Use of Mechanical Turk as a MapReduce Framework for Macular OCT Segmentation

Figure 1

Examples of incorrect segmentations by automated software and user-interface for Mechanical Turk. Panel (a) is an example of a macula SD-OCT image with missing information (arrow) causing a sudden jump in the identification of the Internal-Limiting Membrane (ILM) by automated software included with Heidelberg Spectralis. Panels (b and c) show two similar macular OCT images with different automated segmentations caused by pigment epithelial detachment (arrow) and subretinal fibrosis (arrowhead). Panel (d) is a screenshot of web-based user-interface submitted to Amazon Mechanical Turk for manual segmentations.

(a)

(b)

(c)

(d)