Development of data augmentation methods for infertility diagnosis support system

2021 Fiscal Year Final Research Report

• Project/Area Number: 20K22513
• Research Category: Grant-in-Aid for Research Activity Start-up
• Allocation Type: Multi-year Fund
• Review Section: 0403:Biomedical engineering and related fields
• Research Institution: Maizuru National College of Technology
• Principal Investigator: Mori Kentaro 舞鶴工業高等専門学校, その他部局等, 助教 (90881128)
• Project Period (FY): 2020-09-11 – 2022-03-31
• Keywords: 不妊症 / データ拡張 / 機械学習

Outline of Final Research Achievements

Data augmentation is often used to solve lack of training data problem in machine learning. In this study, we developed new data augmentation methods for medical images. We evaluated the infertility diagnosis support system based on machine learning by using the proposed augmentation data. We focused on uterus movement feature with characteristic propagation velocity, not visual features of image. Images were generated by extracting specific velocities movement from ultrasonic images. The data augmentation was performed by generating intermediate images of these images. We confirmed that the system can adequately learn with a small number of training images by using the proposed method.

Free Research Field

医用画像処理

Academic Significance and Societal Importance of the Research Achievements

本研究では、運動の伝搬速度という特定の特徴量に着目することで、効果的なデータ拡張が実現できることを発見した。医療画像には個人差によるばらつきが非常に大きいという特徴がある。本研究を通して、画像の視覚的な特徴ではなく、数値的な特徴に対しての操作がデータ拡張として効果的であることがわかった。この知見は、広い範囲の医療データに応用できるため、不妊症診断のみならず、多くの機械学習を用いた医療診断支援システムの精度向上につながる結果だと考えられる。