You could try to use an ultrasonic differential Doppler method, but I
suspect that this will be expensive, in development costs and transducers.
This method is metrology instrumentation grade, once you have a tube that
does not vary in dimensions and a constant density liquid. I know that
something like this is used to measure blood flow in artificial kidneys
and other delicate places. If you want to get really picky, also measure
the temperature and pressure in the tube to correct for compressibility
and density variation ;) The differential Doppler method requires an
ultrasonic transmitter and two receivers, all immersed in a tube section
of constant cross-section.  It can measure flow in both directions, and
has a lower speed limit under which no measurement is possible (due to the
phase resolution of the detection equipment). There are no S/N and
preamplifier problems as the transducer output signal is huge and clean.
imho this method is excellent for a PIC, used both to generate the
ultrasonic frequency directly (by bit-bashing) and for data aquisition.

  Another method that does not use moving parts, uses the thermal transfer
properties of the liquid. You have a piece of tubing that is thermally
insulated, and inside it, a heated thermistor, and at some distance
downstream, another one that is not heated. Knowing the thermal
conductvity of the liquid and the temperature read-out at the 2nd
thermistor, and tube diameter etc you can determine flow rate with uncanny
accuracy. A third unheated thermistor is used upflow of the heated one to
give a temperature reference if the liquid temperature is not constant.
This method requires serious calibration, unlike the differential Doppler
one, which needs none. However, it measures flow down to very, very near
zero, in both directions.

hope this helps,

        Peter