Visual perception studies and observer models in medical imaging.

Area of Science:

• Medical Imaging
• Radiology
• Human Observer Performance

Background:

• Receiver operating characteristic (ROC) studies are familiar to academic radiologists.
• Human observer performance is often assessed using forced-choice methods, both rooted in signal detection theory.
• ROC analysis yields a true-positive versus false-positive probability curve, summarized by the area under the curve (A(Z)).

Purpose of the Study:

• To summarize 30 years of research on human signal detection capabilities.
• To review observer models and their evaluation using human experimental data.
• To discuss the application of these findings in imaging system design and image quality metrics.

Main Methods:

• Description of forced-choice experimental methods and data analysis.
• Utilizing signal detection theory as the foundational framework for both ROC and forced-choice methods.
• Comparing summary performance measures like A(Z) from ROC and percentage correct (PC) from forced-choice under controlled conditions.

Main Results:

• Under controlled experimental conditions, theoretical predictions show a simple relationship between summary measures (A(Z) and PC).
• Human results are consistent with theoretical predictions, validating the methods.
• Human experiment results have been instrumental in evaluating and refining theoretical observer models.

Conclusions:

• Observer models correlating well with human performance can aid in preliminary imaging system design.
• These models are valuable for selecting appropriate image quality metrics for comparing equipment performance.
• A comprehensive understanding of human signal detection is essential for the development of improved medical imaging technologies.