Automated collateral assessment restricted to the hypoperfused area for distal vessel occlusions in ischemic stroke

Lucas de Vries, M M Q Robbe, Ivo G.h. Jansen, Saeid Mojtahedi, Jan W. Hoving, Susanne G. H. Olthuis, Robrecht R. M. M. Knapen, Florentina M.e. Pinckaers, Manon Kappelhof, Ludo F.m. Beenen, Alida A. Postma, Robert J. van Oostenbrugge, Diederik W.j. Dippel, Efstratios Gavves, Bart J. Emmer, Charles B.l.m. Majoie, Wim H. van Zwam, Henk A. Marquering, on behalf of the Mr Clean Registry Investigators, Aad van der Lugt, Yvo B. W. E. M. Roos, Jelis Boiten, Jan Albert Vos, Study coordinators, Maxim J. H. L. Mulder, Robert-Jan B. Goldhoorn, Kars C. J. Compagne, Josje Brouwer, Sanne J. den Hartog, Wouter H. Hinsenveld, Lotte van den Heuvel, Local principal investigators, Bob Roozenbeek, Pieter Jan van Doormaal, Jonathan M. Coutinho, Wouter J. Schonewille, Marieke J. H. Wermer, Marianne A. A. van Walderveen, Adriaan C. G. M. van Es, Julie Staals, Jeannette Hofmeijer, Jasper M. Martens, Geert J. Lycklama à Nijeholt, Sebastiaan F. de Bruijn, Lukas C. van Dijk, H. Bart van der Worp, Rob H. Lo, Ewoud J. van Dijk, Hieronymus D. Boogaarts, J. de Vries, Paul L. M. de Kort, Julia van Tuijl, Issam Boukrab, Jo P. Peluso, Puck Fransen, Jan S. P. van den Berg, Heleen M. den Hertog, Boudewijn A. A. M. van Hasselt, Leo A. M. Aerden, René J. Dallinga, Maarten Uyttenboogaart, Omid Eschgi, Reinoud P. H. Bokkers, Tobien H. C. M. L. Schreuder, Roel J. J. Heijboer, Koos Keizer, Rob A. R. Gons, Lonneke S. F. Yo, Emiel J. C. Sturm, Tomas Bulut, Paul J. A. M. Brouwers, Anouk D. Rozeman, Otto Elgersma, Michel J. M. Remmers, Thijs E. A. M. de Jong, Marieke E. S. Sprengers, Sjoerd F. M. Jenniskens, René van den Berg, Albert J. Yoo, Stefan D. Roosendaal, Bas F. W. van der Kallen, Ido R. van den Wijngaard, Joost Bot, Anton Meijer, Elyas Ghariq, Marc P. van Proosdij, G. Menno Krietemeijer, Wouter Dinkelaar, Auke P. A. AppelmanEuropean 2025 · DOI

TL;DR

Standard methods for assessing "backup" blood flow (collaterals) in stroke patients often look at the entire brain territory, which leads to overly optimistic results for blockages in smaller, distal vessels. This study introduces ODACS, an automated tool that restricts assessment only to the specific area starved of blood, revealing that collateral support in distal strokes is often much poorer than previously thought.

What Problem Does This Paper Address?

In an ischemic stroke, a blood clot blocks an artery in the brain. However, the brain has a "backup" system called collateral circulation—small vessels that can bypass the blockage to keep tissue alive. Patients with "good collaterals" generally have better outcomes and respond better to treatment.

Currently, both radiologists and automated software typically assess collaterals across the entire Middle Cerebral Artery (MCA) territory. This works well for large, proximal blockages (near the base of the brain) where the whole territory is affected. But for distal vessel occlusions (blockages further downstream), only a small portion of the MCA territory is actually at risk.

The problem? Traditional scoring counts the healthy blood vessels in the unaffected parts of the MCA territory as "good collaterals." This creates a false sense of security, overestimating the patient's actual backup flow and potentially leading to suboptimal treatment decisions.

Methods

The researchers developed the Occlusion-Downstream Area Collateral Score (ODACS). The workflow integrates two types of imaging:

  1. CT Perfusion (CTP): Used to identify the "hypoperfused area"—the specific region where blood flow is delayed (defined as Tmax > 6 seconds). This serves as a patient-specific map of the territory downstream of the clot.
  2. CT Angiography (CTA): An AI (a Residual 3D U-Net) segments all visible blood vessels.

The Calculation: ODACS is the ratio of the vessel volume within the affected downstream area compared to the same area in the healthy, opposite hemisphere.

The team compared ODACS against the standard "whole-MCA" approach (MCA-CS) and against visual scores from two expert neuroradiologists.

Datasets

The study used retrospective data from two major Dutch stroke trials:

  • MR CLEAN-NO IV
  • MR CLEAN Registry

From an initial pool of hundreds, the researchers identified 204 patients who had both high-quality CTA and CTP scans and confirmed occlusions in the proximal M1 (52%), distal M1 (32%), or M2 (16%) segments.

Key Insights and Ideas

  • The "Dilution" Effect: The study confirmed that the further downstream a blockage is, the more the traditional method overestimates collateral health. Because the traditional method looks at a fixed, large area, the "signal" of the actual collateral flow gets diluted by the "noise" of healthy vessels in unaffected areas.
  • Significant Reclassification: The most striking "aha moment" occurred with M2 occlusions. When switching from the traditional whole-territory method to ODACS, 58% of patients were reclassified from having "good" collaterals to "poor" collaterals.
  • Anatomy-Agnostic: Unlike some methods that require a computer to perfectly locate the clot (which is notoriously difficult for small vessels), ODACS uses the effect of the clot (the hypoperfusion map) to define the area of interest. This makes the tool more robust to variations in human vascular anatomy.

