Both filter are actually the same in a different stage of the design.
The one in the bottom is the result of transforming the impedance inverters (2-port ABCD) into a pi network. The series element is a stub and the shunt elements are absorbed by the resonators at both sides. That´s why they resonator stubs are different in both circuits.
Anyway, the filter is well designed, I replicated everything in ADS and the results are what I expected, but not in QucsStudio. So, my conclusion is that I´m using the TL wrongly or the ideal model isn´t correct when it´s in series.
If you are interested in the original LC prototype, you can run the attached schematic.