ESTRO 2024 - Abstract Book

S2918

Interdiscplinary - Other

ESTRO 2024

within MIM Software (Cleveland, OH) using a client-server architecture. Following the standard workflow, 34 consecutively recruited patient datasets including manual contours from the contributing centres were uploaded to the clinical trials portal and these datasets were then transferred to MIM. The MIM extension client is was run, which transferred the MRI image along with the CTV contour to a separate server. The CQA tool runs on the server and generates a PDF report (figure 1) including acceptance of the contour or the need for a manual review and is returned to the MIM client for the user to download. For this investigation all reports were then forwarded to the reviewing radiation oncologist who also provided a manual review after reviewing the report. Based on previous investigations the tolerance criteria on the tool was set at 0.85 for DSC, with the goal of flagging all true violations for manual review, accepting that some true passes would also be flagged for manual review and only flagging acceptance by the tool for true passes. An assessment of the manual review of the contours compared to the automated assessment was undertaken. Sensitivity and specificity were calculated, as well as determining the number of cases where the tool correctly identified as acceptable, with the potential of removing the need for manual review and a subsequent reduced workload.

Figure 1: Sample report generated by the automated CQA tool depicting a case with acceptable protocol compliance.

Results:

The automated CQA tool was successfully implemented in the NINJA clinical trial workflow, meeting required trial timelines for contour review. Figure 2 shows the confusion matrix comparing the assessment of the automated CQA tool with manual review. 12/34 patient datasets were correctly flagged as acceptable, representing a potential reduced workload of 35% compared to a manual review of all cases. The use of a rectal hydrogel spacer was a