Editors' ChoiceCancer

Pecking order

See allHide authors and affiliations

Science Translational Medicine  09 Dec 2015:
Vol. 7, Issue 317, pp. 317ec210
DOI: 10.1126/scitranslmed.aad8028

The continuous improvement of medical image digitization technologies brings along with it a constant need to evaluate and assess these technologies. Existing tools, in the form of evaluation software, can only address a narrow set of evaluations questions, suffer from poor adaptability to new image features, and essentially fail to reflect human performance. Hence, the domain expert remains the gold standard; however, designing and conducting human observer studies are both costly and complex. Levensen et al. set out to find a surrogate to the human observer—pigeons. Known for being adept at visual discrimination tasks, pigeons can distinguish foreground from background and even discriminate famous paintings, such as distinguishing Monet from Picasso. More importantly, pigeons exhibit functional equivalencies with humans with respect to visual pathways. The investigators wanted to determine whether pigeons could: (i) be trained to discriminate between benign (noncancerous) and malignant (cancerous) pathology and radiology images; (ii) go beyond memorization and discriminate novel (new) stimuli; and (iii) perform discrimination tasks even difficult for domain experts, such as classifying benign and malignant breast masses on x-rays.

Pigeons were trained using food reinforcement to peck on colored response buttons flanking pathology and radiology images rendered on a monitor. The pigeons first experienced a training phase lasting a number of days depending upon the complexity of the task, followed by a testing phase. Image quality was varied including orientation. Surprisingly, pigeons mastered discriminating benign or malignant pathology specimens ranging from low (4x) to high (20x) magnifications. This led the investigators to test pigeons’ ability to detect clinically relevant microcalcifications (small calcium deposits in breast tissue) on x-rays. They found that these avian observers faced limitations when classifying benign and malignant masses on mammograms—a task that is challenging for experienced mammographers.

Although the investigators assert that pigeons are unlikely to be called upon to offer clinical diagnostic support, pigeons may ultimately be useful as observers in assessing new image rendering hardware, software, and analysis tools critical to improving digital image–based diagnoses and risk assessments. These findings highlight an important and insightful shift in the use of animal models along to translational pathway.

R. M. Levenson et al., Pigeons (Columba livia) as trainable observers of pathology and radiology breast cancer images. PLOS ONE 10, e0141357 (2015). [Full Text]

Navigate This Article