Nana Fujita
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
Nana Fujita has a total of 10 co-authors. Here are the top co-authors by number of shared publications:
- Koichiro Yasaka
The University of Tokyo, Tokyo, Japan
4 shared publications - Osamu Abe
Department of Radiology, Graduate School of Medicine, The University of Tokyo, Tokyo, Japan
4 shared publications Mao Konishiike
Department of Radiology, The University of Tokyo Hospital, Tokyo, Japan
2 shared publications- Akira Katayama
Department of Radiology, Graduate School of Medicine, The University of Tokyo, Tokyo, Japan
1 shared publication - Jun Kanzawa
Department of Radiology, The University of Tokyo, Tokyo, Japan
1 shared publication - Naomasa Okimoto
Department of Radiology, Graduate School of Medicine, The University of Tokyo, Tokyo, Japan
1 shared publication
Latest Publications
- Open Access
- 2 June 2025
- Published by Springer Nature in Emergency Radiology
- Vol. 32 (4) , 495-501
Fine-tuned large Language model for extracting newly identified acute brain infarcts based on computed tomography or magnetic resonance imaging reports
Purpose: This study aimed to develop an automated early warning system using a large language model (LLM) to identify acute to subacute brain infarction from free-text computed tomography (CT) or magnetic resonance imaging (MRI) radiology reports. Methods: In this retrospective study, 5,573, 1,883, and 834 patients were included in the training (mean age, 67.5 ± 17.2 years; 2,831 males), validation (mean age, 61.5 ± 18.3 years; 994 males), and test (mean age, 66.5 ± 16.1 years; 488 males) datasets. An LLM (Japanese Bidirectional Encoder Representations from Transformers model) was fine-tuned to classify the CT and MRI reports into three groups (group 0, newly identified acute to subacute infarction; group 1, known acute to subacute infarction or old infarction; group 2, without infarction). The training and validation processes were repeated 15 times, and the best-performing model on the validation dataset was selected to further evaluate its performance on the test dataset. Results: The best fine-tuned model exhibited sensitivities of 0.891, 0.905, and 0.959 for groups 0, 1, and 2, respectively, in the test dataset. The macrosensitivity (the average of sensitivity for all groups) and accuracy were 0.918 and 0.923, respectively. The model’s performance in extracting newly identified acute brain infarcts was high, with an area under the receiver operating characteristic curve of 0.979 (95% confidence interval, 0.956–1.000). The average prediction time was 0.115 ± 0.037 s per patient. Conclusion: A fine-tuned LLM could extract newly identified acute to subacute brain infarcts based on CT or MRI findings with high performance.
- 2 June 2025
- Published by Springer Nature in Emergency Radiology
- Vol. 32 (4) , 495-501
Open AccessFine-tuned large Language model for extracting newly identified acute brain infarcts based on computed tomography or magnetic resonance imaging reports
Purpose: This study aimed to develop an automated early warning system using a large language model (LLM) to identify acute to subacute brain infarction from free-text computed tomography (CT) or magnetic resonance imaging (MRI) radiology reports. Methods: In this retrospective study, 5,573, 1,883, and 834 patients were included in the training (mean age, 67.5 ± 17.2 years; 2,831 males), validation (mean age, 61.5 ± 18.3 years; 994 males), and test (mean age, 66.5 ± 16.1 years; 488 males) datasets. An LLM (Japanese Bidirectional Encoder Representations from Transformers model) was fine-tuned to classify the CT and MRI reports into three groups (group 0, newly identified acute to subacute infarction; group 1, known acute to subacute infarction or old infarction; group 2, without infarction). The training and validation processes were repeated 15 times, and the best-performing model on the validation dataset was selected to further evaluate its performance on the test dataset. Results: The best fine-tuned model exhibited sensitivities of 0.891, 0.905, and 0.959 for groups 0, 1, and 2, respectively, in the test dataset. The macrosensitivity (the average of sensitivity for all groups) and accuracy were 0.918 and 0.923, respectively. The model’s performance in extracting newly identified acute brain infarcts was high, with an area under the receiver operating characteristic curve of 0.979 (95% confidence interval, 0.956–1.000). The average prediction time was 0.115 ± 0.037 s per patient. Conclusion: A fine-tuned LLM could extract newly identified acute to subacute brain infarcts based on CT or MRI findings with high performance.
