RPS 2311 - Stroke: improving diagnosis through imaging

Author Block: A. Bani Sadr, L. Mechtouff, M. Hermier, C. De Bourguignon, A. Martin, E. Tommasino, T-H. Cho, N. Nighoghossian, Y. Berthezene; Lyon/FR
Purpose: The aim of this study was to investigate the relationship between cerebral collateral status and blood-brain barrier (BBB) permeability at admission MRI in a cohort of acute ischemic stroke (AIS) patients treated with mechanical thrombectomy.
Methods or Background: The HIBISCUS-STROKE cohort is a single-centre observational study that prospectively included AIS patients with anterior circulation occlusion treated with mechanical thrombectomy. Admission dynamic-susceptibility MRI were post-processed to provide K2 maps with arrival-time correction as a marker of BBB permeability. After co-registration with ADC maps, 90th percentiles of K2 were extracted in the infarct core and normalised to contralateral white matter. Cerebral collateral status was assessed by the hypoperfusion intensity ratio (HIR). Good collaterals were defined by HIR<0.4. Multiple variable logistic regression analysis was adjusted to investigate factors associated with poor collaterals.
Results or Findings: One hundred and fifty seven patients were included (67.916.6 years, 52.8% male) with a median HIR of 0.43 (interquartile range (IQR): [0.32; 0.65]) and a median K2 of 1.67 (IQR: [0.45; 65.94]). Patients with poor collaterals (n=81, 51.6%) had worse NIHSS score (P=0.01), larger infarct core (P<0.0001) and higher K2 (median: 3.1, IQR: [0.5; 85.1] versus 1.2; IQR: [0.5; 3.4]; P=0.002). They were less likely to achieve successful recanalization (P=0.004) and had a higher rate of haemorrhagic transformation (P=0.02). On multiple variable analysis, poor collaterals were independently associated with larger infarct core volume (odds ratio (OR)=1.10; 95% confidence interval (CI): [1.06; 1.13]; P=0.002) and K2 (OR=1.46; 95% CI: [1.18; 1.84]; P=0.004).
Conclusion: Poor collaterals are independently associated with a larger infarct core and increased BBB permeability at admission MRI.
Limitations: This study was a retrospective analysis of a single-centre cohort, limiting its scope.
Funding for this study: This study was funded by the RHU MARVELOUS (ANR-16-RHUS-0009) of Université de Lyon, within the program “Investissements d'Avenir” operated by the French National Research Agency.
Has your study been approved by an ethics committee? Yes
Ethics committee - additional information: This study was ethically approved by the IRB number: 00009118.

Author Block: A. N. Khoruzhaya, K. M. Arzamasov, D. V. Burenchev, E. I. Kremneva, A. Vladzymyrskyy, Y. Vasilev; Moscow/RU
Purpose: The objective of this study was to evaluate the performance metrics of an artificial intelligence (AI) service aimed at diagnosing intracranial haemorrhage on head CT during a year-long multicentre clinical follow-up.
Methods or Background: As part of the Moscow Computer Vision Experiment, an AI service with known characteristics obtained from calibration testing (sensitivity - 0.89, specificity - 0.96, ROC AUC - 0.96) was connected on 28 April 2022 to the CT machines of 56 inpatient medical institutions. At the end of April 2023, it had analysed a total of 133,506 head CT. In order to assess accuracy rates, monthly clinical monitoring was conducted throughout the year, during which 1112 randomly selected head CT scans were independently assessed by three radiologists with more than 3 years' experience, assessing the actual presence of ICH (0/1) and the result provided by the AI service (trigger threshold 0.75). According to the results of physician judgement (GT), 440 CT studies (39%) contained signs of ICH.
Results or Findings: Full concordance of brain CT assessment by radiologists and AI service was achieved in 57% (633 CTs), partial concordance in 22% (249). The number of false positive responses by the AI service was 19% (212) and false negative responses were 1% (11). Thus, sensitivity was 97.5%, specificity was 68.4%, and ROC AUC was 0.94.
Conclusion: The AI service in a year-long multicentre clinical monitoring demonstrated a higher sensitivity than calibration testing, but lost significantly in specificity. This suggests good potential for the service to be used in acute care and to perform triage.
Limitations: The decrease in AI performance metrics should be taken into account for long-term use in practice and feedback should be given to developers for further retraining of AI services to reduce the number of FP results.
Funding for this study: This study was funded by the Program of the Moscow Healthcare Department “Scientific Support of the Capital's Healthcare” for 2023–2025: АААА-А21-121012290079-3.
Has your study been approved by an ethics committee? Yes
Ethics committee - additional information: The study was conducted in accordance with the Declaration of Helsinki, and approved by the Independent Ethics Committee of MRO RORR (protocol code 2/2020, the date of approval February 20, 2020). Clinical trial: NCT04489992.

