Date of Award
5-2024
Document Type
Thesis
Degree Name
Master of Science (MS)
Department
Applied Psychology
Committee Chair/Advisor
Dr. Dawn Sarno
Committee Member
Dr. Mark Neider
Committee Member
Dr. Patrick Rosopa
Abstract
Radiologists use computer-aided detection (CAD) systems to assist with detecting and assessing breast cancers in mammograms, digital breast tomosynthesis (DBT), and other types of breast imaging; however, the usefulness of such automation aids has been debated since their implementation. Initial studies on CAD systems had mixed results, while more recent studies have shown that they can improve diagnostic outcomes (i.e., greater sensitivity and fewer false alarms) and decrease the reading time required for images. Three CAD types are currently in use: binary, analog, and interactive CAD. However, studies rarely explore the differences between the CAD methods. Additionally, recent work has suggested the benefit of including Breast Imaging Reporting and Database System (BI-RADS) ratings in CAD systems to help standardize assessments and improve diagnostic performance. The present study adds to the literature by comparing the three existing types of CAD (binary, analog, and interactive) and including a novel CAD system with BI-RADS ratings. Fifty undergraduate students completed a visual search task with mock DBT images and were assigned to either one of the CAD conditions or a control. Participants also completed surveys regarding their propensity to trust automation in general and their perceptions of the system’s usability and trustworthiness. Results suggested that the binary and analog CAD systems improved participants’ hit rate and sensitivity (d’). Additionally, participants in the analog and interactive CAD conditions appeared to trust their CAD aids more than those in the binary and BI-RADS conditions. Regarding perceived usability, participants in the binary, analog, and interactive CAD conditions rated their CAD aids with higher usability scores than those in the BI-RADS CAD condition. Exploratory analyses provided support for the trust and usability findings in the present study and suggested that participants in the BI-RADS condition were more conservative than those in the other CAD conditions. Together, the results from the present study provide further support that the use of CAD aids, particularly those that provide additional information, can improve hit rate and sensitivity when assessing DBT images. Although the novel BI-RADS CAD aid did not perform as strongly as the other CAD systems, it is promising that this variation did not harm reader performance. More research is required to explore how different implementations of BI-RADS ratings within CAD systems may be beneficial in the assessment of DBT images. Findings from this research contribute to the development of more user-centered CAD systems to ultimately improve radiologists’ diagnostic performance, particularly for DBT images.
Recommended Citation
Sabo, Katharine, "Detection and Assessment of Targets in Mock Digital Breast Tomosynthesis Images Using Computer-aided Detection Systems" (2024). All Theses. 4219.
https://tigerprints.clemson.edu/all_theses/4219
Author ORCID Identifier
0009-0000-8793-9325