Abstract
Perceptual learning (PL) is not only a strong tool for understanding visual plasticity (Sasaki, Nanez and Watanabe, 2011; Watanabe & Sasaki, 2015) but also for reducing symptoms of eye diseases. Here, we show that PL can also be a strong tool to increase the detectability of breast cancers in mammography scans only if an appropriate response feedback during training is given for different types of breast cancer. Response feedback is known to facilitate PL (Herzog and Fahle, 1998; Shibata et al., 2011), although a detailed role of feedback remains unclear. We examined the role of response feedback for two types of breast cancer: calcification cancer and architectural distortion cancer. The former is easier to detect than the latter. Thus, we examined how different contents of feedback are effective on PL for different breast cancers. Three 30-min long training sessions were conducted on separate days. Detection performance was measured in pre-training and post-training tests. Three different groups of participants (N = 48 subjects in total) without medical training background were trained with different contents of feedback. In the detailed feedback condition, participants were given feedback regarding the location of cancer as well as the correctness of detection. In the partial feedback condition, participants were provided with feedback only about the correctness of detection. In the no feedback condition, participants received no feedback. In the detailed condition detectability improved for architectural distortion and calcification cancers, whereas in the partial feedback condition detectability improved only for calcification. No learning for either cancer was observed in the no feedback condition. These results suggest that PL significantly improves the detectability of breast cancers in mammograms. More importantly, the content of feedback leads to different learning outcomes for different breast cancers, indicating the importance of future studies on the effect of content of feedback on PL.
Acknowledgement: NIH R01EY019466, NIH R21EY028329, NIH R01EY027841, BSF2016058