CT-based radiomics for differentiating post-thrombectomy hyperdensities

Hyperdensities on post-thrombectomy CT scans are common (incidence: 23–84 %). Identifying the different causes can have clinical implications, particularly in distinguishing between contrast retention and haemorrhagic transformation. While contrast retention is reabsorbed within 24–48 h, haemorrhage tends to remain in the extravascular space for several days, which may delay the reinitiation of antithrombotic treatment in stroke patients. This work proposes a radiomics-based quantitative analysis to differentiate the hyperdense hemisphere from the contralateral one using specific features, paving the way to identify some relationships with hyperdensity types. 80 patients with post-thrombectomy CT hyperdensities (April–December 2024) were included in the study, including both contrast retention and haemorrhagic transformation cases. Hemorrhage subtypes were classified using the ECASS score. A neuroimaging pre-processing pipeline typically used for MRI, was adapted to CT images to perform steps such as orientation correction, thresholding, scalp extraction, linear registration, ventricle segmentation, and erosion. The left and right hemispheres were extracted from the CT images of each patient using a custom-developed automatic procedure. A radiomic analysis was then performed using the open-source package Pyradiomics to extract 18 first order features from the HU histogram of both the hyperdense hemisphere and the contralateral hemisphere. A Wilcoxon paired test was used to investigate differences between the two regions. The comparison between the two hemispheres shows that 16 out of 18 first order radiomic features show significant differences. These variations in the HU histogram reflect tissue differences that can be effectively described by these features. Preliminary results also suggest a possible relationship between some features and clinical procedural information, such as contrast type and quantity. The proposed automatic radiomics-based analysis demonstrated the sensitivity of the radiomic features to distinguish the hyperdense hemisphere from the contralateral one. Further analysis will also investigate the relationship between these features and clinical and procedural factors.