Click here to


Are you sure ?

Yes, do it No, cancel

JACK KROHMER JUNIOR INVESTIGATOR COMPETITION WINNER: Model Observers Predict Radiologists' Detection of Retained Foreign Objects in Homogenous and Anthropomorphic Backgrounds

K Blunt1*, E Marshall2, J Cruz Bastida3, A Clarey4, Z Lu5, A Dachman6, I Reiser7, (1) The University of Chicago Medicine, Chicago, IL, (2) University of Chicago, Boynton Beach, FL, (3) University of Chicago, Chicago, IL, (4) University of Chicago, ,,(5) University of Chicago, Chicago, IL, (6) University of Chicago, Chicago, ,(7) The University of Chicago, Chicago, IL


(Tuesday, 7/14/2020) 2:00 PM - 3:00 PM [Eastern Time (GMT-4)]

Room: Track 1

Purpose: To correlate quantitative image quality metrics with radiologists’ detectability score and to develop optimized radiographic techniques for identifying retained suture needles.

Methods: A suture needle phantom was constructed by arranging nine taper-style steel suture needles measuring from 5 mm to 13 mm in length in a 3x3 grid. Portable radiographs of the needle phantom were obtained in various thicknesses of homogenous soft tissue-equivalent material (Superflab bolus material, Mick Radio-Nuclear Instruments, Inc.) and at various locations in an anthropomorphic phantom (Alderson phantom patient). One board-certified abdominal radiologist with 30+ years of experience and three radiology residents scored suture needle visibility on a diagnostic review monitor using a 4-point Likert scale (not visible, barely visible, visible, obviously visible). A template-matching observer model was implemented to calculate suture needle index of detectability (d’) in homogenous backgrounds, and suture needle contrast-to-noise ratio (CNR) was computed in anthropomorphic phantom backgrounds. Correlation between these quantitative image quality metrics and radiologists’ qualitative rankings were assessed using Kendall’s tau.

Results: Radiologists’ assessment correlated to d’ with a Kendall’s tau of 0.77 (p<0.01), and to CNR with a Kendall’s tau of 0.62 (p<0.01). To achieve a “visible” score, a d’ of 26 in a homogenous background and a CNR of 12 in anthropomorphic background are required.

Conclusion: We have correlated d’ and CNR with radiologists’ assessment of needle visibility in homogenous and anthropomorphic phantom backgrounds, respectively. This work will facilitate the development of a patient- and needle size-specific radiographic technique chart based on target d’ and CNR values.


Not Applicable / None Entered.


Not Applicable / None Entered.

Contact Email