Advanced imaging techniques in cerebrovascular disease

Abstract

This dissertation aims to investigate how advanced imaging techniques can improve diagnosis, prognostication and treatment delivery in patients with cerebrovascular disease. Six original publications are presented in which perfusion imaging and new software algorithms are used to address clinical need in two specific cerebrovascular disease entities: acute ischemic stroke (AIS) and dural arteriovenous fistulas (DAVFs).

AIS patients with anterior circulation large vessel occlusions (LVO) can be successfully treated with mechanical thrombectomy. Patient triage to thrombectomy requires identification of an LVO, the target of therapy, as well as the likelihood of benefit from reperfusion and the potential risk of complication. This assessment must be accurate, fast and efficient since stroke is a time-critical emergency. This poses a challenge for stroke centres because an increasing number of patients are being screened since extension of the thrombectomy window to 24 hours. Additionally, non-tertiary level hospitals are now required to perform acute stroke imaging despite lacking around-the-clock neuroradiology expertise. The first four publications explore how perfusion imaging and automated software algorithms can be used to expedite triage while maintaining high diagnostic accuracy for identifying patients who are likely to benefit from treatment.

The first study introduces and describes a new fully-automated deterministic software algorithm for detecting LVOs on computed tomography (CT) angiography, then validates it in a large cohort of 926 AIS patients that was enriched for LVOs. The algorithm had high sensitivity (97%) and moderate specificity (74%) for detecting LVOs. The second study then applied this algorithm to a consecutive cohort of 477 “code stroke” patients presenting to a large regional hospital, with the aim of field testing it in the real world clinical setting where automated LVO detection tools are most likely to be used. The high sensitivity (94%) and negative predictive value (98%), combined with fast processing times, suggest that it can be used as a screening tool to assist radiologists and expedite diagnosis of LVOs.

Patients with LVOs who have large infarct cores are unlikely to benefit from thrombectomy and have an increased risk of complication. CT perfusion (CTP) with fully automated post-processing is widely used to exclude patients with large infarct cores from treatment. Previous studies that validated CTP for this purpose had some key limitations. These were addressed in the third study, which sought to determine whether automated estimation of the infarct core on perfusion, based on reduced relative cerebral blood flow (rCBF), is sufficiently accurate for patient triage to thrombectomy. A novel approach was adopted, allowing almost perfectly temporally and volumetrically matched diffusion and perfusion data to be compared in a cohort of 119 prospectively enrolled patients. 94% of patients were correctly triaged using reduced rCBF, suggesting that fully automated perfusion-based measurement of the infarct core can be used for individual patient triage.

Despite widespread use of CT, magnetic resonance imaging (MRI) remains the first-line modality for stroke patients in Europe and Asia. One of the most time-consuming sequences is T2*-weighted gradient recalled echo (T2*GRE), which is used to detect haemorrhage that contraindicates reperfusion therapies. Dynamic susceptibility contrast perfusion weighted imaging (DSC-PWI), which is used primarily to delineate the ischemic penumbra, is also sensitive to haemorrhage. The agreement between DSC-PWI and T2*GRE for detection of haemorrhage was assessed in the fourth study on 393 MRI scans from a cohort of 221 AIS patients. Almost perfect agreement (k > 0.90) was shown for detection of acute haemorrhage. This suggests that DSC-PWI is sufficient for haemorrhage screening when it is included in the AIS MRI protocol.

Arterial spin label (ASL) is an entirely non-invasive MR perfusion technique that is an alternative to DSC-PWI. While its use in AIS is limited, it has been serendipitously discovered that ASL signal in venous structures indicates the presence of shunting. Intracranial DAVFs are a type of shunting lesion that can be difficult to detect on structural imaging.

The diagnostic performance and added value of ASL for detection of DAVFs was assessed in the fifth study, in a cohort of 156 patients. Venous ASL signal had a high sensitivity (94%), negative predictive value (98%) and specificity (88%) for the presence a DAVF. Including ASL in the MRI protocol improved diagnostic confidence and performance. The sixth study assessed the accuracy of ASL for identifying cortical vein drainage, the main risk factor for haemorrhage in DAVF patients. In a cohort of 34 patients, ASL was found to have a sensitivity of 91% and specificity of 96% for the presence of cortical vein drainage. These findings suggest that an MRI protocol augmented with ASL can be used to non-invasively screen for DAVFs and differentiate between high-risk fistulas requiring treatment and low-risk lesions that can be managed with observation.

To conclude, a summary of the findings is presented along with the impact of the work, its limitations and future directions for research.