All positive feedback on the previous period, "Do you have to understand the imaging of Truth" series, small series encouraged by new lecture topic again today - noise ratio.
You do fluorescence imaging must have the experience of it, you carefully prepare a good sample, adjusting the microscope and software, click the camera button, the results were the following image.
figure 1
To see this image is a bit frustrating? However, experienced users have been aware of the problem and manually extend the exposure time or increase the intensity of the excitation light, when you see the image may be the following two.
figure 2
image 3
Wow, image quality improvement immediate ah! Yes, you look and feel of this three image quality is a problem today, Xiao Bian think chatter chatter - the image quality is the most important basis to judge - noise ratio.
Before the commencement of this question, please consider the following two questions:
1. Noise Figure 1 and Figure 3 compared to what a bigger?
2. The signal to noise ratio in the end what is it?
Professional answer to the question dividing line - Think first look at the answer Oh!
Well, we publish answers to slightly! I want to see their own right.
correct answer:
1. FIG three noise, a small FIG. My Force told me that a lot of people got it wrong!
2. The SNR is the ratio of signal and noise. Oh, the answer is nothing wrong, but just a literal interpretation, it is quite abstract. To put this concept clear, or have the formula:
It looks a lot of new concepts, but it is not difficult to understand. Below we explain one by one:
First is the signal. The signal is actually our experiments, GFP fluorescence green or red fluorescent RFP. While the number of photons of fluorescence intensity is quantified expression. A digital imaging device, the required optical signals into electrical signals can be measured, for example, a confocal microscope in the photomultiplier tube (the PMT), or the camera (CCD, CMOS camera) wide-field fluorescence imaging use. And the quantum efficiency (Quantum Efficiency, QE) is the camera or the PMT efficiency of converting the optical signal to an electrical signal. The most important properties of a detection device, is to have a higher quantum efficiency.
然后我们说噪声。噪声源实际上也有很多种，大家都知道相机自身是会产生噪声的，而对于和活细胞弱荧光成像相似的应用，相机的读出噪声（Readout Noise）是对图像质量影响很大的一个参数。除了读出噪声，相机实际上还有其它很多种噪声，例如暗噪声（Dark Noise），或是对CMOS相机成像质量影响较大的相关性噪声（Correlated Noise）。除某些特定成像应用外，一般暗噪声或是相关性噪声的贡献很小，在这篇文章中小编就不过多讨论。忽略掉其它的噪声源，上面的公式可以简化成为
这里还有一个非常重要的噪声源，称为散粒噪声（Shot Noise），由于从相机硬件层面无法降低散粒噪声，所以很多时候这个最为重要的噪声反而并没有太多人提及。（注：硬件层面没办法，并不意味着软件层面没办法哦！）
散粒噪声是光信号自身的噪声
上面的公式就进一步变化为
这个时候，我们分析信噪比随信号变化的曲线，会得到如下结果。其中实线表示无噪声的理想相机，虚线是实际的相机。可见，读出噪声越大，相机曲线与理想曲线偏离越远。
从以上公式可以看出，在信号比较强的情况下，散粒噪声远远大于读出噪声，信噪比其实等于
这时，决定信噪比的仅仅是信号强度和量子效率！本文开始时提到的，通过增加曝光时间或是加强激发光，能提高图像的原因就在这里了。不过，实际拍照时，由于各种原因，例如避免光漂白、光毒性，或是为了提高速度，活细胞成像的信号常常处在相对较低的水平，这时候我们就需要借助信噪比来判断图像质量好坏与否了。
用上面这张图为范例，信噪比的计算方法如下，
除了判断图像质量，对于很多需要做荧光定量测量的应用，例如简单的钙离子成像，FRET/FRAP，或是复杂如单分子定位的超高分辨率成像（STORM/PALM），这也是一个非常重要的概念。
最后我们回到开始的问题，为什么图3的噪声会高过图1呢？原因很简单，图3本身信号强，散粒噪声占主导地位，噪声远大于图1。图3质量好，并不是噪声小，而是信噪比高。
以上，我们讨论了信噪比的一些基本概念。现在给大家再出两个问题，请判断以下说法是否正确：
-
成像的追求，就是要不断提高信噪比；没有最好，只有更好
-
图像的信噪比，有经验后视觉判断就可以了, 定量计算很难，也没有必要
期待大家踊跃留言回答，小编将会在下一篇文章中给出解答。