1. news
  2. 2020

Craig receives NIH grant to improve radiologists' breast cancer screening performance

The work will focus on how visual adaptation may facilitate the reading of breast cancer screening image batches.
August 19, 2020
Fig 1 | Example of an Invasive Carcinoma imaged with FFDM (left) and DBT (right). Images from Rafferty et al., 2013.
Fig 1 | Example of an Invasive Carcinoma imaged with FFDM (left) and DBT (right). Images from Rafferty et al., 2013.

VIU Lab member Craig Abbey is one of four principle investigators of a newly awarded National Institutes of Health (NIH) grant, entitled Sequential Reading Effects in Digital Breast Tomosynthesis, with funding provided by the National Cancer Institute (NCI). The grant investigates visual adaptation as a mechanism for improved performance in sequential batch reading in breast cancer screening images (Figure 1).

Fig. 2 | Batch reading results from Taylor-Philips et al., 2017. The cancer detection rate (left) remains flat while the recall rate (right) decreases across the position of a case within a batch. Median number of cases per data point is 21,931.
Fig. 2 | Batch reading results from Taylor-Philips et al., 2017. The cancer detection rate (left) remains flat while the recall rate (right) decreases across the position of a case within a batch. Median number of cases per data point is 21,931.

Radiologists often read high-volume exams, such as breast-cancer screening images, in uninterrupted batches. Historically there has been concern that this practice would lower accuracy due to decreased vigilance. However, a number of studies in the US and Europe have found the opposite effect (Figure 2), with batch reading leading to higher accuracies (Burnside et al., 2005) that continue to improve as readers progress through an image batch (Taylor-Philips et al., 2017; Abbey et al., 2020). Currently, there is no accepted explanation for this improvement.

This grant will explore visual-cognitive adaptation as a mechanism for this improvement. Visual-cognitive pathways are able to recalibrate in response to new visual environments, and we hypothesize that this process leads to the observed improvements in batch reading diagnostic accuracy. If confirmed, this explanation could have important implications for how to improve and optimize radiologist performance in breast cancer screening.