(Wednesday, 7/17/2019) 10:15 AM - 12:15 PM
Room: 304ABC
Purpose: Comprehensive response assessment of patient imaging entails visual analysis of individual lesions, which is time-consuming and error-prone. In this work, we present a registration-based automated method for soft-tissue lesion matching in longitudinal PET/CT image series and assess its performance when preceded by different registration methods.
Methods: ¹�F-FDG PET/CT scans of 19 metastatic melanoma patients were retrospectively collected for analysis. Disease sites larger than 1 cm³ were identified from clinical reports and visually matched between sequential scan pairs (N=76). Two levels of registration were investigated: whole-body rigid registration (WBR), and body-segment articulated registration (ARR). ARR follows WBR and consists of splitting the patient image into skeletal segments and rigidly registering each segment independently. In the automated matching that followed registration, positive matching decisions were made for lesions that fell within 6 cm of each other. One-to-one lesion mapping was guaranteed by global minimization of distances between lesions. Using visual matching as ground truth, we categorized each matching decision as either true positive, false positive, true negative, or false negative and assessed the accuracy, sensitivity, and specificity of lesion matching for each registration method. Differences in matching performance were assessed using McNemar's test.
Results: With 553 matching decisions made, ARR outperformed WBR by achieving an accuracy of 99% vs 96%, sensitivity of 98% vs 93%, and specificity of 99% vs 98%. The matching improvement from WBR to ARR was statistically significant (p≤0.001). For ARR-based matching, 73/76 of sequential scan pairs completely agreed with visual matching. The 3 cases with incorrect matching required 70, 68, and 13 matching decisions respectively, compared to the population median of 4.
Conclusion: The presented method exhibited high accuracy, sensitivity, and specificity. The use of articulated registration resulted in performance superior to whole-body rigid registration. Incorrect matching decisions were most prevalent in cases of high disease burden.
Funding Support, Disclosures, and Conflict of Interest: Robert Jeraj is a co-founder of AIQ Solutions. This work was supported by the University of Wisconsin Carbone Cancer Center Support Grant P30 CA014520.
Keywords
PET, Registration, Image Analysis
Taxonomy
IM- Multi-modality imaging systems: CT/PET - human
Contact Email
