Median Baseline Filtering; Questions

[Rob Burgess]

Hi All,

I've just recieved an article concerning this. From a first glance it appears that it can smooth out baseline rises and dips resulting from gradients and make them appear flat. I'd imagine it may help with integration and making the chromatogram appear "prettier". What other advantages and/or disadvantages can anyone think of for using such "smoothing" techniques?

Further, is anyone out there using such baseline filtering software in a regulated environement? Is it allowed by the regulated authorities even?

[AA]

"Further, is anyone out there using such baseline filtering software in a regulated environement? Is it allowed by the regulated authorities even?"

Interesting question, but a simple answer. Of course it is allowed.

Think of it this way, all data coming out of any detector is filtered, it is usually filtered using something called a time constant or filter constant or something like that. This filter (used by everybody) is designed to filter out high frequency noise. A median baseline filter (not widly available, but offered by some vendors) is used to filter out low frequency noise.

From a regulated environment perspective, there is no difference, a filter is a filter. This makes the assumption that your method was validated using median baseline filter of course.

A word of warning however if you choose to use MBF, it will filter out (ie remove) broad wide peaks. As part of your validation I would suggest running a set of samples with and without it to confirm that no information is lost.

[Dan]

If I understand Rob's comment correctly, the median baseline filter (MBF) is a function (perhaps a mathematical one) used to smooth out the raw data for better integration analysis.

The time constant is something (usually an electronic circuit) that is used in the generation of the raw data. So, the time constant and the MBF are not the same.

I have not heard of someting like an MBF being used in the regulated industry. It is possible, but, I would sat that the FDA would expect that an SOP must exist for its proper use.

If the MBF is just another type of time constant device/function, then OK. If it is something to process the raw data, then be careful in its use.

Regards,
Dan

[AA]

Just to clarify for you Dan, time constants are mathimatical formulas appplied to raw data usually (always?) done on board the detector (ie within the firmware of the detector). In the only instance of a comerically avaiable MBF equipped detector that I know of, this MBF filtering also happens on board the detector. Also, just for interest, most CDS packages (Empower, Chemstation ect) also allow post data collection smoothing of data. Again, if you validate your method using smoothing as part of your integration method, it is valid and quite acceptable to any regulatory body.

[Dan]

AA,

Not all time constants are mathematical formulas.

Most detectors still make use of an electronic circuit to filter the analog signal. This circuit simply consists of a resistor and a capacitor. Resistance X capacitance = time. Hence, the term 'time constant'. A variable resistor is used to get the different filter times (0.5 s 1 s, 2 s, etc.). Detectors such as a PDA that use processing techniques to generate the detector signal can use a mathematical filter in stead of the traditional time constant (Example: the Waters PDAs use a Hamming filter).

I am aware of the data processing that is performed by the algorithms used by a CDS. However, that is usually only for peak integration purposes. In the regulated labs that I have been in, they didn't allow for any "smoothing" of the data, at least not in the forms for which I understand that term to be used.

However, I am still not clear on how the MBF functions. Is this of the same type of data processing used by a CDS for data integration?

Rob, can you provide a reference for that article? Thanks.

Regards,
Dan

[AA]

I agree, if you are using the analog out, you are not dealing with a math function except for the fact that as soon as the data hits the A to D converter, welcome back math filtering. I disagree that most detectors are are using anything but algorithmic filtering, older detectors, perhaps, but todays generation of detectors not so (of course, each vendor does it differently).

I have delevoped, validated and use several electochemical methods that use post data collection filtering. I have never had a problem with any regulatory agency.

MBF (which I have used) is a filter for low frequency noise, I have the reference somewhere, I'll try and find it on Monday.

[AA]

Median Filtering for Removal of Low-Frequency Background Drift
A.W. Moore and J.W.Jogenson
Anal. Chem 1993, 65, 188-191

[Rob Burgess]

Hi Dan / AA,

Thanks for your informative posting's on this subject... they are very interesting and much appreciated.

That was the original ref. that I subsequently obtained. The other was a white paper from Waters entitled: "Improving Chromatographic Visualization: the Median Baseline Filter Explained". I'm not sure if this papaer is in the public domain yet, but it may be worth approaching your local Waters rep. about this.

Uwe, if you are out thre, is this paper in the public domain and if so can you provide a link perchance?

[Dan]

AA and Rob,

Thanks for the references. I will need to track down the Anal.Chem. article. Meanwhile, I found the link to the Waters article:

http://www.waters.com/WatersDivision/Si ... stration=1

You will need to be registered (for free) at the Waters site to access the paper.

I'll take a look at the references and see what this is all about.

Thanks,
Regards,
Dan

[Dan]

Rob,

Sorry for a delayed response. I finally got to read the Waters white paper on MBF. This paper does a great job of explaining the MBF functionality.

In summary, the Median Baseline Filter (MBF) is an intriguing and usefull tool. I am not sure of its acceptance in the regulated lab as there is a drawback.

First, you start with the signal from the detector which was obtained using the sampling rate and filter time constant of the detector. The MBF is applied, providing an additional filter functionality. The three steps of the MBF function are: (1) a median filter determines the baseline from the selected chromatogram, (2) a smoothing filter is applied to the median filter baseline to provide the "model baseline" and (3) the model baseline is subtracted from the selected chromatogram.

It is this third step that worries me for the regulated lab. A modern CDS will allow for baseline subtraction for chromatograms (it is a bit generic and not as useful as the MBF). In the regulated labs that I have worked, that subtraction function was either not allowed for use and maybe even prohibited from use (either via a software switch to disable the function or via an SOP to disallow its use).

The conlusion of this paper is that the MBF "is a useful tool for enhancing peak visualization and baseline reproducibility."

Do we need that? The chromatograms that have had the MBF applied do look better. That may help us chromatographers. However, the example chromatogram could easily be integrated by a modern CDS. Do we need the cosmetic enhancement? As to the reproducibility, there is only one example given in the paper, more data would be needed to evaluate if the improvement in the baseline repsoducibility is actually of use in general practice. I suspect that it may, but I would like to see more examples.

So, an overall summary:
1) The MBF can be very useful to the chromatographer.
2) The regulatory authorities may need to be convinced about use of the subtraction function.
3) The use of the MBF delays the final chromatogram by 25% (see the paper). So there goes the observing of the "real time" chromatogram.

AA,

I now see what you meant by the smoothing algorithm being a filter. Thanks for the clarification. The way I learned it, the smoothing algorithm was not considered a filter, but now I see what you mean, it is a filter.

Rob and AA,

Thanks for the interesting topic and the useful info. We chromatographers can use all the tools that are made available; anything to make our jobs easier. Now we just need to convince those in the FDA, EMEA, etc.

Regards,
Dan

[Mark Tracy]

The algorithm given in the Analyt. Chem. article is easy to implement. Even my modest programming skill is enough to write one. I'll try it and see what is does. I actually have a use for this filter.

My concern about this technique is that it may be hard to transfer from say a Waters LC system to say a Dionex LC. Not only do the MBF parameters need to match between systems, even the data rates must match. Suppose you want to transfer your method to a contract laboratory that uses a different CDS than yours? Can you imagine USP's response?