Objective : To develop an automated quantitative and pattern-based method for the characterization of Lipiodol deposits on 24h post conventional transarterial chemoembolization (cTACE) CT to establish Lipiodol as an early imaging biomarker of radiographic tumor response.
Methods : This was a retrospective review of prospectively collected clinical trial data (NCT01877187, NCT02753881) of 42 primary and secondary liver cancer patients who underwent cTACE from 2012-2018. On 24h CT, the presence and density of Lipiodol deposition in 65 tumors was automatically characterized using Hounsfield Unit (HU) cut-off values, determined through analysis of intensity distributions in the parenchyma as well as variance-based thresholding techniques applied to the tumor. Also, patterns including homogeneity, sparsity, rim deposition, and peripheral deposition of Lipiodol were automatically characterized using morphological operations on the tumor segmentation mask. After diffeomorphic registration of MRI to CT, Lipiodol deposition was correlated with enhancement on baseline (BL) MRI and tumor response on follow-up (F/U) MRI, using Wilcoxon signed-rank test, Mann-Whitney U test, Kruskal Wallis test, Spearman's rank correlation, and linear regression.
Results : Automatic characterization of Lipiodol densities (low, mid, high) on 24 CT was performed for all lesions using cut-off values of 87 HU, 155 HU and 241 HU (Figure 1a). Tumor areas that deposited Lipiodol became necrotic at a higher rate on F/U MRI than areas without Lipiodol (p=0.0475). Compared to areas with no Lipiodol deposition, the decrease of viable tumor tissue represented by enhancing tumor volume, depositing low, medium and high density Lipiodol was -0.87% ± -15.98 (p=0.3393), -9.32% ± -22.20 (p=0.0066) and -17.91% ± -23.42 (p=0.0003) respectively. With regards to Lipiodol deposition patterns (Figure 1b), homogeneous deposition (p=0.0006), non-sparse deposition (p<0.0001), rim deposition within sparse tumors (p=0.045), and peripheral deposition (p<0.0001) of Lipiodol showed improved response on F/U MRI. Also, tumor enhancement on BL MRI was significantly associated with Lipiodol deposition on 24h CT (p<0.0001), with 8.22% ± 14.59 more Lipiodol coverage in enhancing areas than non-enhancing areas of the tumor on BL MRI.
Conclusions : This study demonstrates that the threshold-based automated characterization of Lipiodol is feasible and that the density and patterns of Lipiodol deposition are strongly correlated with therapeutic effectiveness of cTACE, making them promising imaging biomarkers. The techniques for automated volumetric characterization of Lipiodol deposition presented here are workflow efficient and can be easily incorporated into a standardized framework for cTACE management, paving the way for earlier and more precise response assessment from 24h post cTACE CT.