Texture Analysis of Sub-massive Pulmonary Embolism to Predict Mechanical Thrombectomy Success

To identify textural characteristics of sub-massive pulmonary embolism (PE) on CT imaging that predict technical difficulty and success of mechanical thrombectomy.

Materials and Methods:

All patients with sub-massive PE with post contrast CT confirming diagnosis treated from 2020 to 2022 were included. Image analysis was done using 3D Slicer. The clot in each pulmonary artery and its proximal branches was segmented and analyzed separately. Grey-level correlation matrices were generated yielding eight distinct textural outputs. Clot yields were classified as low, moderate, or high yield (1,2,3) based on physician documentation. Number of aspirations, and device time were also recorded. To capture technical difficulty, a ratio of number of aspirations / clot yield was calculated.

Results:

Fourteen cases of patients undergoing mechanical thrombectomy for sub-massive PE using the Inari FlowTriever device (Irvine, CA) were included. Eleven cases had bilateral clot removal yielding a total of 25 clots for analysis. Mean age was 63.5 years (SD: 17.6). 13/25 clots had high yields, 7/25 had moderate yields, and 5/25 had low yields. The median number of aspirations was 5 (3, 6) and the median device time was 60 min (43.5, 80). One-way ANOVA analysis revealed the Haralick textural coefficient was significantly different (p 0.042) between clots in the low yield (median 1.19×10⁷ IQR 1.01×10⁷- 2.12×10⁷), moderate yield (median 2.44×10⁷ IQR 1.74×10⁷ - 2.85×10⁷), and high yield (median 3.83×10⁷ IQR 1.72×10⁷- 6.03×10⁷) groups. Further, a binary logistic regression model demonstrated that a higher Haralick coefficient had predictive value for high clot yield compared to low clot yield (AUC 0.80). Linear regression showed the Haralick coefficient inversely correlated with the number of aspirations normalized to clot yield (r = -0.47, p 0.017). Two additional texture features, cluster shade and cluster prominence correlated with device time (r = 0.57, p 0.004 and r = 0.52, p 0.009, respectively).

Conclusion:

Our results suggests that higher Haralick coefficient, a measure of correlation of voxels relative to each other, is a predictor for higher clot yield and lower difficulty in mechanical thrombectomy of PE. Lower cluster shade and cluster prominence values, measures of asymmetry in voxels, were associated with decreased FlowTriever device time. These metrics may provide insight into the chronicity of clot and aid in patient selection and technical strategy to optimize chances of procedural success.