Purpose: Timely triage in acute PE demands rapid, reliable risk stratification. Automated CTA clot-burden assessment can accelerate mobilization and represents a step forward in the continued efforts to extract clinically predictive signals from diagnostic PE CTs. Using a publicly available dataset, we derive an interpretable exam-level score linked to central emboli and RV strain (RV/LV ≥1). Because clot distribution and pulmonary vascular reserve may predict response to intra-arterial therapy, this score may help guide care {1}.
Materials and Methods: Using the RSNA STR Pulmonary Embolism Detection dataset, we excluded studies with motion or flow artifacts, and trained a 2.5D per-slice classifier on mediastinal-window images (inputs: [slice−1, slice, slice+1]). A study-level test set was held out. The model produced per-slice probabilities. On a validation fold, we chose a slice-probability threshold t* that maximized correlation between a density measure and RV strain, then fixed t* for testing. To summarize clot burden, we combined extent (how many slices are positive) and strength (how convincing the positives are):
density_t = (number of slices with probability ≥ t) / N
S = density_t* × topk_mean
where N is the number of slices per study and topk_mean is the mean of the top 3% of slice probabilities. Higher S indicates greater burden. We evaluated slice- and study-level detection, central emboli, and association with RV strain.
Results: On the test cohort, the per-slice classifier achieved AUC 0.988. Aggregated to the exam level using S, AUC was 0.894 (95% CI 0.861-0.927) with precision 0.839, recall 0.867, and F1 0.851 at t*. S discriminated central emboli (AUC 0.921) and correlated with RV/LV ≥ 1 (Spearman r = 0.688).
Conclusion: An interpretable exam-level score that combines extent and severity quantifies PE burden, aligns with central emboli and RV strain. Embedded in CTA workflows, S can provide rapid, reproducible triage signals, and help identify patients who may benefit from escalation—including intra-arterial therapies.
