Crowdsourcing for relevance evaluation
ACM SIGIR Forum
Proceedings of the international conference on Multimedia information retrieval
Overview of the second workshop on medical content---based retrieval for clinical decision support
MCBR-CDS'11 Proceedings of the Second MICCAI international conference on Medical Content-Based Retrieval for Clinical Decision Support
Crowdsourcing for affective-interaction in computer games
Proceedings of the 2nd ACM international workshop on Crowdsourcing for multimedia
| Hi-index | 0.00 |
As in many other scientific domains where computer--based tools need to be evaluated, also medical imaging often requires the expensive generation of manual ground truth. For some specific tasks medical doctors can be required to guarantee high quality and valid results, whereas other tasks such as the image modality classification described in this text can in sufficiently high quality be performed with simple domain experts. Crowdsourcing has received much attention in many domains recently as volunteers perform so--called human intelligence tasks for often small amounts of money, allowing to reduce the cost of creating manually annotated data sets and ground truth in evaluation tasks. On the other hand there has often been a discussion on the quality when using unknown experts. Controlling task quality has remained one of the main challenges in crowdsourcing approaches as potentially the persons performing the tasks may not be interested in results quality but rather their payment. On the other hand several crowdsourcing platforms such as Crowdflower that we used allow creating interfaces and sharing them with only a limited number of known persons. The text describes the interfaces developed and the quality obtained through manual annotation of several domain experts and one medical doctor. Particularly the feedback loop of semi--automatic tools is explained. The results of an initial crowdsourcing round classifying medical images into a set of image categories were manually controlled by domain experts and then used to train an automatic system that visually classified these images. The automatic classification results were then used to manually confirm or refuse the automatic classes, reducing the time for the initial tasks. Crowdsourcing platforms allow creating a large variety of interfaces for judgements. Whether used among known experts or paying for unknown persons, they allow increasing the speed of ground truth creation and limit the amount of money to be paid.