A Semi-Automated Usability Evaluation Framework for Interactive Image Segmentation Systems.

Area of Science:

• Computer Vision
• Human-Computer Interaction (HCI)
• Medical Image Analysis

Background:

• Fully automated segmentation accuracy is limited for complex tasks.
• Semi-automatic methods provide user benefits for precise segmentation of limited datasets.
• Objective comparison of HCI aspects in novel interactive segmentation systems (ISS) is often lacking.

Purpose of the Study:

• To propose an objective method for comparing interactive segmentation systems (ISS).
• To evaluate user experience and interface design in interactive image segmentation.
• To reduce the resources needed for evaluating variations in ISS prototypes.

Main Methods:

• Extensive user studies were conducted to gather qualitative and quantitative data.
• Qualitative content analysis of user feedback (visual and verbal).
• Direct assessment using System Usability Scale (SUS) and AttrakDiff-2 questionnaires.
• Approximation of usability findings from system-measurable user actions.

Main Results:

• User experience varies substantially between ISS prototypes, even with identical algorithms.
• Users prefer simpler interfaces and greater control over segmentation steps.
• An automated evaluation scheme predicted questionnaire results with an 8.9% average relative error.
• The proposed method approximates usability findings from system-measurable user actions.

Conclusions:

• An objective, automated method for evaluating ISS usability is feasible and effective.
• This approach can significantly reduce the resources required for ISS development and refinement.
• Understanding HCI aspects is crucial for improving user experience in interactive image segmentation.