Isolating analog signals while maintaining accuracy is hard. Optocouplers tend to be rather nonlinear, and can have wide unit-to-unit and temperature variations in their CTR. They can be okay for something like an SMPS where the signal is maintained very close to a threshold, but they are not a great choice for isolating a signal with a large dynamic range. There are dual-output optocouplers designed to help with those problems by using a second phototransistor to provide feedback to linearize the output, but they only help so much.
Anyway, if your MCU shares a ground connection with the power stage, then isolating the Vds sense connection alone isn't very beneficial. A simple divider circuit, with adequate protection, will probably be fine.
One other note: since the shunt resistor is between source and ground, you need to subtract that voltage from the drain voltage to get Vds. The ADCs built into most MCUs cannot sample channels simultaneously, so if you digitize both Vd and Vs and then do the subtraction in software, you will get an error from the difference in sampling time. You could solve that by using an MCU with multiple ADCs that can be synchronized, an external simultaneous-sampling ADC (or multiple synchronized external ADCs), or a differential amplifier that does the subtraction in the analog domain. On the other hand, with such a low shunt value and a relatively large range on Vd, the error is probably negligible and you can treat Vd as close enough to Vds.