Supervised Learning Occurs in Visual Perceptual Learning of Complex Natural Images

Abstract

There have been long-standing debates regarding whether supervised or unsupervised learning mechanisms are involved in visual perceptual learning (VPL) [1-14]. However, these debates have been based on the effects of simple feedback only about response accuracy in detection or discrimination tasks of low-level visual features such as orientation [15-22]. Here, we examined whether the content of response feedback plays a critical role for the acquisition and long-term retention of VPL of complex natural images. We trained three groups of human subjects (n = 72 in total) to better detect "grouped microcalcifications" or "architectural distortion" lesions (referred to as calcification and distortion in the following) in mammograms either with no trial-by-trial feedback, partial trial-by-trial feedback (response correctness only), or detailed trial-by-trial feedback (response correctness and target location). Distortion lesions consist of more complex visual structures than calcification lesions [23-26]. We found that partial feedback is necessary for VPL of calcifications, whereas detailed feedback is required for VPL of distortions. Furthermore, detailed feedback during training is necessary for VPL of distortion and calcification lesions to be retained for 6 months. These results show that although supervised learning is heavily involved in VPL of complex natural images, the extent of supervision for VPL varies across different types of complex natural images. Such differential requirements for VPL to improve the detectability of lesions in mammograms are potentially informative for the professional training of radiologists.

Keywords: breast cancer; feedback; high-level vision; mammogram; natural stimuli; radiology; supervised learning; unsupervised learning; visual perceptual learning.

Figure 1

Example mammograms used for VPL. Left and right breasts are presented side-by-side. The yellow circle shows the location of the lesion and was only visible during trial-by-trial feedback. (A) Calcification, defined as fine white specks, tightly grouped together. (B) Architectural distortion, defined as lines radiating to a central point (similar to the spokes of a wheel). (C) No lesion. See also Figure S1.

Figure 2

Examples of different contents of feedback provided at the end of each trial during training. (A) In the detailed feedback condition, subjects were informed about the correctness of their response and the true location of the lesion. To this purpose, the mammogram was shown again, and the true location of the lesion was enclosed by a yellow circle. Subjects could compare the true location with the location they had identified (denoted by blue crosshair). If there was no lesion, the mammogram was presented without any yellow circle. (B) In the partial feedback condition, subjects were only informed about the correctness of their response. (C) In the no feedback condition, subjects did not receive any feedback at the end of the trial and moved on to the next trial. See also Figure S1.

Figure 3

Results of the experiment. Pre and Post on the x-axis correspond to the pretest and posttest, which were conducted before and after training, respectively. Gray bars represent the mean (± standard-error-of-the-mean, SE) observer sensitivity (d’) for the trained lesion (n = 12 subjects for each training group and content of feedback condition). Markers connected by lines show individual subject data. Each color denotes a different subject. (A) Subjects trained on calcifications with detailed feedback during training. (B) Subjects trained on distortions with detailed feedback during training. (C) and (D): Same as (A) and (B) but with partial feedback during training. (E) and (F): Same as (A) and (B) but without feedback during training. Asterisks show the results of post hoc paired sample t-tests between pre- and posttests (*** p < 0.001, ** p < 0.01, * p < 0.05, n. sig. = no significant difference). See also Figure S2 and Figure S3.

Figure 4

Retention of VPL. Pre on the x-axis refers to the pretest. The retest was conducted on available subjects six months after the posttest. Gray bars represent the mean (± SE) d’ for the trained lesion. Markers connected by lines show individual subject data. Each color denotes a different subject. (A) Subjects (n = 9) trained on calcifications with detailed feedback during training. (B) Subjects (n = 9) trained on distortions with detailed feedback during training. (C) Subjects (n = 9) trained on calcifications with partial feedback during training. Asterisks show the results of post hoc paired sample t-tests between retest and pretest (** p < 0.01, * p < 0.05, n. sig. = no significant difference). See also Figure S4.