Md Sazidur Rahman, Kjersti Engan, Kathinka Dæhli Kurz, Mahdieh Khanmohammadi
6 min
Abstract
Computed tomography perfusion (CTP) at admission is routinely used to estimate the ischemic core and penumbra, while follow-up diffusion-weighted MRI (DWI) provides the definitive infarct outcome. However, single time-point segmentations fail to capture the biological heterogeneity and temporal evolution of stroke. We propose a bi-temporal analysis framework that characterizes ischemic tissue using statistical descriptors, radiomic texture features, and deep feature embeddings from two architectures (mJ-Net and nnU-Net). Bi-temporal refers to admission (T1) and post-treatment follow-up (T2). All features are extracted at T1 from CTP, with follow-up DWI aligned to ensure spatial correspondence. Manually delineated masks at T1 and T2 are intersected to construct six regions of interest (ROIs) encoding both initial tissue state and final outcome. Features were aggregated per region and analyzed in feature space. Evaluation on 18 patients with successful reperfusion demonstrated meaningful clustering of region-level representations. Regions classified as penumbra or healthy at T1 that ultimately recovered exhibited feature similarity to preserved brain tissue, whereas infarct-bound regions formed distinct groupings. Both baseline GLCM and deep embeddings showed a similar trend: penumbra regions exhibit features that are significantly different depending on final state, whereas this difference is not significant for core regions. Deep feature spaces, particularly mJ-Net, showed strong separation between salvageable and non-salvageable tissue, with a penumbra separation index that differed significantly from zero (Wilcoxon signed-rank test). These findings suggest that encoder-derived feature manifolds reflect underlying tissue phenotypes and state transitions, providing insight into imaging-based quantification of stroke evolution.
Alex: Okay, so it's like sorting photos of kids into groups based on who was healthy at the start and who got sick later—to spot early warning signs in their faces. That gives six groups total?
Sam: Yes—core that stayed dead, core that somehow recovered, penumbra that survived, penumbra that infarcted, healthy that stayed fine, and healthy that unexpectedly died. Inside each zone, they pull out hidden patterns from the first scan using special computer tools trained on tons of brain images. These tools scan the image in layers, first spotting basic edges and brightness, then building up to complex textures and flows that humans miss—like how a video game AI learns to recognize enemies by piecing together pixels into shapes and movements.
Alex: And those patterns from the computer tools... do they naturally bunch up by whether the tissue lived or died?
Sam: Precisely—the study finds that deep patterns cluster tightly: salvageable penumbra looks a lot like healthy tissue in this pattern space, while doomed areas stand apart. They measure this by comparing how similar patterns are within the same fate group versus across groups, using a score called cosine separation—think of it as the angle between arrows pointing to each tissue's description; small angles mean similar, wide angles mean different. The difference, or Δcos, shows clear separation for deep features, suggesting these scans capture invisible tissue traits right at admission that predict evolution. This is a notable step in understanding stroke variety from images alone.
Alex: That's a clear edge for the deep ones... but does summarizing patches risk ignoring subtle variations within a zone?
Sam: It does simplify, focusing on strongest signals per zone, which suits group differences but might smooth over fine patient variety—the paper notes this as a direction for future work with full maps. Still, the separation holds across their cohort, pointing to real predictive power in admission images.
Alex: So in essence, fusing the zones lets these features reveal tissue destinies unfolding... a meaningful way to peek beyond snapshots.
Alex: With just 18 patients, how solid is this for real clinics?
Sam: The paper highlights key constraints. The small group size limits statistical strength and how widely results apply—it's framed as a feasibility check, not final proof. Co-registration between the angled CT perfusion scans and follow-up MRI can misalign tiny spots due to head tilt differences, potentially blurring tissue matches. All data came from one center, so patient variety might differ elsewhere.
Alex: Fair points—misalignments and small numbers could muddy the clusters. Still, if it scales, how might this change stroke treatment calls?
Sam: The approach suggests admission scans hold hidden tissue traits that predict survival, letting doctors spot salvageable areas more reliably without waiting. In practice, it could guide personalized blood-flow restoration—say, pushing aggressive treatment for penumbra leaning salvageable based on feature clusters. Future steps include bigger, multi-center tests to confirm.
Alex: Huh... so despite limits, it's a feasible step toward reading tissue fates from first scans alone. Ties those early patterns directly to outcomes.
Sam: Exactly—this framework shows deep features from perfusion data can phenotype stroke tissue evolution meaningfully. It opens paths for better predictions, though validation needs larger groups.
Alex: That's a grounded look at using bi-temporal imaging to uncover stroke tissue destinies. Thanks for joining us on ResearchPod.