How to do semi quantitative detection in lateral flow ?

An idea, there might be better ones:

If you design the lower concentration limit to be at X concentration, you can do. It won't be very precise, but in general it should work. If you have a good binding antibody for the test line, then you reduce its concentration till you just get a coloring when your concentration is high enough.

Dieter Spiehl Thank you, but I need more specific method

In many cases this cut-off value is obtained by using a series of calibration solutions. However, consider that this limit may vary according your test conditions and the sample. It is not easy to adjust the assay to value you might prefer.

The barcode/ladder-bar lateral flow assay format may suit you needs. It involves printing multiple test lines on the same LFA strip with varying concentrations of antibody so that each test line will have a different "cut-off value" which is the concentration of the target analyte required for that line to become visible by the naked-eye. While useful this method is practically limited to 3-5 different concentration ranges due to limited resolution and room for printing on the membrane so it won't work for all applications. I have used this method for the development of an LFA for semi-quantitative detection of beta-trace protein for detecting CSF leaks which you can find on my profile. It's also been used for quantification of C-reactive protein in blood and gliadins in food samples. Other than this, the best approach would be using an electronic bench top or cell phone reader to compare the assay signal to an internal standard curve.

This paper below may help you understand how to set up this barcode style LFA or even just adjust the detection limit for an assay that has one test line. Basically by varying the concentration of the antibody immobilized on the membrane, the amount of your antibody-labeled probe, and the sample dilution prior to running the LFA, you can adjust the minimum analyte concentration required for the test line to become visible. It is a highly empirical process so you will have to screen many different conditions of these 3 variables to find the best setup.

Thank you so much @ Daniel, it will certainly solve my problem, I will get back to you for further help.