Department of Medical Physics, Ninewells Hospital and Medical School, Dundee, Angus DD1 9SY, UK.
Med Phys. 2011 Sep;38(9):5058-66. doi: 10.1118/1.3622605.
Texture analysis (TA) has proved to be useful to distinguish different tissues and disease states using magnetic resonance imaging (MRI). TA has been successfully applied clinically to improve identification of abnormalities in the brain, liver, and bone and, more recently, has been used to enhance the specificity of breast MRI. This preclinical study used a custom-made phantom containing different grades of reticulated foam embedded in agarose gel to assess the capability of TA to distinguish between different texture objects, under different imaging conditions. The aim was to assess whether TA could be used reliably with clinical protocols that were not optimized for texture analysis and also to investigate the effect that changing imaging sequence parameters would have on the outcome of TA.
Clinical fast gradient echo sequences and two different breast RF coils were used in order to reflect standard clinical practice. Three protocols were used: (1) a high spatial resolution protocol run on a 1.5 Tesla (T) MRI scanner, (2) a parameter matched sequence run on a 3.0 T magnet, and (3) a high temporal resolution protocol also run on a 3.0 T magnet.For each protocol, three sequence parameters (repetition time, bandwidth/echo time, and flip angle) were altered from the baseline values to assess the impact of changes in acquisition parameters on the outcome of TA.
TA was performed using MAZDA software and clearly differentiated four foam phantoms when using the wavelet transform method (WAV), also moderately so with the co-occurrence matrix method (COM). The outcome was generally improved for imaging protocols acquired on the 3.0 T scanner, particularly for the high spatial resolution protocol where changes to the acquisition parameters influenced the TA, especially changes to the bandwidth/echo time. For the other protocols, TA outcome was less affected by changes to the imaging parameters.
This phantom study shows that acquisition parameters and protocols that are typically used for clinical breast imaging can result in good TA. Our findings suggest that changes to sequence parameters may not greatly influence the outcome of texture analysis, but rather that spatial resolution may be the most important factor to consider.
纹理分析（TA）已被证明可用于通过磁共振成像（MRI）来区分不同的组织和疾病状态。TA 已成功应用于临床，以提高对大脑、肝脏和骨骼异常的识别能力，最近，它已被用于增强乳腺 MRI 的特异性。本临床前研究使用了一个定制的包含不同等级的泡沫网格嵌入琼脂糖凝胶的幻影来评估 TA 在不同成像条件下区分不同纹理物体的能力。目的是评估 TA 是否可以可靠地用于未针对纹理分析进行优化的临床方案，以及研究改变成像序列参数对 TA 结果的影响。
使用临床快速梯度回波序列和两种不同的乳腺射频线圈来反映标准的临床实践。使用了三种方案：（1）在 1.5T 磁共振扫描仪上运行的高空间分辨率协议；（2）在 3.0T 磁体上运行的参数匹配序列；（3）也在 3.0T 磁体上运行的高时间分辨率协议。对于每个协议，从基线值改变三个序列参数（重复时间、带宽/回波时间和翻转角），以评估采集参数变化对 TA 结果的影响。
使用 MAZDA 软件进行 TA，当使用小波变换方法（WAV）时，清晰地区分了四个泡沫幻影，当使用共生矩阵方法（COM）时也能适度区分。对于在 3.0T 扫描仪上采集的成像协议，TA 的结果通常会得到改善，特别是对于高空间分辨率协议，采集参数的变化会影响 TA，特别是带宽/回波时间的变化。对于其他协议，成像参数的变化对 TA 结果的影响较小。
这项幻影研究表明，通常用于临床乳腺成像的采集参数和协议可以产生良好的 TA。我们的发现表明，序列参数的变化可能不会对纹理分析的结果产生重大影响，而是空间分辨率可能是需要考虑的最重要因素。
