Dogs have been shown to detect certain cancers at later stages (which greatly limits use as a screener.. not to mention how does the sensitivity/specificity compare to state of the art). This is not surprising since as cancers advance they cause gross biochemical changes.
The further issue diagnosing cancer/no cancer or epilepsy/no is an easy problem. But train a dog/machine on 1000 or more rare diseases. Now selectivity is an actual issue. And unless you manage to overcome that, how is all this scent stuff clinically useful?