Key Results

  • Lower Scores Across the Board: ODACS yielded significantly lower collateral scores than the MCA-based approach for all locations ($p < 0.001$).
  • Bias Increases with Distality: The mean difference between the traditional score and ODACS was 0.15 for proximal M1, 0.23 for distal M1, and 0.34 for M2 occlusions.
  • Human-Level Performance: The agreement between the automated ODACS and human experts (weighted Cohen’s kappa $\kappa = 0.35$ to $0.51$) was comparable to the agreement between the two human experts themselves ($\kappa = 0.45$). This suggests the automated tool is as reliable as a trained radiologist but provides a more objective, quantitative value.

Limitations and Future Work

The authors acknowledge several hurdles before ODACS becomes a clinical standard:

  • CTP Dependency: The method requires CT Perfusion imaging. While common in major stroke centers, it isn't available everywhere and adds radiation/contrast dose.
  • The "Rapid Collateral" Paradox: If a patient has exceptionally fast collaterals, their Tmax might stay below 6 seconds. In these cases, ODACS might underestimate the downstream area because the tissue doesn't look "starved" enough on the map.
  • Clinical Outcomes: This study focused on the accuracy of the score. The next logical step is to prove that ODACS is a better predictor of actual patient recovery or treatment success than current methods.

In short, ODACS offers a "high-resolution" look at stroke impact, ensuring that when it comes to backup blood flow, we aren't counting chickens in the healthy parts of the brain that haven't even hatched.

Summary updated

Abstract

Question CT angiography-based collateral assessment in the MCA territory is inadequate to assess the collateral status in patients with distal vessel occlusions. Findings Our automated ODACS revealed lower collateral scores than traditional whole-territory assessment, especially in distal vessel occlusions. Clinical relevance The more precise evaluation of affected brain territories through automated occlusion-downstream area assessments prevents an overestimation of collateral status in distal occlusions, which could lead to improved patient selection and treatment decisions in acute stroke care.
BibTeX
@inproceedings{vries2025automated,
  author = {Lucas de Vries and M M Q Robbe and Ivo G.H. Jansen and Saeid Mojtahedi and Jan W. Hoving and Susanne G. H. Olthuis and Robrecht R. M. M. Knapen and Florentina M.E. Pinckaers and Manon Kappelhof and Ludo F.M. Beenen and Alida A. Postma and Robert J. van Oostenbrugge and Diederik W.J. Dippel and Efstratios Gavves and Bart J. Emmer and Charles B.L.M. Majoie and Wim H. van Zwam and Henk A. Marquering and on behalf of the MR CLEAN Registry Investigators and Diederik W. J. Dippel and Aad van der Lugt and Charles B. L. M. Majoie and Yvo B. W. E. M. Roos and Robert J. van Oostenbrugge and Wim H. van Zwam and Jelis Boiten and Jan Albert Vos and Study coordinators and Ivo G. H. Jansen and Maxim J. H. L. Mulder and Robert-Jan B. Goldhoorn and Kars C. J. Compagne and Manon Kappelhof and Josje Brouwer and Sanne J. den Hartog and Wouter H. Hinsenveld and Lotte van den Heuvel and Local principal investigators and Bob Roozenbeek and Pieter Jan van Doormaal and Bart J. Emmer and Jonathan M. Coutinho and Wouter J. Schonewille and Marieke J. H. Wermer and Marianne A. A. van Walderveen and Adriaan C. G. M. van Es and Julie Staals and Jeannette Hofmeijer and Jasper M. Martens and Geert J. Lycklama à Nijeholt and Sebastiaan F. de Bruijn and Lukas C. van Dijk and H. Bart van der Worp and Rob H. Lo and Ewoud J. van Dijk and Hieronymus D. Boogaarts and J. de Vries and Paul L. M. de Kort and Julia van Tuijl and Issam Boukrab and Jo P. Peluso and Puck Fransen and Jan S. P. van den Berg and Heleen M. den Hertog and Boudewijn A. A. M. van Hasselt and Leo A. M. Aerden and René J. Dallinga and Maarten Uyttenboogaart and Omid Eschgi and Reinoud P. H. Bokkers and Tobien H. C. M. L. Schreuder and Roel J. J. Heijboer and Koos Keizer and Rob A. R. Gons and Lonneke S. F. Yo and Emiel J. C. Sturm and Tomas Bulut and Paul J. A. M. Brouwers and Anouk D. Rozeman and Otto Elgersma and Michel J. M. Remmers and Thijs E. A. M. de Jong and Marieke E. S. Sprengers and Sjoerd F. M. Jenniskens and René van den Berg and Albert J. Yoo and Ludo F. M. Beenen and Alida A. Postma and Stefan D. Roosendaal and Bas F. W. van der Kallen and Ido R. van den Wijngaard and Joost Bot and Pieter-Jan van Doormaal and Anton Meijer and Elyas Ghariq and Marc P. van Proosdij and G. Menno Krietemeijer and Rob Lo and Wouter Dinkelaar and Auke P. A. Appelman},
  title = {{Automated collateral assessment restricted to the hypoperfused area for distal vessel occlusions in ischemic stroke}},
  booktitle = {European},
  year = {2025},
  doi = {10.1007/s00330-025-11442-2},
}
Figure