The correct Statistical tecnique for change in proportions based on paired data is called McNemar. You find it in SPSS under "Non-Parametric" tests.
You have to find a sample size calculator that can calculate sample size based on this statistical test
However, I see that you are talking about patients and an intervention. You should prefarably work out a stronger design With at least a cross-over study or comparison With a Control Group (se medcalc link below)
See also: Nussbaum: Categorical and Nonparametric data analysis (book)
There are several explanations and calculators on the web, for example: