2022 Fiscal Year Final Research Report
Deep learning for automatic diagnosis of chest X-ray images and its computer-aided diagnosis
Project/Area Number |
19K17232
|
Research Category |
Grant-in-Aid for Early-Career Scientists
|
Allocation Type | Multi-year Fund |
Review Section |
Basic Section 52040:Radiological sciences-related
|
Research Institution | Kobe University |
Principal Investigator |
Mizuho Nishio 神戸大学, 医学部附属病院, 助教 (50581998)
|
Project Period (FY) |
2019-04-01 – 2023-03-31
|
Keywords | 新型コロナウィルス肺炎 / 胸部単純レントゲン写真 / 深層学習 / 人工知能 / 医用画像処理 |
Outline of Final Research Achievements |
The purpose of this study was to (i) develop deep-learning-based automatic diagnosis model (DL model) of coronavirus disease 2019 (COVID) on chest X-rays (CXRs), (ii) evaluate the diagnostic performance of our system by external validation, and (iii) investigate whether the diagnostic performance of radiologists was improved using our model. Our model was developed using EfficientNet and more than 10000 CXRs. 180 CXRs was used for external validation of our model. Eight radiologists performed two reading sessions of the 180 CXRs with and without our DL model. The accuracy of our DL model was 0.733, and that of the eight radiologists without DL was 0.696 ± 0.031. There was a significant difference in AUC between the radiologists with and without DL for COVID vs. NORMAL or PNEUMONIA (p = 0.0038). Our DL model significantly improved the diagnostic performance of radiologists for COVID vs. NORMAL or PNEUMONIA.
|
Free Research Field |
放射線診断学
|
Academic Significance and Societal Importance of the Research Achievements |
今回の研究の結果から、胸部単純レントゲン写真(CXR)で新型コロナウィルス肺炎の診断を深層学習のソフトウェアで行うことは可能であり、開発されたソフトウェアは放射線科医の診断能を有意に改善させることが分かった。
|