Author Block: S. Steinmetz, M. A. Abello Mercado, A. Sanner, M. A. Brockmann, A. Othman; Mainz/DE
Purpose: The aim of this study was to examine the impact of deep-learning augmented contrast enhancement on diagnostic performance of poorly contrasted computed tomography angiography (CTA) in acute stroke.
Methods or Background: This retrospective single centre study included patients with suspected ischaemic stroke who underwent comprehensive CT imaging including cranial computed tomography (CCT), whole brain volume perfusion CT (VPCT) and computed tomography angiography (CTA) and had poorly contrasted CTA (defined as <350HU in the proximal MCA) between 01/2021 and 12/2022. 58/102 patients had vascular occlusion with correlate in perfusion. Datasets were reconstructed both conventional iterative methods (conventional CTA) and additionally using a pre-trained deep learning model allowing selective boosting of iodine-based contrast agents (enhanced CTA). Subjective image analysis was conducted by three readers, who rated general and vessel specific parameters on a 4-point Likert-scale. Furthermore, they evaluated both datasets for presence / absence of cerebral vessel occlusions. VPCT served as reference standard for calculating sensitivity and specificity.
Results or Findings: Enhanced CTA revealed significantly higher subjective image quality parameters (p<.001). Readers significantly improved sensitivity by enhanced CTA compared to conventional CTA (Reader 1 and 3: 55/58 [95%; 95% CI: 85.62% to 98.92%] vs. 48/58 [83%; 95% CI: 70.57% to 91.41%]; Reader 2: 53/58 [91%; 95% CI: 81.02% to 97.14%] vs. 46/58 [79%; 95% CI: 66.65% to 88.83%]. Reader 1 and 3 yielded no false positive findings on conventional CTA or enhanced CTA (specificity 44/44 [100%; 95% CI: 91.96% to 100%]), reader 2 yielded one false positive on enhanced CTA (specificity 43/44) [98%; 95% CI: 87.98% to 99.94%].
Conclusion: Deep-learning augmented contrast enhancements improves image quality and diagnostic performance in poorly contrasted CT angiographies. This could contribute to improve diagnostic certainty in acute stroke.
Limitations: Retrospective study design limits this study.
Funding for this study: No funding was obtained for this study.
Has your study been approved by an ethics committee? Yes
Ethics committee - additional information: This was a retrospective study and was approved by the Ethics committee, application number: 2022-16477.

Author Block: J. Schwarting, D. Trost, C. Albrecht, C. Zimmer, M. Wostrack, B. Meyer, J. H. W. Bodden, T. Boeckh-Behrens; Munich/DE
Purpose: Large-artery vasospasm (CVS) after aneurysmatic subarachnoid hemorrhage (aSAH) can reduce cerebral perfusion and cause severe neurological deficits. Delayed recognition of CVS risks the success of endovascular spasmolysis. An analysis of potential risk factors could be used to enable risk stratification for early therapeutic interventions. Therefore, the aim of this study was to confirm established and identify unknown risk factors for CVS at the time of aneurysm occlusion.
Methods or Background: In a single-center retrospective cohort study design, we compared 853 SAH patients (mean age 57; 67% female) between 01/2006 and 03/2020. Patients with and without CVS were compared based on demographic, clinical, and radiographic parameters at the time of aneurysm occlusion. CVS was defined as a blood flow velocity of >200 cm/s in large intracranial arteries or the occurrence of secondary neurological deficits without competing causes. Cohort differences were included as predictors in a multivariate analysis to address confounding. Logistic regression models were used to determine odds ratios (OR) for the presence of CVS for each predictor.
Results or Findings: 32% of SAH patients developed CVS. CVS was associated with age, female sex, aneurysm location, modified Fisher score, Barrow Neurological Institute (BNI) score, and surgical interventions in univariate analysis. Multivariable regression analysis identified multiple risk factors (Table 1). BNI score (OR: 1.33, p = 0.002), de-compressive craniectomy (OR: 1.93, p = 0.005), and aneurysm clipping (OR: 2.22, p < 0.001), were identified as only independent risk factors after correction for age, sex, aneurysm site and clinical severity.
Conclusion: Patients undergoing surgical interventions or patients with thick layers of subarachnoid blood should be monitored most intensively after aneurysm occlusion for early detection of CVS and endovascular intervention.
Limitations: The most important limitation was the retrospective, single-center study design.
Funding for this study: We did not receive any 3rd party funding for this study.
Has your study been approved by an ethics committee? Yes
Ethics committee - additional information: The study was approved by the local Research Ethics committee of the Technical University of Munich (186/20S).

