Investigation of Human and Model Observer Performance of Mass Detection in a Prototype Digital Breast Tomosynthesis System

Abstract

Task-based assessment of image quality using a theoretical observer model has recently gained attention in medical imaging fields. Observer models whose performance can suitably match that of a human observer under various imaging conditions have been considered a key concept for virtual clinical studies. This work focused on experimental study in which a prototype digital breast tomosynthesis (DBT) system, developed by the Korea Electrotechnology Research Institute, was used to compare the task-based Fourier metrics of a detectability index with the performance of human observers. Different angular ranges of ±7°, ±10.5°, ±14°, ±17.5°, ±21°, and ±24.5° were used to detect four different sizes of spheroidal masses with the same 15-projection samplings. A total of 16 human observers participated in the study. Human observer performance was measured using four-alternative forced choice (4AFC) tests for different detection tasks, including detecting spheroidal masses. To determine the task-based detectability index (d′), the non-prewhitening matched filter observer was calculated by analyzing the task function, local spatial resolution, and local noise of spheroidal masses. The percentage of signals correctly detected (Pc) with the 4AFC tests was then compared with the detectability index for tasks presented in 2D slices (dslice′2). The average Pc for 16 human observers was 0.87 (range = 0.56–1). Results showed that Pc decreased as the angular range increased from ±7° to ±24.5° with different mass sizes. Our result also showed that the performance of the task-based theoretical model observer could reasonably estimate d′ for all inserted lesions of different sizes and acquired under different conditions, while presenting a trend similar to those of the human observers. Consequently, our study is expected to contribute to quantitative imaging performance analysis under various DBT imaging acquisition parameters.