Interobserver agreement of gradient-based and threshold-based segmentation for longitudinal TTV and TLA on post-therapy $$\mathrm {[^{177}Lu]}$$ Lu-PSMA-I&T SPECT/CT

Abstract

Quantitative post-therapy SPECT/CT imaging is increasingly used to derive total tumour volume (TTV) and total lesion activity (TLA), which have emerged as potential early biomarkers of response to \(\mathrm {[^{177}Lu]}\) Lu-PSMA-I&T (imaging and therapy). This study compared the interobserver reproducibility of longitudinal TTV and TLA assessment using gradient-based and threshold-based segmentation methods on post-therapy SPECT/CT. Six patients underwent four cycles of \(\mathrm {[^{177}Lu]}\) Lu-PSMA-I&T, with post-therapy SPECT/CT acquired 19 ± 1 h after each cycle using a Spectrum Dynamics Veriton 200 system. Twenty-four post-therapy \(\mathrm {[^{177}Lu]}\) Lu-SPECT/CT scans were independently analysed by three expert readers. For the gradient-based segmentation, lesions were visually identified and delineated using an edge-detection algorithm. The threshold-based approach applied fixed lean-body-mass-normalised standardised uptake value ( \(\mathrm {SUV_{LBM}}\) ) thresholds of 2, 3, 4, 6, 8, and 10, followed by reader-based exclusion of physiological uptake regions (LesionID, MIM Software). Segmentation time was recorded for each method. Interobserver agreement was assessed using the intraclass correlation coefficient ( ICC (2, 1)) with absolute agreement, and longitudinal reproducibility was quantified using the minimal detectable change ( \(MDC_{\Delta }\) ) derived from between-cycle changes in TTV and TLA. Associations between the gradient- and threshold-based measurements were evaluated using Pearson correlation and Bland-Altman analysis. Gradient-based segmentation required a median of 4.8 min [interquartile range (IQR): 2.5−10.6] per scan, whereas all threshold-based segmentations were completed within 2.5 min. Gradient TTV ( \(\mathrm {TTV_{Gr}}\) ) demonstrated moderate interobserver agreement ( \(ICC(2,1)=0.57\) ) and a corresponding \(MDC_{\Delta }\) of 78.3%. In contrast, gradient TLA ( \(\mathrm {TLA_{Gr}}\) ) showed excellent agreement ( \(ICC(2,1)=0.92\) ) with an \(MDC_{\Delta }\) of 50.9%. All threshold-based approaches achieved near-perfect interobserver agreement ( \(ICC(2,1)\approx 1.0\) ) and substantially lower \(MDC_{\Delta }\) values, ranging from 0.1 to 6.8% for TTV and from 0.1 to 4.6% for TLA. Threshold-based TLA ( \(\mathrm {TLA_{Th}}\) ) showed moderate correlation with \(\mathrm {TLA_{Gr}}\) at lower thresholds (2-6) and strong correlation at higher thresholds (8 and 10). Gradient-based TLA assessment showed higher reproducibility than gradient-based TTV for longitudinal response evaluation on post-therapy \(\mathrm {[^{177}Lu]}\) Lu-PSMA-I&T SPECT/CT. The high \(MDC_{\Delta }\) of gradient-based TTV indicates limited sensitivity for detecting smaller longitudinal changes, whereas threshold-based segmentation provided a faster and more reproducible approach with substantially lower \(MDC_{\Delta }\) values. Further validation is required to determine the optimal \(\mathrm {SUV_{LBM}}\) threshold and its clinical outcome relevance.