MRI
正在载入扫描仪与解剖数据Loading the scanner and anatomy
MRI自旋到图像From spin to image
自旋Spins

拨一下,看它们散开又聚拢Tip them, watch them fan out and regroup

头部切面上的每个箭头,是一小团处在略有不同磁场里的自旋。按下面的按钮激发它们:先一起转,再慢慢散开;自旋回波能让它们重新聚拢。

Each arrow on the slice is a small group of spins in a slightly different local field. Excite them with the buttons below: they turn together, then drift apart; a spin echo brings them back.

原理与公式Physics and formulas

箭头不是"单个质子像指南针一样排列",而是等色团:一小团处于相同局部磁场中的自旋的平均磁化。在主磁场 B₀ 中它们以 Larmor 频率 进动;射频脉冲把净磁化扳离 z 轴,随后它按 Bloch 方程弛豫。这里用 2048 个等色团精确求解(硬脉冲旋转 + 指数弛豫算子),失谐 Δω 服从洛伦兹分布,所以整体 FID 按 T2* 衰减。自旋回波见 Hahn 1950。

The arrows are not single protons lining up like compass needles but isochromats: the mean magnetisation of a small group of spins sharing the same local field. In B₀ they precess at the Larmor frequency ; an RF pulse tips the net magnetisation away from z, after which it relaxes according to the Bloch equation. Here 2048 isochromats are solved exactly (hard-pulse rotations plus exponential relaxation operators) with Lorentzian-distributed off-resonance, so the ensemble FID decays with T2*. For the spin echo see Hahn 1950.

对比度Contrast

同一个大脑,换个时间就是另一幅图Same brain, different timing, different picture

MRI 没有固定的"照片"。等多久再激发(TR)、等多久再读信号(TE),决定哪种组织亮。拖动滑块,扫描仪里的切面会跟着变。

MRI has no single “photograph”. How long you wait before exciting again (TR) and before reading the signal (TE) decides which tissue is bright. Drag the sliders and the slice inside the scanner follows.

原理与公式Physics and formulas

图像对比来自同一组织参数(PD、T1、T2、T2*)在不同时间参数下的组合。TR 控制纵向恢复多少(T1 权重),TE 控制横向衰减多少(T2 / T2* 权重)。每个体素按教科书信号方程合成,组织参数来自人体在体测量。注意 CNR、SNR 和扫描时间:真实的取舍从来不只是对比好不好。

Contrast comes from the same tissue parameters (PD, T1, T2, T2*) combined under different timings. TR controls how much longitudinal recovery there is (T1 weighting), TE how much transverse decay (T2 / T2* weighting). Every voxel is synthesised from the textbook signal equations with in-vivo tissue values. Watch CNR, SNR and scan time: the real trade-off is never contrast alone.

k 空间k-space

图像是用条纹拼出来的The image is assembled from stripes

扫描仪采到的不是像素,而是各种疏密和方向的条纹各占多少。少采一些、截掉一些、中途动一下,看图像各自坏成什么样。

The scanner does not record pixels; it records how much of each stripe pattern the image contains. Skip some, cut some off, or move halfway through, and see how the image breaks in each case.

原理与公式Physics and formulas

梯度场让频率随位置变化,于是接收到的信号就是物体的傅里叶变换。这里的"物体"是 0.5 mm 的高分辨率解剖;扫描只取它频谱中央的 N×N 块,所以 Gibbs 振铃、卷褶、运动鬼影都是对 k 空间做了相应操作后自然出现的。

Gradients make frequency depend on position, so the received signal is the Fourier transform of the object. The “object” here is anatomy at 0.5 mm; a scan keeps only the central N×N block of its spectrum, so Gibbs ringing, wrap-around and motion ghosts appear by themselves once k-space is treated accordingly.

扩散Diffusion

看水分子往哪儿走Watching where water moves

水分子一直在乱走。在神经纤维里它顺着纤维走得更远,所以能从信号里读出纤维的方向。切面上的颜色就是方向:红 左右,绿 前后,蓝 上下。

Water molecules wander all the time. Inside nerve fibres they travel further along the fibre, so fibre direction can be read from the signal. The colours on the slice are directions: red left–right, green front–back, blue up–down.

原理与公式Physics and formulas

一对扩散梯度(Stejskal–Tanner)让移动的水分子积累无法重聚的相位,信号按 衰减。b 值由梯度幅度 G、时长 δ、间隔 Δ 决定;白质的张量各向异性让信号依赖于梯度方向。六个方向加 b=0 即可解出张量、FA 与主方向。纤维方向由白质形态估计,属示意。

A pair of diffusion gradients (Stejskal–Tanner) leaves moving water with phase that cannot be refocused, so the signal decays as . The b-value follows from gradient amplitude G, duration δ and spacing Δ; tensor anisotropy in white matter makes the signal depend on gradient direction. Six directions plus b=0 determine the tensor, FA and principal direction. Fibre directions are estimated from white-matter shape and are illustrative.

fMRI

看大脑哪里在工作Seeing which part of the brain is at work

脑区一活跃,血流跟着增加,那里的信号会升高百分之几。切面后部亮起的是视觉皮层。变化这么小,要靠实验设计和统计才能从噪声里找出来。

When a brain region becomes active, blood flow follows and the signal there rises by a few per cent. The patch lighting up at the back of the slice is visual cortex. A change this small has to be pulled out of the noise by experimental design and statistics.

原理与公式Physics and formulas

神经活动 → 局部血流增加超过耗氧 → 脱氧血红蛋白减少 → R2* 下降 → T2* 加权信号上升几个百分点。这里:刺激序列与 SPM 典范 HRF卷积得到 ΔR2*(t),再加入扫描仪漂移、生理噪声(∝ 信号)与热噪声,最后用含余弦漂移项的一般线性模型逐体素求 t 值。

Neural activity → blood flow rises more than oxygen use → less deoxyhaemoglobin → R2* falls → the T2*-weighted signal rises by a few per cent. Here the stimulus is convolved with the SPM canonical HRF to give ΔR2*(t); scanner drift, physiological noise (∝ signal) and thermal noise are added, and a general linear model with cosine drift terms gives a t value per voxel.