2023 Fiscal Year Research-status Report
MRIを用いた深層学習による胸部領域のPET吸収補正法の開発
| Project/Area Number |
20K16758
|
|---|
| Research Institution | Kobe University |
|---|
Principal Investigator |
曽 菲比 神戸大学, 医学部附属病院, 医員 (50837680)
|
|---|
| Project Period (FY) |
2020-04-01 – 2025-03-31
|
| Keywords | PET/MRI / 減弱補正 |
|---|
| Outline of Annual Research Achievements |
In our previous study, pseudo-CT (pCT) with bone components from 360 ZTE-MRI training data set using the 2-dimensional (2D) deep learning (DL) method including unsupervised generative adversarial networks (GAN) yielded μmaps with bone components (μmapbone-2D) in combination with conventional bone-lacking μmaps (μmapno-bone) in the chest. We aimed to assess the feasibility of pCT and μmap (μmapbone-2.5D) from ZTE using 2.5 dimensional (2.5D) DL method and compare the quantitative values of the normal tissues and lesions between μmapno-bone-, μmapbone-2D-, and μmapbone-2.5D-besed reconstruction.
|
| Current Status of Research Progress |
Current Status of Research Progress
2: Research has progressed on the whole more than it was originally planned.
Reason
μmapbone-2.5D was significantly larger than μmapbone-2D in the normal bone and the liver (p=0.0009, p< 0.0001). The SUVmax by μmapbone-2.5D was significantly larger than μmapbone-2D in all regions (p= 0.0176). The correlation coefficients of the histogram for μmapbone-2.5D and μmapbone-2D were significantly higher than μmapno-bone in the bone (p<0.0001). The image-quality scores of pCT bone-2.5D for bone delineation and bone continuity were significantly higher than pCTbone-2D in both readers (p= 0.0106, p< 0.0001). μmapbone-2.5D are assumed to yield smaller differences in SUV with μmapCT than μmapno-bone and μmapbone-2D in normal tissues and lesions. The result supports the feasibility of using the 2.5D method for the generation of bone components from ZTE with DL.
|
| Strategy for Future Research Activity |
Attenuation correction (AC) of bone components in the chest on positron emission tomography-magnetic resonance images (PET/MRI) is challenging due to the complicated anatomical structures and limited delineation of the bone on MRI. A deep learning approach using unsupervised generative adversarial networks (GANs) with adaptive layer-instance normalization for image-to-image translation (U-GAT-IT) in combination with a modality-independent neighborhood descriptor (MIND) performed in our previous study yielded a pseudo-CT (pCT) generation with bone components from Zero echo-time (ZTE) MRI for AC in the chest. The purpose of this study was the external validation of the deep learning approach.
|
| Causes of Carryover |
研究プロジェクトが進行中で、実験材料や装置、人件費などの必要な経費が来年度も発生することが予想されたため。
|
Research Products
(1 results)
