Longitudinal imaging markers for pancreas cancer tumor response via Eshelby biomechanics

<strong>Purpose</strong>: Pancreatic ductal adenocarcinoma (PDAC) often presents with poorly defined lesions and low imaging contrast, which impedes accurate assessment of therapeutic response. Current non-invasive clinical markers are limited to RECIST and blood-based biomarkers CA19-9 and CEA, which are widely regarded to be insufficient descriptors of PDAC response to therapy. We propose a contrast-independent and modality-independent biomechanical image registration workflow for characterizing therapeutic response by leveraging the Eshelby inclusion problem to estimate the transformation strain of the tumor from whole pancreas segmentations across the treatment interval. <br/><br/> <strong>Methods</strong>: An Eshelby inclusion model was constructed to parameterize pancreatic mass effect that arises due to transformation strain of the tumor. This model was integrated into a biomechanical image registration framework to simultaneously resolve changes in tumor and parenchymal shape between baseline and restaging imaging. In 25 patients with measured pathologic response to a standardized neoadjuvant chemotherapy regimen, Eshelby transformation strain was compared to RECIST v1.1 score and changes in serum CA19-9 and CEA as indicators for therapeutic response. Correlations between each variable and pathologic response were assessed using Spearman’s <i>ρ</i>, and prognostic value towards stratifying pathologic response was evaluated via Wilcoxon rank sum tests at a significance level of <i>α</i>=0.05. <br/><br/> <strong>Results</strong>: Tumor transformation strain was significantly associated with pathologic response (<i>ρ</i>=0.55, <i>p</i>=0.008), whereas the change in CA19-9 (<i>ρ</i>=0.29, <i>p</i>=0.19), RECIST score (<i>ρ</i>=0.09, <i>p</i>=0.71), and change in CEA (<i>ρ</i>=0.07, <i>p</i>=0.75) did not exhibit significant correlation. Current clinical markers for therapeutic response were not able to stratify pathologic response based on above- versus below-median values of each marker (RECIST: <i>p</i>=0.49; ΔCA19-9: <i>p</i>=0.24; ΔCEA: <i>p</i>=0.54), whereas the Eshelby transformation strain effectively distinguished patients who achieved more favorable pathologic response (<i>p</i>=0.009). <br/><br/> <strong>Conclusions</strong>: The Eshelby biomechanical model of tumor transformation strain offers a modality- and intensity-independent mechanism for characterizing therapeutic response that surpasses current clinical markers for PDAC.