| Abstrakt: |
Introduction:Infarct volume measured from 2,3,5-Triphenyltetrazolium Chloride (TTC)-stained brain slices is critical for in vivostroke models. However, current image tools are not automated and severely limited in their capability. Here, we developed an interactive, trainable software application that computes whole-brain infarct volume metrics from batches of serial TTC-stained brain sections.Methods:Male Wistar Rats (n=21) received middle cerebral artery occlusion and were euthanized after 24 h. Brains were serially sectioned (2 mm, coronal), stained with TTC, and optically scanned (anterior and posterior sides). Manual annotations of infarcted regions and calculation of infarct volume were used as ground truth. We performed computational image analysis to segment brain tissue and develop a custom semantic segmentation model for infarct detection (n=3 training cases) (Figure, top). Stroke parameters, including brain volume, infarct volume, infarct/non-infarct volume ratio (V), and % infarct were computed by measuring each feature per section, averaging over anterior and posterior faces, and summing across all slices. Results were compared against manual ground truth (n=18 testing cases) using accuracy, f1-score, and Spearman correlation. The pipeline was then packaged as a standalone, opensource software named Tectonic Infarct Analysis (Figure, bottom).Results:The software performed well for brain (accuracy=0.95, f1-score=0.90) and infarct (accuracy=0.96, f1-score=0.89) segmentation. Strong, significant correlation was observed between manual and computational calculation of infarct parameters: Vbrain(ρ=0.91, p<0.001), Vinfarct(0.97, p<0.001), Vnon-infarct(0.96, p<0.001), and % infarct (0.81, p=0.001). The software reduced analysis time by 25 min per case vs. manual effort.Conclusions:Tectonic Infarct Analysis software offers a robust and adaptable approach for rapid TTC-based stroke assessment. |
