Table 6. List of design recommendations for improving the system from the user study.

Category	Recommendation
Review	1. Allow users to define their custom color schemes for highlights
	2. Include negation rules for keyword search. For example, differentiate between: “mass” and “no mass” 50
	3. Enable top feature highlighting as explanations for the predictions
	4. Distinguish between different kinds of sections in the reports (e.g., Impression and Findings vs. other sections). Allows users to quickly jump to specific sections
Feedback	1. All but one participant gave feedback only at the sentence level even though the tool allowed them to provide feedback at report and section levels as well. Feedbacks may be provided with a single right-click instead of triggering a contextual menu first. Options for other levels may then be provided with a pop-up menu over these highlighted feedback items
	2. Display intelligent blurbs in the feedback list that drew attention to the main findings or keywords (e.g., “ mass ” or “ nodule ”) instead of just the leading part of the sentence
Retrain	1. Allow some free-form comments along with the feedback marking incidental findings. Not only this can serve as a helpful annotation for the other members of the team, the learning pipeline may use that as an additional input to improve models
	2. Some of the predefined search keywords (in pink) raised a lot of false-positives (e.g., “ note ”). An automated mechanism to suggest addition and removal of these terms may be useful