Yes, in theory you should be able to use transcript counts to do differential expression of isoforms. But be aware that differential isoform expression is not the same thing as differential splicing.

For example, imagine a gene with two isoforms expressed for the same promoter. If transcription is up regulated for this gene, then (in the ideal world) both isoforms will be DE, even though there has been no change in splicing. If there is a change in splicing, you would expect one isoform to go up and one go down, and in the absence of acompanying transcriptional differences for the some of the number of isoform 1 transcripts and isoform 2 transcripts to remain constant (remember counts =/= # of transcripts, so the sum of counts probably won't stay the same). Of course in either of these situations one or both of the isoforms may fail to make the threshold for significance, wither because there is too much variation, or the number of reads was too low.

There are many packages specifically designed to find alternate splicing as opposed to isoform specific DE. As mentioned by @arta, DEXseq from the DESeq2 authors is one of them.

You can use DEXSeq from same guy (Simon Anders) to compare differential exon usage.

DEXseq seems to be a very good option to estimate differential usage of exons but I don't have biological replicates, can I still use DEXseq?