On the detectability by novices, radiologists, and computer algorithms of smallest increases in local single dot size in random-dot images

TL;DR

This study demonstrates that a self-organizing map (SOM) based metric can reliably detect small local changes in image series, surpassing human visual detection capabilities, with implications for medical and environmental monitoring.

Contribution

The paper introduces a SOM-based metric for detecting subtle local changes in image time series, improving sensitivity over visual inspection by humans.

Findings

SOM metric signals critical local changes undetectable visually.

The method outperforms human detection in small change identification.

Applicable to medical and environmental image analysis.

Abstract

Time-series of images may reveal important information about changes in medical or environmental conditions, depending on context. Visual inspection of images by humans (experts or laymen) may fail in detecting very small differences between images, yet, small but visually undetectable differences may carry important significance. Computer algorithms may help overcome this problem, and the use of computer driven image analysis in medical practice or for the tracking of small but critical changes in natural environments attracts a lot of interest. In many contexts relevant to society, the preprocessing of large sets of image series will soon no longer be the exclusive realm of a few scientists. Here we show that a metric obtained from self-organizing map analysis (SOM) of image contents in time series of images of one and the same object or environment reliably signals potentially critical local changes in images that may not be detectable visually by a layman or even an expert.