Back to all papers

Deep learning-reconstructed time-maximum intensity projection versus iterative reconstruction for collateral assessment in acute anterior circulation ischemic stroke.

November 17, 2025pubmed logopapers

Authors

Zheng QF,Sun CF,Zhang YJ,Zhu YX,Dong HB

Affiliations (4)

  • Department of Radiology, The Affiliated Lihuili Hospital of Ningbo University, Ningbo 315041, China.
  • Department of Neurosurgery, The Affiliated Lihuili Hospital of Ningbo University, Ningbo 315041, China.
  • Shanghai United Imaging Healthcare Co., Ltd., Shanghai 201807, China.
  • Department of Radiology, The Affiliated Lihuili Hospital of Ningbo University, Ningbo 315041, China. Electronic address: [email protected].

Abstract

To evaluate the deep learning-reconstructed time-maximum intensity projection (DLR t-MIP) in acute patients with anterior circulation ischemic stroke, and to compare its ability to assess collateral circulation and predict 90-day outcomes with that of single-phase CTA (SP-CTA), multiphase CTA (MP-CTA), and CT perfusion (CTP) volumetrics. This retrospective study analyzed data from 75 AIS patients with unilateral anterior circulation large-vessel occlusion who underwent one-stop CT angiography and CT perfusion (CTA-CTP) scanning. The raw data were reconstructed using iterative reconstruction (IR) algorithms and deep learning algorithms. DLR t-MIPs were generated from dynamic brain scan data; axial MIPs were 24/4 mm. Subjective vessel delineation and noise (5-point scales) and objective attenuation, noise, signal-to-noise ratio (SNR), and contrast-to-noise ratio (CNR) were compared. Collaterals were graded by Menon single-phase (SP-CTA, DLR t-MIP) or multiphase criteria (MP-CTA). Ninety-day mRS defined outcomes (good ≤ 2 vs poor ≥ 3). Prognostic performance was evaluated with a ROC curve and pairwise DeLong tests. Subjective quality scores revealed that the vascular imaging quality was significantly greater in the DLR t-MIP group (p < 0.05), with high interobserver agreement (kappa = 0.716). Objectively, DLR t-MIP demonstrated a significantly lower level of image noise (p < 0.001) and significantly higher vascular CT attenuation values, SNRs, and CNRs (all p < 0.001) than SP-CTA. Collateral scores correlated inversely with the final infarct volume (FIV): DLR t-MIP r =  - 0.73 (95 % CI, -0.80 to -0.51), SP-CTA r =  - 0.64 (-0.78 to -0.45), and MP-CTA r =  - 0.72 (-0.83 to - 0.57); all p < 0.001. The AUCs for 90-day outcomes were as follow: DLR t-MIP 0.84 (95 % CI 0.74-0.92), SP-CTA 0.73 (95 % CI 0.62-0.82), MP-CTA 0.84 (95 % CI 0.71-0.91), ischemic core volume 0.83 (95 % CI 0.73-0.92), and hypoperfusion volume 0.85 (0.75-0.94). DeLong testing showed that DLR t-MIP > SP-CTA (ΔAUC + 0.113; p = 0.012) and ≈ MP-CTA (ΔAUC + 0.002; p = 0.957). DLR t-MIP consolidates temporal information from CTP into a single high-quality angiographic dataset, yielding superior image quality and better prognostic discrimination than SP-CTA, and comparable performance to MP-CTA and CTP volumetrics-without additional acquisition. This approach may streamline acute-stroke triage and collateral-guided decision-making.

Topics

Journal Article

Ready to Sharpen Your Edge?

Join hundreds of your peers who rely on RadAI Slice. Get the essential weekly briefing that empowers you to navigate the future of radiology.

We respect your privacy. Unsubscribe at any time.