2023 Fiscal Year Final Research Report
Non-invasive lumbar myelography using dual-energy CT and deep learning
| Project/Area Number |
21K15854
|
|---|
| Research Category |
Grant-in-Aid for Early-Career Scientists
|
| Allocation Type | Multi-year Fund |
|---|
| Review Section |
Basic Section 52040:Radiological sciences-related
|
|---|
| Research Institution | Akita Cerebrospinal and Cardiovascular Center |
|---|
Principal Investigator |
Shinohara Yuki 秋田県立循環器・脳脊髄センター(研究所), 放射線医学研究部, 主任研究員 (60462470)
|
|---|
| Project Period (FY) |
2021-04-01 – 2024-03-31
|
| Keywords | 腰椎ミエログラフィー / デュアルエネルギーCT / 逐次近似画像再構成 / 仮想単色X線画像 / 金属アーチファクト低減技術 / 手術支援画像 / 腰部脊柱管狭窄症 / 腰椎椎間板ヘルニア |
|---|
| Outline of Final Research Achievements |
The purpose of this study was to create lumbar spine surgery-assisted image from non-contrast dual-energy CT (DECT). In virtual monochromatic imaging (Mono+, Siemens), the best contrast between dural canal and ligamentum flavum or dural canal and intervertebral disc was 100 keV and 190 keV images, respectively. However, these contrasts in 100 kV/Sn140 kV mixed images were higher than those in Mono+ images and improved as the strength of advanced modeled iterative reconstruction (ADMIRE, Siemens) was increased (3<4<5). The imaging findings of the dural canal and nerve roots in non-contrast CT myelography (NC-CTM) generated from mixed images with ADMIRE5 were consistent with the intraoperative findings. In NC-CTM, favorable image quality was obtained by using dedicated metal artifact reduction software (iMAR, Siemens). As a result of training and cross-validation using 3D U-Net, the NC-CTM generated from the trained 3D U-Net model showed high similarity with the training images.
|
| Free Research Field |
放射線医学
|
| Academic Significance and Societal Importance of the Research Achievements |
本研究では非造影DECTと深層学習による腰椎硬膜管/神経根画像(NC-CTM)作成の可能性を示した。腰椎低侵襲除圧術の術前画像作成にはMRMとCTの両者を用いることが多いが，MRI検査の長い撮像時間や体内金属・閉所恐怖症等による撮影不可，MRMとCTとの画像融合時の位置ずれなどの課題がある．造影剤を用いずに通常の腰椎CTとほぼ同等の被曝線量で，かつ一回の短時間DECT撮影でNC-CTMを作成できれば，患者に与えるリスクや負担も減らすことができる。また深層学習の応用によりNC-CTMを自動生成する技術が発展すれば，作成者による完成画像の精度の違い，即ち再現性の問題の解決にも繋がる可能性がある。
|
