||5R01CA236791-02 Interpret this number
||University Of California Los Angeles
||Perceptual and Adaptive Learning in Cancer Image Interpretation
Cancer screening from visual displays, as in dermatology and radiology, depends crucially on the expertise of
medical practitioners, but current data indicate that even among experienced professionals there are significant
and persistent error rates. While there have been impressive advances in the technologies of medical imaging,
considerably less attention has been paid to the learning processes involved in the training of medical image
interpretation. Research in perception and cognition indicates that the central process by which people become
able to detect and classify complex and subtle patterns and structures in visual images is a process known as
perceptual learning. Through perceptual learning mechanisms, with appropriate practice in a given domain, the
brain progressively improves information extraction to optimize task performance. These mechanisms are
largely unaffected by the traditional didactic instruction common in medical education; instead, they depend on
interaction with large numbers of examples with task-relevant feedback. Recent work has shown that
application of principles of perceptual learning can dramatically accelerate accuracy and fluency in medical
learning domains. Evidence suggests that these training methods can be markedly enhanced, and customized
for individual learners, by incorporating novel adaptive learning algorithms based on principles of learning and
The primary aim of this project is to investigate principles and mechanisms of perceptual and adaptive learning
in the learning of multiple diagnostic categories in dermatologic screening and mammography, with the
ultimate aim of improving training and proficiency in cancer image interpretation. Studies with novices in lab
settings will establish basic principles and hypotheses, and selective studies with nurse melanographers,
residents, and physicians will test validation with actual practitioners. Culminating studies of melanographers in
actual dermatologic screening settings will compare practitioners who train with best-practices perceptual-
adaptive learning modules (PALMs) to control participants. Specific studies will investigate the incorporation of
signal detection concepts into adaptive perceptual learning systems; the role of comparisons in defining and
differentiating perceptual categories; the relative benefits of passive and active learning episodes across
learning phases; and the relationship between the stringency of mastery criteria and the degree to which
resulting performance is accurate, fluent, generalizable, and long-lasting.
Local features and global shape information in object classification by deep convolutional neural networks.
, Lu H.
, Erlikhman G.
, Kellman P.J.
Vision research, 2020 07; 172, p. 46-61.