Author Block: J. Schwarting¹, M. Froelich², J. S. Kirschke¹, J. H. W. Bodden¹, K. Dimitriadis¹, J. Ricke¹, C. Zimmer¹, T. Boeckh-Behrens¹, W. G. Kunz¹; ¹Munich/DE, ²Mannheim/DE
Purpose: Endovascular thrombectomy (EVT) is the standard of care for acute large vessel occlusion stroke. Recently, the ANGEL-ASPECT and SELECT 2 trials showed improved outcomes in patients with acute ischaemic Stroke presenting with large infarcts. The cost-effectiveness of EVT for this subpopulation of stroke patients has only been calculated with data from the previously published RESCUE-Japan LIMIT trial. It is therefore limited in its generalizability to an international population. The aim of this study was, therefore, to simulate patient-level costs, analyze the economic potential of EVT for patients with large ischaemic stroke from a public health payer perspective based on the recently published data, and identify significant determinants of cost-effectiveness.
Methods or Background: A Markov model was developed to compare outcome and cost parameters for patients treated with EVT or only with the best medical care based on the three recent prospective clinical trials ANGEL-ASPECT, SELECT2 and RESCUE-Japan LIMIT. Treatment outcomes were derived from the most recent literature. Deterministic and probabilistic sensitivity analyses addressed uncertainty. Willingness to pay was set at $100,000 per quality-adjusted life year (QALY).
Results or Findings: Endovascular treatment yielded an incremental gain of 1.32 QALYs per procedure with cost savings of $17,318 per patient. Lifetime costs were most sensitive to the costs of the endovascular procedure.
Conclusion: EVT is a cost-saving (i.e., dominant) strategy for patients presenting with large ischemic cores defined by inclusion criteria of the recently published ANGEL-ASPECT, SELECT2, and RESCUE-Japan LIMIT trials in comparison to best medical care in our simulation. Prospective data of individual patients need to be collected to validate these results.
Limitations: Simulation based on simplified linear pathways for diagnostics and therapy with limitations derived from the availability, quality, and validity of input variables.
Funding for this study: No third party funding was used for this study.
Has your study been approved by an ethics committee? Not applicable
Ethics committee - additional information: This was a Cost-effectiveness analysis without the use of individual patient data.