- Open Access
- 26 August 2024
- Published by Springer Nature in Journal of Imaging Informatics in Medicine
- Vol. 38 (2) , 865-872
The Fine-Tuned Large Language Model for Extracting the Progressive Bone Metastasis from Unstructured Radiology Reports
Early detection of patients with impending bone metastasis is crucial for prognosis improvement. This study aimed to investigate the feasibility of a fine-tuned, locally run large language model (LLM) in extracting patients with bone metastasis in unstructured Japanese radiology report and to compare its performance with manual annotation. This retrospective study included patients with “metastasis” in radiological reports (April 2018–January 2019, August–May 2022, and April–December 2023 for training, validation, and test datasets of 9559, 1498, and 7399 patients, respectively). Radiologists reviewed the clinical indication and diagnosis sections of the radiological report (used as input data) and classified them into groups 0 (no bone metastasis), 1 (progressive bone metastasis), and 2 (stable or decreased bone metastasis). The data for group 0 was under-sampled in training and test datasets due to group imbalance. The best-performing model from the validation set was subsequently tested using the testing dataset. Two additional radiologists (readers 1 and 2) were involved in classifying radiological reports within the test dataset for testing purposes. The fine-tuned LLM, reader 1, and reader 2 demonstrated an accuracy of 0.979, 0.996, and 0.993, sensitivity for groups 0/1/2 of 0.988/0.947/0.943, 1.000/1.000/0.966, and 1.000/0.982/0.954, and time required for classification (s) of 105, 2312, and 3094 in under-sampled test dataset (n = 711), respectively. Fine-tuned LLM extracted patients with bone metastasis, demonstrating satisfactory performance that was comparable to or slightly lower than manual annotation by radiologists in a noticeably shorter time.
- 26 August 2024
- Published by Springer Nature in Journal of Imaging Informatics in Medicine
- Vol. 38 (2) , 865-872
Open AccessThe Fine-Tuned Large Language Model for Extracting the Progressive Bone Metastasis from Unstructured Radiology Reports
Early detection of patients with impending bone metastasis is crucial for prognosis improvement. This study aimed to investigate the feasibility of a fine-tuned, locally run large language model (LLM) in extracting patients with bone metastasis in unstructured Japanese radiology report and to compare its performance with manual annotation. This retrospective study included patients with “metastasis” in radiological reports (April 2018–January 2019, August–May 2022, and April–December 2023 for training, validation, and test datasets of 9559, 1498, and 7399 patients, respectively). Radiologists reviewed the clinical indication and diagnosis sections of the radiological report (used as input data) and classified them into groups 0 (no bone metastasis), 1 (progressive bone metastasis), and 2 (stable or decreased bone metastasis). The data for group 0 was under-sampled in training and test datasets due to group imbalance. The best-performing model from the validation set was subsequently tested using the testing dataset. Two additional radiologists (readers 1 and 2) were involved in classifying radiological reports within the test dataset for testing purposes. The fine-tuned LLM, reader 1, and reader 2 demonstrated an accuracy of 0.979, 0.996, and 0.993, sensitivity for groups 0/1/2 of 0.988/0.947/0.943, 1.000/1.000/0.966, and 1.000/0.982/0.954, and time required for classification (s) of 105, 2312, and 3094 in under-sampled test dataset (n = 711), respectively. Fine-tuned LLM extracted patients with bone metastasis, demonstrating satisfactory performance that was comparable to or slightly lower than manual annotation by radiologists in a noticeably shorter time.
- 5 October 2023
- Published by SAGE Publications in Canadian Association of Radiologists Journal
- Vol. 75 (2) , 351-358
Detection of Vertebral Mass and Diagnosis of Spinal Cord Compression in Computed Tomography With Deep Learning Reconstruction: Comparison With Hybrid Iterative Reconstruction
Purpose: To compare the impact of deep learning reconstruction (DLR) and hybrid-iterative reconstruction (hybrid-IR) on vertebral mass depiction, detection, and diagnosis of spinal cord compression on computed tomography (CT). Methods: This retrospective study included 29 and 20 patients with and without vertebral masses. CT images were reconstructed using DLR and hybrid-IR. Three readers performed vertebral mass detection tests and evaluated the presence of spinal cord compression, the depiction of vertebral masses, and image noise. Quantitative image noise was measured by placing regions of interest on the aorta and spinal cord. Results: Deep learning reconstruction tended to improve the performance of readers with less diagnostic imaging experience in detecting vertebral masses (area under the receiver operating characteristic curve [AUC] = 0.892-0.966) compared with hybrid-IR (AUC = 0.839-0.917). Diagnostic performance in evaluating spinal cord compression in DLR (AUC = 0.887-0.995) also improved compared with that in hybrid-IR (AUC = 0.866-0.942) for some readers. The depiction of vertebral masses and subjective image noise in DLR were significantly improved compared with those in hybrid-IR ( P < .041). Quantitative image noise in DLR was also significantly reduced compared with that in hybrid-IR ( P < .001). Conclusion: Deep learning reconstruction improved the depiction of vertebral masses, which resulted in a tendency to improve the performance of CT compared to hybrid-IR in detecting vertebral masses and diagnosing spinal cord compression for some readers.
