I would definitively go for Salmon (or Kallisto) for three reasons:

1) Isoform/transcript based quantification can handle multi-mapping reads - which otherwise gives a bias in genes with less unique sequences - see this post.

2) Isoform/transcript based quantification tools such as Salmon and Kallisto can perform quite advanced bias corrections meaning you will get rid of some confounding factors (see their respective papers).

3) Analysis of transcript/isoforms (such as isoform switches etc) is a growing field and seems to be quite important in some sitiuations - so why not also enable youself to do that. An option for this could be my tool IsoformSwitchAnalyzeR

P.s. remember to use the tximport R package to import the Kallisto/Salmon with the offset option to pass on the bias corrections to your de analysis.

/Kristoffer