Author Block: Y. Zhou, Y. Zhai, K. Xue, Y. Yang, D. Zhang, J. Ni, F. Feng, Y. Wang; Beijing/CN
Purpose: The aim of this study was to compare the efficacy of 5.0T and 3.0T TOF MRA in evaluating collateral circulation and visualising abnormal vascular networks in patients diagnosed with moyamoya disease (MMD).
Methods or Background: Using both 5.0T and 3.0T MRI systems, a retrospective analysis was conducted on 30 symptomatic hemispheres from 21 ischemic-type MMD patients (10 males, age 7–39 years). A 0-5 grading system evaluated the visibility of moyamoya vessels across 5 regions (basal ganglion, anterior communicating artery, MCA–ICA tip, posterior communicating artery-PCA, basilar artery tip areas). Leptomeningeal anastomoses were evaluated with scores ranging from 0 to 6, comprising three parts of the collateral networks (pPCA→ ACA, the anterior temporal branch of the PCA anastomoses to the temporal branch of the MCA, pPCA→ MCA ). High-signal-intensity regions in basal ganglia was manually counted. Comparative analysis of moyamoya vessels and leptomeningeal system scores between 3.0 and 5.0T MRA was executed using the Wilcoxon matched-pair signed-rank test. A paired t-test was employed to juxtapose the number of high-signal-intensity regions between 3.0T and 5.0T MRA.
Results or Findings: The 5.0T TOF MRA demonstrated enhanced detection capabilities, showing a more significant number of moyamoya vessels and leptomeningeal anastomoses compared to 3.0T MRA. Furthermore, 5.0T MRA was more adept at detecting high-signal-intensity regions in the basal ganglia. Overall, 5.0T MRA provided clearer visualisation of the abnormal vascular networks associated with MMD.
Conclusion: The 5.0T TOF MRA presents a promising diagnostic tool for MMD, offering superior visualization of abnormal vascular networks and potentially aiding in more accurate disease assessment and prognosis.
Limitations: The parameters for 3.0T and 5.0T MRI couldn’t be perfectly matched. The small sample size might lead to false-negative findings.
Funding for this study: This study was funded by the Major International (Regional) Joint Research Project of National Natural Science Foundation of China (Grant Nos. 82020108018, 2020), the Beijing Natural Science Foundation (Grant Nos. Z210013, 2021) and National High Level Hospital Clinical Research Funding (2022-PUMCH-B-027).
Has your study been approved by an ethics committee? Yes
Ethics committee - additional information: The study was approved by the Ethics Committee at Peking Union Medical College Hospital (Reference number: K3147).

Author Block: S. Ingala¹, M. Perslev¹, S. Yamazaki Jensen¹, K. Dmitriev², J. Schlemper², C. Truwit², M. B. Bachmann Nielsen¹, M. Sofka², A. Pai¹; ¹Copenhagen/DK, ²Palo Alto, CA/US
Purpose: Cerebriu apollo brain is a software designed to detect acute cerebral infarcts on standard (1.5T and 3T) magnetic resonance images (MRIs). In this work, we aim to investigate the generalisation of the out-of-the-box software in detecting acute cerebral infarcts on ultra-low field MRI (ULF-MRI).
Methods or Background: Background: Apollo Brain was trained on 1300 standard MRI scans to detect acute cerebral infarcts, intracranial tumors, and intracranial haemorrhages. In this work, only the cerebral infarct detection feature is evaluated. Apollo brain accepts T2-FLAIR and DWI sequences with an option to add either T2* or susceptibility weighted images (SWI). No additional training or fine-tuning was performed for generalisation to ULF-MRI sequences. Methods: MRI images using Hyperfine Swoop scanner (software version > 8.4.0) collected between July 27, 2022, and June 13, 2023 were gathered. Inclusion criteria were a human non-contrast MRI imaging study indicating ischaemic stroke or intracranial hemorrhage (intraparenchymal or subarachnoid haemorrhage). We excluded MRI images displaying significant motion and other artifacts. All the scans were reviewed by an experienced radiologist. Then, Apollo brain performance was tested.
Results or Findings: A total of 30 subjects were included. Of these, 19 were diagnosed with acute cerebral infarct, including one with signs of haemorrhagic transformation of the infarct. Apollo brain software was able to detect the infarcts with a sensitivity of 94% and specificity of 83%. The software had one false negative , and two false positives.
Conclusion: Out-of-the-box Apollo Brain software for detection of acute intracerebral infarcts exhibited excellent performance with ULF-MRI images. These results demonstrate the possibility for an accelerated adoption of ULF-MRI in low resource settings for detecting acute cerebral infarcts coupled with Apollo brain software.
Limitations: We expect to improve the study by increasing the sample size.
Funding for this study: No information was provided by the submitter.
Has your study been approved by an ethics committee? Not applicable
Ethics committee - additional information: No information was provided by the submitter.

