Many Can Work Better than the Best: Diagnosing with Medical Images via Crowdsourcing

Abstract

We study a crowdsourcing-based diagnosis algorithm, which is against the fact that currently we do not lack medical staff, but high level experts. Our approach is to make use of the general practitioners’ efforts: For every patient whose illness cannot be judged definitely, we arrange for them to be diagnosed multiple times by different doctors, and we collect the all diagnosis results to derive the final judgement. Our inference model is based on the statistical consistency of the diagnosis data. To evaluate the proposed model, we conduct experiments on both the synthetic and real data; the results show that it outperforms the benchmarks.

Keywords

• medical images based diagnosis
• crowdsourcing
• entropy
• Kullback–Leibler divergence