- 5 October 2023
- Published by SAGE Publications in Canadian Association of Radiologists Journal
- Vol. 75 (2) , 351-358
Detection of Vertebral Mass and Diagnosis of Spinal Cord Compression in Computed Tomography With Deep Learning Reconstruction: Comparison With Hybrid Iterative Reconstruction
Purpose: To compare the impact of deep learning reconstruction (DLR) and hybrid-iterative reconstruction (hybrid-IR) on vertebral mass depiction, detection, and diagnosis of spinal cord compression on computed tomography (CT). Methods: This retrospective study included 29 and 20 patients with and without vertebral masses. CT images were reconstructed using DLR and hybrid-IR. Three readers performed vertebral mass detection tests and evaluated the presence of spinal cord compression, the depiction of vertebral masses, and image noise. Quantitative image noise was measured by placing regions of interest on the aorta and spinal cord. Results: Deep learning reconstruction tended to improve the performance of readers with less diagnostic imaging experience in detecting vertebral masses (area under the receiver operating characteristic curve [AUC] = 0.892-0.966) compared with hybrid-IR (AUC = 0.839-0.917). Diagnostic performance in evaluating spinal cord compression in DLR (AUC = 0.887-0.995) also improved compared with that in hybrid-IR (AUC = 0.866-0.942) for some readers. The depiction of vertebral masses and subjective image noise in DLR were significantly improved compared with those in hybrid-IR ( P < .041). Quantitative image noise in DLR was also significantly reduced compared with that in hybrid-IR ( P < .001). Conclusion: Deep learning reconstruction improved the depiction of vertebral masses, which resulted in a tendency to improve the performance of CT compared to hybrid-IR in detecting vertebral masses and diagnosing spinal cord compression for some readers.
- 11 April 2023
- Published by SAGE Publications in Canadian Association of Radiologists Journal
- Vol. 74 (4) , 688-694
Assessing the Effects of Deep Learning Reconstruction on Abdominal CT Without Arm Elevation
Purpose: To evaluate the effects of deep learning reconstruction (DLR) on image quality of abdominal computed tomography (CT) in patients without arm elevation compared with hybrid-iterative reconstruction (Hybrid-IR) and filtered back projection (FBP). Methods: In this retrospective study, axial images of 26 patients who underwent CT without arm elevation were reconstructed using DLR, Hybrid-IR, and FBP. Streak artifact index (SAI) was calculated by dividing the standard deviation of CT attenuation in the liver or spleen by that in fat. Two other blinded radiologists evaluated streak artifacts on images (in the liver, spleen, and kidney), depiction of liver vessels, subjective image noise, and overall quality. They were also asked to detect space-occupying lesions other than cysts in the liver, spleen, and kidney. Results: The SAI (liver/spleen) in DLR images was significantly reduced compared with Hybrid-IR and FBP. Regarding qualitative image analysis, streak artifacts in the 3 organs, qualitative image noise, and overall quality in DLR images were rated by both readers as significantly improved compared with Hybrid-IR ( P ≤ .012) and FBP ( P < .001). Both blinded readers detected more lesions on DLR images than on Hybrid-IR and FBP ones. Conclusion: DLR resulted in significantly better-quality abdominal CT images in patients scanned without elevating their arms with reducing streak artifacts compared with Hybrid-IR and FBP.
- 11 April 2023
- Published by SAGE Publications in Canadian Association of Radiologists Journal
- Vol. 74 (4) , 688-694
Assessing the Effects of Deep Learning Reconstruction on Abdominal CT Without Arm Elevation
Purpose: To evaluate the effects of deep learning reconstruction (DLR) on image quality of abdominal computed tomography (CT) in patients without arm elevation compared with hybrid-iterative reconstruction (Hybrid-IR) and filtered back projection (FBP). Methods: In this retrospective study, axial images of 26 patients who underwent CT without arm elevation were reconstructed using DLR, Hybrid-IR, and FBP. Streak artifact index (SAI) was calculated by dividing the standard deviation of CT attenuation in the liver or spleen by that in fat. Two other blinded radiologists evaluated streak artifacts on images (in the liver, spleen, and kidney), depiction of liver vessels, subjective image noise, and overall quality. They were also asked to detect space-occupying lesions other than cysts in the liver, spleen, and kidney. Results: The SAI (liver/spleen) in DLR images was significantly reduced compared with Hybrid-IR and FBP. Regarding qualitative image analysis, streak artifacts in the 3 organs, qualitative image noise, and overall quality in DLR images were rated by both readers as significantly improved compared with Hybrid-IR ( P ≤ .012) and FBP ( P < .001). Both blinded readers detected more lesions on DLR images than on Hybrid-IR and FBP ones. Conclusion: DLR resulted in significantly better-quality abdominal CT images in patients scanned without elevating their arms with reducing streak artifacts compared with Hybrid-IR and FBP.