Assessing the accuracy and clinical utility of an algorithm-based PET/CT service for quantifying treatment response in older adults with metastatic cancer.

Abstract

e13649 Background: There is substantial variability in PET/CT interpretation, influenced by technical, procedural, and reader-dependent factors. This variability is increased in older adults due to age related physiologic changes, highlighting the need for standardization of PET/CT interpretation. This study aims to assess the accuracy of a novel software, TRAQInform IQ, in evaluating change over time in PET/CT imaging for older adults with metastatic cancer and investigate perceived clinical usefulness. Methods: Serial PET/CT scans for 17 older adults aged 65 and older with metastatic cancer were retrospectively analyzed using the algorithm-based software TRAQInform IQ which assesses spatial location and classification of individual lesions in addition to overall disease response. An independent radiologist reviewed clinically significant lesions as determined by oncologists for contour, matching precision, and treatment classification across serial scans. The software report was compared to existing original PET/CT radiology reports for overall disease assessment. To assess usefulness, 4 oncologists reviewed specialty-specific cases and completed a questionnaire comparing the algorithm-based reports to the PET/CT scans. Descriptive statistics evaluated demographics, report findings, and usefulness. Results: Patients had a mean age of 72 years old and a majority had stage IV (71%) disease at diagnosis. Seven primary cancers were represented, including cervical, melanoma, lymphoma, vulvar, vaginal, uterine, and urothelial. Of 73 lesions deemed clinically significant, the accuracy of lesion contours, matching, and classification across serial scans was 97% per an independent radiologist. The software reported disease that was increasing in 3 (18%), and decreasing in 13 (76%) patients, with 1 patient (6%) with no lesions reported. There were distinctions in overall disease assessment between the software and PET/CT report in 7 (41%) cases. Treating oncologists deemed 54% of the reports useful and found the software most useful for spatial information (75%), followed by systemic treatment decision and patient education (54%). Conclusions: The novel algorithm-based software TRAQInform IQ created to quantify and classify change in lesions across serial PET/CT scans in conjunction with standard radiology reports was highly accurate. There was notable variability in appreciated usefulness across treating physicians. This analysis provides novel data to better understand how this technology can be integrated in practice and assist with standard clinical judgment and management decisions.