Perceptual learning of lesions in mammograms induced by response feedback during training

Detecting subtle lesions in mammograms indicative of early breast cancer usually requires years of experience. Well-designed training paradigms could be a strong tool for promoting perceptual learning (PL) with rapid and long-lasting improvement in detectability of these subtle mammographic lesions....

Full description

Saved in:
Bibliographic Details
Published inbioRxiv
Main Authors Sebastian Martin Frank, Qi, Andrea, Ravasio, Daniela, Sasaki, Yuka, Rosen, Eric, Watanabe, Takeo
Format Paper
LanguageEnglish
Published Cold Spring Harbor Cold Spring Harbor Laboratory Press 02.09.2019
Subjects
Breast cancer
Calcification
Feedback
Learning
Mammography
Therapeutic applications
Online AccessGet full text

Cover

More Information
Summary:Detecting subtle lesions in mammograms indicative of early breast cancer usually requires years of experience. Well-designed training paradigms could be a strong tool for promoting perceptual learning (PL) with rapid and long-lasting improvement in detectability of these subtle mammographic lesions. Given that PL occurs without feedback about the accuracy of subjects' responses, the role of feedback has been completely ignored in clinical applications of PL. However, in this study, we found that the contents of the feedback profoundly and differentially influence the formation and retention of PL to detect calcification and architectural distortion lesions, two types of mammographic lesions that are frequently missed in mammographic screenings. We trained subjects to detect one type of lesion in a mammogram and manipulated the content of the response feedback during training for 3 groups (no feedback, correctness only, and both correctness and location of the lesion). We found that PL occurred for both lesions when both correctness and location feedback were provided. PL also occurred for calcifications but not for distortions when only correctness was provided. No learning occurred without feedback for either lesion. A retest conducted six months later showed that PL was retained only in the group with both correctness and location feedback for both types of lesions. In contrast to the general consensus of basic PL studies, our results demonstrate that the content of the response feedback is a determining factor in forming and retaining PL to detect mammographic lesions.
DOI:10.1101/752246