{
  "abstract": "Introduction Neurointerventional outcomes for acute ischemic stroke treatment depend on clot composition and may also be influenced by clot contraction. For instance, RBC-rich clots are softer and more responsive to aspiration and thrombolysis, whereas fibrin-rich clots resist both. Thus, pre-interventional assessment of clot composition and contraction could inform treatment strategy and improve patient outcomes. However, current imaging markers are indirect and qualitative. To address this gap, we conducted the first in vitro study to test whether MRI and radiomics can reliably predict clot composition and contraction.Materials To this end, we prepared blood clots spanning clinically observed compositions (0-80% RBCs) in contracted and uncontracted states. We imaged clots using quantitative sequences (T1, T2, ADC mapping) and standard clinical sequences (T2 RARE, SWI, T1 GRE). Using these data, we tested whether MRI signal intensities, quantitative parameters, and radiomic features could predict clot hematocrit and classify clots by composition (RBC-rich vs. fibrin-rich) and contraction state.Results Because clot hematocrit influences treatment, we first tested associations between hematocrit and MRI signal. Quantitative parameters (T1, T2, ADC) decreased with increasing hematocrit (R 2 up to 0.85). Clinical sequences showed similar trends but weaker associations (R2 up to 0.62). To extract additional information from MRI, we used radiomic features to predict hematocrit, which improved model performance. Quantitative sequences remained strongest (R2 up to 0.87), with T1 and ADC mapping yielding the most accurate predictions. Clinical sequences again provided moderate predictions (R2 up to 0.66; ADC and T2 RARE regression shown in figure 1).Distinguishing RBC- from fibrin-rich clots may be sufficient to guide treatment, so we next tested radiomic feature-based classification. Quantitative MRI achieved the strongest discrimination (AUCs > 0.94), with ADC features achieving near-perfect classification. Importantly, clinical sequences also provided robust classification (AUCs > 0.84), indicating that clinically acquired MRI may already capture compositional information (ADC and T2 RARE ROC curves shown in figure 1). Since clot contraction can also influence treatment, we tested whether MRI could distinguish contracted versus uncontracted clots. This proved more challenging, but features from T1 and T2 maps still achieved strong discrimination (AUCs of 0.88), as did T2 RARE (AUC of 0.81).Conclusion Overall, MRI-based radiomic analysis can quantitatively characterize clot composition and contraction in vitro. While quantitative MRI provides the strongest characterization, standard clinical stroke sequences also perform well. Our findings demonstrate the feasibility of MRI-based radiomics for pre-intervention clot characterization, though in vivo validation is needed for clinical translation.Disclosures G. Bechtel: None. J. Fuhg: None. J. Tamir: None. H. Saber: None. M. Rausch: 2; C; Edwards Lifesciences.Abstract E-149 Figure 1",
  "authors": [
    {
      "affiliations": [
        "Department of Biomedical Engineering, The University of Texas at Austin, Austin, TX"
      ],
      "name": "G Bechtel"
    },
    {
      "affiliations": [
        "Department of Aerospace Engineering and Engineering Mechanics, The University of Texas at Austin, Austin, TX"
      ],
      "name": "J Fuhg"
    },
    {
      "affiliations": [
        "Department of Electrical and Computer Engineering, The University of Texas at Austin, Austin, TX"
      ],
      "name": "J Tamir"
    },
    {
      "affiliations": [
        "Department of Neurology and Neurosurgery, Dell Medical School, The University of Texas at Austin, Austin, TX"
      ],
      "name": "H Saber"
    },
    {
      "affiliations": [
        "Department of Biomedical Engineering, The University of Texas at Austin, Austin, TX"
      ],
      "name": "M Rausch"
    }
  ],
  "title": "E-149 MRI-based radiomic analysis of clot composition and contraction to guide neurointervention",
  "uid": "0281ad8e-ecad-5ce0-af23-0e1aad969cda"
}