2022 Fiscal Year Final Research Report
Construction of a comprehensive diagnostic workflow for salivary gland tumors based on the comparison of images and pathology
Project/Area Number |
19K08092
|
Research Category |
Grant-in-Aid for Scientific Research (C)
|
Allocation Type | Multi-year Fund |
Section | 一般 |
Review Section |
Basic Section 52040:Radiological sciences-related
|
Research Institution | Chiba University |
Principal Investigator |
|
Project Period (FY) |
2019-04-01 – 2023-03-31
|
Keywords | 深層学習 / 唾液腺腫瘍 / 画像診断 / MRI |
Outline of Final Research Achievements |
We aimed to construct a benign-malignant diagnostic model for salivary gland tumors using deep learning techniques, specifically convolutional neural networks, and comparing the results with pathological findings. To facilitate deep learning, tumor regions were manually extracted by a radiologist from MRI images, including T2-weighted images, dynamic MRI, and diffusion-weighted images, followed by image preprocessing. By comparing the performance of individual modality training and combined training using all modalities, we found that the combined training achieved a higher diagnostic accuracy than training with individual modalities alone. In fact, the combined training outperformed specialized radiologists in the head and neck region, achieving a diagnostic accuracy of 87%.
|
Free Research Field |
放射線画像診断
|
Academic Significance and Societal Importance of the Research Achievements |
唾液腺腫瘍は稀な腫瘍であり、種類や内部構造が多様のこともあり、良性悪性の鑑別をはじめとした画像診断は難しい。頻度が低く、症例数が少ないため、深層学習を使った評価は今までほとんどされていなかった。本研究では、唾液腺腫瘍のMRIに対し、深層学習を使うことによって、頭頸部領域専門の放射線科医よりも高い正診率を持つモデルを作成した。理論上は腫瘍範囲を設定することができれば利用可能であり、このモデルを用いれば非放射線科、頭頸部領域以外を専門とする放射線科医であっても、正確な診断を行うことが可能になると予想された。
|