How the heck did I miss that? How visual attention limits visual perception Jeremy Wolfe Brigham & Women s Hospital/ Harvard Medical School

Why the heck did I miss that? Profusion Spatial uncertainty Inattentional blindness Prevalence Ambiguity and Decision Criteria

Profusion: Too much data Andriole, K. P., Wolfe, J. M., Khorasani, R., Treves, S. T., Getty, D. J., Jacobson, F. L., et al. (2011). Optimizing Analysis, Visualization, and Navigation of Large Image Data Sets: One 5000-Section CT Scan Can Ruin Your Whole Day. Radiology, 259(2), 346-362.

Spatial Uncertainty Optical Illusions http://www.moillusions.com/hidden-coffeefaces-and-bugs/

Inattentional Blindness: Looking for nodules in lung CT Round white nodules

There was something odd about this case

Not subtle but missed by 20 of 24 radiologists Note that radiologists were guiding to small white nodules

REALLY IMPORTANT NOTE: This is not a criticism of radiologists! Just evidence that experts use the human search engine, too. Drew, T., Vo, M. L.-H., & Wolfe, J. M. (2013). The Invisible Gorilla Strikes Again: Sustained Inattentional Blindness in Expert Observers. Psychological Science, 24(9), 1848 1853.

The problem is the attentional bottleneck

Prevalence The Problem of Rare Targets Can you find a threat?

Can you find a threat?

Does prevalence matter? Let s take 20 bags with guns and knives And put them in a stack of 40 bags 50% Prevalence Or 1000 bags 2% Prevalence

When targets are present in half of the bags people miss about 20% of them NOTE: THESE ARE VOLUNTEERS, NOT AIRPORT SECURITY OFFICERS

When targets are present in 2% of the bags people miss over 40% of them! Same threats, just rarer

False alarm errors go the other way. Which kind of error do you want to minimize?

That just the lab. Would this ever happen in a real clinical setting?

The Radiology Result Evans, K. K., Birdwell, R. L., & Wolfe, J. M. (2013). If You Don t Find It Often, You Often Don t Find It: Why Some Cancers Are Missed in Breast Cancer Screening.. PLoS ONE 8(5): e64366

Oh and these are real transportation security officers (TSOs) J. of Vision (any moment now)

Ambiguity and decision Knife?

Ambiguity and decision No

Ambiguity and decision Yes

A little model of the problem NOT a threat Maximal THREAT

There are various sorts of threat Typical threat Less Very Rare Maybe threat? Not very threatening Maximal threat Maximally threatening

Then there are non-threats Typical threat Less Very Rare Maybe threat? Maximal threat Not a Threat Not very threatening Maximally threatening

If you need to decide about Threat or not threat, that is pretty easy. NOT THREAT THREAT Less Very Rare

Call this green line your criterion NOT THREAT THREAT Less Very Rare

The problem is that the distributions overlap Less NOT THREAT? THREAT? Very Rare

If you have to make a decision, all you can do is put a criterion, somewhere Less NOT THREAT? THREAT? Very Rare

Now you have four possibilities It was a threat and you called it a threat Less Very Rare

Now you have four possibilities It was not a threat and you did not call it a threat Less Very Rare

Less Now you have four possibilities ERROR: It was not a threat and you called it a threat Very Rare

Now you have four possibilities Less ERROR: It was a threat and you missed it Very Rare Not bear-like Maximally bear-like

If MISS errors are bad Less Very Rare Move the criterion

But that is a high false alarm rate Less Very Rare Move the criterion

If you move the other direction, to keep the line moving Don t stop Stop Less Very Rare Missed threats Your choice

In summary Our limited-capacity visual system has too much to look at in unknown locations while hunting for rare and ambiguous targets

The problem is ubiquitous The solutions are not