in Endobronchial ultrasonography image using a multi-branch framework and voting ensemble
- 17 September 2025
- Published byĀ ElsevierĀ inĀ Computers in Biology and Medicine
- Vol.Ā 197 Ā (Pt B) , 111064
Open AccessClassification of peripheral pulmonary lesions in Endobronchial ultrasonography image using a multi-branch framework and voting ensemble