2021 Fiscal Year Final Research Report
Next-Generation Ultrasonic Beamformer with Innovative Transducer and Deep Learning
| Project/Area Number |
19K22891
|
|---|
| Research Category |
Grant-in-Aid for Challenging Research (Exploratory)
|
| Allocation Type | Multi-year Fund |
|---|
| Review Section |
Medium-sized Section 62:Applied informatics and related fields
|
|---|
| Research Institution | University of Toyama |
|---|
Principal Investigator |
|
|---|
| Co-Investigator(Kenkyū-buntansha) |
藤原 久美子 富山大学, 学術研究部医学系, 助教 (60404737)
長岡 亮 富山大学, 学術研究部工学系, 准教授 (60781648)
高 尚策 富山大学, 学術研究部工学系, 准教授 (60734572)
|
|---|
| Project Period (FY) |
2019-06-28 – 2022-03-31
|
| Keywords | 超音波 / ビームフォーミング / 深層学習 |
|---|
| Outline of Final Research Achievements |
A beamformer is a indispensable component in medical ultrasound imaging. This study explored beamforming methods through deep learning for accurate reconstruction of target structures. In the present study, a simulation platform was first developed for generation of numerous data mimicking ultrasonic echoes from biological tissues because a lot of data are required to train the artificial neural network appropriately. Generating data in simulation has another merit, i.e., the true property of the source of the simulated echo signals is known and can be used as a teacher data for training of the artificial neural network. Using the neural network trained by such simulated data sets, ultrasonic images were obtained with less influences of the characteristics of the ultrasonic imaging system. The artificial neural network was also trained by the data sets produced with a more sophisticated beamformer, and the trained neural network produced more accurate ultrasonic images.
|
| Free Research Field |
医用超音波工学
|
| Academic Significance and Societal Importance of the Research Achievements |
医用超音波画像の品質はビームフォーマに大きく依存するため，ビームフォーマの高性能化に関する研究開発が精力的に行われている．一方，そのような高度なビームフォーマは計算負荷が高く，医用超音波画像の特徴の1つであるリアルタイム性を損なわずに実装できない場合が多い．本研究は，深層学習を用いて高度なビームフォーマと同様の画像を出力する手法について検討を行い，超音波伝搬シミュレーションにより人工ニューラルネットワークを学習させるための多数のデータを生成することで深層学習を用いてより高精度な画像が得られることを示した．本研究成果は，医用超音波画像の品質向上に資するものである．
|
