Nodule Detection with Eye Movements

Abstract

Radiologists often miss nodules that may represent lung cancer on chest radiographs. We investigated whether eye movements collected during the search for lung nodules by large samples of laypeople may provide information that could assist radiologists in their detection. For that purpose, we developed a partially invisible Markov model with partially unobserved states and analyzed eye tracking data of over 100 laypeople who reviewed 14 chest X-ray images, of which seven contained a potentially cancerous nodule. We used the luminance value of the pixels in the X-ray images as prior information on the possible location of a nodule and identified six regions of interest on each image that may contain a nodule. Our study demonstrated that the eye movements recorded from laypersons contained information that may assist radiologists in the detection of nodules in chest X-rays, which has important implications for crowdsourcing of search for pulmonary nodules, which are discussed.