Author Block: S. Aludin, L-P. Schmill, P. Langguth, O. Jansen, N. Larsen, F. Wodarg, T. Klintz, S. Seehafer, A. Horr; Kiel/DE
Purpose: CT-angiography (CTA) and CT-perfusion (CTP) are part of acute stroke imaging. CTP thereby estimates the irreversibly damaged parenchyma, the infarct core (IC), and the potentially salvageable parenchyma, the penumbra (PEN), but requires additional contrast and radiation. Spectral-detector-CT (SDCT) enables spectral imaging like iodine-density imaging or virtual-monoenergetic imaging and the aim of this study was to assess its potential in differentiating IC and PEN using only monophasic CTA.
Methods or Background: Twenty patients with stroke in the media territory were analysed. Areas of IC and PEN derived from CTP-analysis as well as their healthy hemisphere’s counterparts were transferred to spectral maps of the CTA as regions of interest. The areas were compared to each other and between the different spectral maps (measurements of Hounsfield-Unit in monoenergetic images (MonoE) at 40 keV, 70 keV, and 120 keV, plus iodine-density (ID) and electron-density (ED) values). Unilateral absolute values and ratios formed to the healthy hemisphere’s values were evaluated. Visual infarct delineation was also rated in each map.
Results or Findings: IC and PEN could be distinguished from their healthy counterpart by absolute values (p<0.01). IC could be differentiated from PEN by absolute values (p<0.0001, except for ED) and by the ratio formed to the healthy counterpart (p<0.01). Thereby, discrimination of IC and PEN by ratio was good in MonoE40 (AUC=0.74, p<0.0001) and best in ID (AUC=0.92, p<0.0001). Visual delineation was best in ID and MonoE40.
Conclusion: SDCT-derived spectral maps from monophasic CTA, such as ID or MonoE40, allow detection and differentiation of IC and PEN. As a result, additional CTP may be removed from acute stroke CT protocols in the future, potentially saving time, contrast and radiation exposure.
Limitations: The limitations of the study are the low number of individuals, thus further studies are needed.
Funding for this study: No funding was sought for this study.
Has your study been approved by an ethics committee? Yes
Ethics committee - additional information: The study was ethically approved by the institutional review board and written informed consent was available from all participants.

Author Block: B. T. Catelani¹, A. Lopez Rueda², M. E. Santos Armentia³, G. B. Harvey¹, A. Pérez Fernandez³; ¹Rosario/AR, ²Barcelona/ES, ³Vigo/ES
Purpose: This study aimed to validate the novel combined indicator Hematoma Maturity Score using non-contrast computed tomography (NCCT) scans in patients with intracerebral haemorrhage (ICH) for predicting functional outcomes at discharge, in comparison to individual radiological NCCT signs.
Methods or Background: A retrospective analysis of patients with ICH who had undergone baseline CT scans was performed. The Black Hole Sign, Blend Sign, Island Sign, Swirl Sign, and the Hematoma Maturity Score were independently assessed by two radiologists who were blinded to clinical information. Patients were dichotomised based on dependency and mortality at discharge using the Modified Rankin Scale (mRS): no symptoms or no significant/mild disability (mRS 0–2); moderate/severe disability or mortality (mRS 3-6).
Results or Findings: Sixty-four patients with confirmed spontaneous ICH identified on NCCT were included. The Swirl Sign, Island Sign, and Hematoma Maturity Score exhibited statistically significant associations with the clinical outcomes of the patients (p < 0.01). There was almost perfect agreement between readers regarding the maturity score (kappa = 0.84), and substantial agreement for the other signs.
Conclusion: The novel concept of the Hematoma Maturity Score demonstrated the highest impact on clinical outcomes when
compared to other evaluated radiological signs, with almost perfect agreement between readers.
Limitations: No limitations have been identified for this study.
Funding for this study: No funding was received for this study.
Has your study been approved by an ethics committee? Yes
Ethics committee - additional information: This study has been approved by the local hospital Ethics Committee.

