You don't need to use DE information for that, instead you could just use expression values (either RPKM, read counts or similar) to compare isoform proportions in your different conditions.
If you really need to use DE information, then you could try to find isoforms that are significantly more/less differentially expressed than their parent gene. For instance use this score :
abs(foldChange(Isoform) - foldChange(parentGene))
You may need to filter out lowly expressed isoforms for more robust results.