| Autor: |
Hays, Savannah P., Lianrui Zuo, Dewey, Blake E., Remedios, Samuel W., Cassard, Sandra D., Fishman, Ann, Jiachen Zhuo, Carass, Aaron, Mowry, Ellen M., Newsome, Scott D., Prince, Jerry L. |
| Předmět: |
|
| Zdroj: |
International Journal of MS Care; 2024 Supplement, Vol. 26, p56-57, 2p |
| Abstrakt: |
BACKGROUND: MRI is used to diagnose and monitor people with multiple sclerosis (PwMS). Standard T1-weighted (T1w) images, which emphasize the contrast between gray and white matter, often fail to distinctly reveal key regions of interest such as the thalamus. Thalamic lesions are present in a significant proportion of PwMS and correlate with disability. Hence, a more effective approach to assessing this region involves acquiring 2 T1w images with different inversion times (TIs), allowing the calculation of proton density and T1 parameter maps. These maps can then generate various T1w images at any TI, thus enhancing visibility of specific regions for segmentation. However, acquiring T1w images with varied TIs is uncommon in MS clinical practice. OBJECTIVES: To demonstrate the ability to synthesize multi-TIs images using a T1w, T2w, and T2w fluid-attenuated inversion recovery (T2w-FLAIR) MRI. METHODS: We used 3-dimensional high-resolution T1w, T2w, and T2w-FLAIR brain MR images (n = 18) acquired from the University of Maryland (UMD) for training our U-Net model. The T1w images were acquired with similar acquisition parameters and 2 different TIs, 400 milliseconds and 1400 milliseconds. This allowed for direct calculation of the parameter maps. After training, we applied our model to images acquired according to the Consortium of Multiple Sclerosis Centers guidelines in the TREAT-MS trial (n = 20). We used 3 images: T1w, T2w, and T2w-FLAIR. We used HACA3 to harmonize the TREAT-MS trial images with the UMD training images. We used the synthetic parameter maps to calculate multi-TI images ranging from 400 milliseconds to 1400 milliseconds. The T1w image has a 1400-millisecond TI that we can directly compare with the corresponding synthetic multi-TI image. Since the other multi-TI images do not have ground truths, we cannot compute the same imaging metrics. RESULTS: For the synthetic T1w image, the peak-signal-to-noise ratio and structural similarity index measure are 22.82 ± 1.23 and 0.9737 ± 0.0031, respectively. These calculations were computed using only the brain voxels (excluding the background and skull). CONCLUSIONS: We demonstrated that multi-TI images can be accurately synthesized using 3 images that are commonly acquired following the Consortium of Multiple Sclerosis Centers guidelines. When interested in the thalamus, we can use a 400-millisecond TI to robustly identify individual thalamic nuclei. With the high presence of thalamic lesions in PwMS, the synthesis of these images could lead to better lesion segmentation and stronger associations with disability. [ABSTRACT FROM AUTHOR] |
| Databáze: |
Complementary Index |
| Externí odkaz: |
|
