Date Approved

5-11-2026

Graduate Degree Type

Thesis

Degree Name

Applied Computer Science (M.S.)

Degree Program

School of Computing and Information Systems

First Advisor

Ali Raza

Second Advisor

Mario Fific

Third Advisor

Denton Bobeldyk

Academic Year

2025/2026

Abstract

Human visual search involves the identification of relevant signals within information-rich environments, which is a fundamental problem in visual perception. While detection accuracy and response time are commonly used to evaluate performance in visual search, these measures do not reveal the underlying cognitive and computational structure that produces observable behavior. A key challenge lies in distinguishing between competing processing architectures, particularly in complex visual domains where different models can produce similar behavioral outcomes. This study addresses this challenge by developing a computational experimental framework for analyzing visual search behavior using System Factorial Technology (SFT). The experimental framework integrates naturalistic medical image data and controlled stimulus manipulation to enable human architecture-level analysis. A total of 191 undergraduate participants from the College of Liberal Arts and Sciences completed three experimental conditions to search for lung cancer nodules. In the first condition, participants performed a target detection task without feedback, establishing baseline performance. The results suggested that the participants were randomly searching, with a lower accuracy of 31 percent. In the second condition, trial-by-trial feedback was introduced to evaluate the effect of feedback on detection accuracy and response patterns. The accuracy of nodule detection for this condition increased by 6 percent. In the third condition, a computationally generated circular stimulus was embedded at target cancer lung nodules, with intensity dynamically adjusted to match local image statistics. We found a significant improvement in visual search to 74 percent accuracy, up from a 31 percent baseline. Results show that masking type, feedback, and cueing influence the stopping rules and cognitive architecture used to interpret lung medical images. We also validated the use of SFT in the diagnosis of complex medical health images. Overall, this work demonstrates a scalable computational framework for analyzing human cognitive performance and informing the design of intelligent visual systems.

Download

Share

COinS