[w1ntermute]

The fatal flaw in your reasoning lies in your assumption that the device will be used to screen for melanoma in the entire population. In reality, it would be used only on patients who show other signs that would normally result in a referral to a specialist (i.e., the chance they have melanoma is much higher than 20/100000). The likelihood of an accurate diagnosis is 85% when this population is examined, and not the general population. As the article states, the likelihood of an accurate diagnosis (85%) is the same regardless of whether it's performed in the traditional manner by a specialist, or if it's performed using this device by a physician who would have otherwise referred the patient to a specialist.

TL;DR: 73.6% of all statistics are made up.