Author Block: M. A. Abello Mercado, S. Steinmetz, A. Sanner, A. Kronfeld, M. A. Brockmann, A. Othman; Mainz/DE
Purpose: The aim of this study was to evaluate the effects of deep-learning image reconstruction on image quality and diagnostic confidence of ultra-high-resolution computed tomography (UHRCT).
Methods or Background: In this single-center study, 100 consecutive patients with acute neurological symptoms underwent CT imaging including cranial computed tomography (CCT) and computed tomography angiography (CTA) using an ultra-high resolution CT scanner. CTA images were reconstructed with normal resolution mode and ultra-high resolution mode using iterative reconstruction. A deep-learning reconstruction algorithm (advanced intelligent clear-IQ engine, AiCE); specifically trained for ultra-high resolution CT-angiography of the brain was utilized to generate a further UHR-CTA datasets (DL-UHR-CTA, matrix 1024 x 1024, slice thickness 0.25 mm). Image quality for all three reconstructions was evaluated visually by two blinded radiologists using a 4-point Likert-scale. Therefore, general (overall image quality, contrast in general, artifacts, diagnostic confidence and image noise) and vessel specific (assessability of proximal, intermediate and subcortical vessels as well as perforators) criteria were assessed. The quantitative features including slope, signal-to-noise ratio (SNR), contrast-to-noise ratio (CNR), noise, entropy and co-occurrence matrix (COOC) were examined and compared using an in-house tool.
Results or Findings: Qualitative analysis revealed highest scores for DL-UHR-CTA, followed by UHR-CTA and NR-CTA, whereas DL-UHR-CTA yielded excellent results for all qualitative parameters and was significantly superior to UHR-CTA and NR-CTA (all p<0.001). The quantitative analysis was in line with the qualitative findings with significantly superior results for DL-UHR-CTA (slope: p<.01, SNR/CNR: p=0.004, entropy p<.01, COOC: p<.01).
Conclusion: Deep-Learning image reconstruction significantly improves image quality of ultra-high resolution neurovascular CT-angiography allowing for higher diagnostic confidence, potentially improving the detection of subtile but oftentimes-significant pathologies.
Limitations: Deep-learning image reconstruction improves the quality of UHR-CTA images, leading to higher diagnostic confidence and potentially aiding in the detection of subtile but clinically significant pathologies.
Funding for this study: No funding was received for this study.
Has your study been approved by an ethics committee? Yes
Ethics committee - additional information: Approval was received from the Ethics committee, application number: 2021-15948:1-retrospektiv.

Author Block: B. Sui, B. B. XiaoYan; Beijing/CN
Purpose: Haemodynamic factors have been proved to play essential roles in the formation and development of vascular diseases. This study aimed to investigate the blood flow distribution in the different types of circle of Willis and the changes of blood flow at different phases of the cardiac cycle using 4D Ffow imaging.
Methods or Background: Four-dimensional flow magnetic resonance imaging (4D flow MRI) was performed in 30 healthy subjects (mean 27.5±4.1 years; range 24-38 years old). CoW was classified into five types according to the vessel anatomical structures. The cerebral blood flow distribution at different arteries and the total cerebral blood flow (tCBF) were analyzed and compared between different types of CoWs. The changes of cerebral blood flow during different phases of the cardiac cycle were observed and compared between different types of CoWs.
Results or Findings: Twelve subjects (40%) were found with complete CoW, and eighteen (60%) with incomplete CoW. No significant difference was found in tCBF between different types of CoWs (P=0.787). In subjects with a fetal-type PCA (Type V), the average flow rates (FRavg) at contralateral ICA was higher than that of the ipsilateral ICA (P=0.003). During the peak systolic and end diastole, compared with Type I, in subjects with unilateral PCoA opening (Type IV), the Δflow rates (ΔFR) at bilateral MCA were significantly higher than that of Type I (P=0.012 and P=0.006, respectively); in subjects with unilateral fetal-type PCA (Type V), the ΔFR at contralateral MCA were significantly higher than that of Type I (P=0.007).
Conclusion: Blood flow distribution exists a difference in different types of CoWs during the different phases of the cardiac cycle, especially in type IV and type V at peak systolic.
Limitations: small samples for type V and VI.
Funding for this study: This study was funded by the Beijing Municipal Natural Science Foundation (NO. 7162056, 7212028).
Has your study been approved by an ethics committee? Yes
Ethics committee - additional information: This study was approved by the local ethics committee of Beijing Tiantan Hospital, and all volunteers gave written informed consent.