Designing a novel pre-hospital treatment decision model based on transcranial ultrasound and clinical assessment for patients with suspected acute stroke: exploratory study

Background and objectives Earlier initiation of reperfusion therapy for patients with acute ischaemic stroke is associated with better outcomes. The current exploratory study presents a planned interim analysis aiming to validate a theoretical treatment decision model to select subjects who could benet from early thrombolysis and/or direct transfer for mechanical thrombectomy. Methods Consecutive subjects suspected of having suffered a stroke within the last 72 hours were recruited. Transcranial grey-scale and Doppler ultrasonography was performed within 24 hours of brain computed tomography. The National Institutes of Health Stroke score (NIHSS), modied Rankin Score and medical history were collected retrospectively. Hospital discharge diagnosis was used as the gold standard. Comparative analyses were performed to identify variables that could assist in designing the treatment decision model. Sensitivity, specicity, positive (PPV) and negative predictive values (NPV) were calculated. Results Among a total number of 50 recruited patients, 27 had a nal diagnosis of ischaemic stroke, including 10 cases with large vessel occlusion (LVO). Stroke patients were more likely to be older (p=0.03) and to have NIHSS total score ≥ 9 (p=0.021). NIHSS ≥ 4 (p=0.032), atrial brillation (p=0.033) and cortical signs (p=0.005) were signicant indicators of LVO. The proposed model based on grey-scale/Doppler ndings and clinical assessment showed 77% sensitivity and 92% specicity (77% PPV, 92% NPV) for making a decision to initiate reperfusion therapy. Ultrasound ndings improved the performance of a model based on clinical data alone (50% sensitivity, 89% specicity, 64% PPV, 82% NPV). Subjects with a stroke mimicking condition were more likely to have been given reperfusion therapy inappropriately using this model, but no subjects with haemorrhage would have received intravenous thrombolysis. Insucient acoustic window was seen in approximately 40% of included participants. Conclusions The proposed treatment decision model might potentially be used for selecting subjects with ischaemic stroke who would benet from early initiation of reperfusion

signs (p=0.005) were signi cant indicators of LVO. The proposed model based on grey-scale/Doppler ndings and clinical assessment showed 77% sensitivity and 92% speci city (77% PPV, 92% NPV) for making a decision to initiate reperfusion therapy. Ultrasound ndings improved the performance of a model based on clinical data alone (50% sensitivity, 89% speci city, 64% PPV, 82% NPV). Subjects with a stroke mimicking condition were more likely to have been given reperfusion therapy inappropriately using this model, but no subjects with haemorrhage would have received intravenous thrombolysis. Insu cient acoustic window was seen in approximately 40% of included participants. Conclusions The proposed treatment decision model might potentially be used for selecting subjects with ischaemic stroke who would bene t from early initiation of reperfusion therapy or immediate transfer for endovascular intervention. A larger sample is required to establish performance of the model with an acceptable degree of precision.

Background
Standard treatment for acute ischaemic stroke patients include intravenous (IV) administration of a thrombolytic drug. However severe strokes with a mortality rate reaching 80% [1] are often caused by a large vessel occlusion (LVO), and can be optimally managed by mechanical thrombectomy preceded by IV thrombolysis [2]. While thrombolysis is offered in many regional hospitals, mechanical thrombectomy can only be performed in specialised endovascular centres with neurointerventional facilities.
Administration of each treatment is highly time sensitive. Outcomes are better with shorter delays: the probability of a good clinical outcome decreases by about 11% with every 30 minute delay from onset to successful reperfusion [3]. However, only 2-10% of stroke patients worldwide receive IV thrombolysis [4,5] due to various reasons, including late arrival to the hospital, variability in the patient selection criteria, etc [6]. As demonstrated by Ebinger et al (2014), thrombolysis given in the ambulance signi cantly improved treatment rates with time reduction by at least 15 minutes [7].
For patients with suspected LVO, immediate thrombolytic therapy with direct transfer to an endovascular centre ("drip and ship") can potentially lead to more positive outcomes [8]. This model is particularly relevant to remote and rural areas with long travel times. Designing a triage system for patients with acute ischaemic stroke who would be considered for early reperfusion therapy and/or endovascular intervention has attracted much attention in the last decade.
The National Institute of Health Stroke Scale (NIHSS) cutoff ³11 and Rapid Arterial Occlusion Evaluation Scale (RACE) cutoff ³5 demonstrated the best diagnostic performance however their false-negative rates remain higher than 25% [9]. Thus, at least 20% of subjects with LVO would be incorrectly transferred to a local hospital with no neurointerventional facilities available, causing unnecessarily poorer outcomes [9]. Importantly, diagnosis of acute intracranial haemorrhage and/or stroke mimicking condition is often an exclusion criterion in these studies, and therefore their ndings are unlikely to be capable of translation into the pre-hospital setting [10].
As suggested by Beume et al (2018), cortical symptoms such as aphasia and/or neglect are indicative of LVO with 91% sensitivity and 70% speci city when compared to 85% sensitivity and 53% speci city of motor symptoms alone [10]. Similarly, the use of clinical assessment scale thresholds, such as NIHSS ³10 [9,11,12], the Field Assessment Stroke Triage for Emergency Destination ³4 [12][13][14][15] and RACE ³5 [12,14,15] which were designed to evaluate cortical symptoms as well as motor symptoms, showed higher sensitivity and speci city when compared to other tools (Antipova et al., 2019, manuscript in preparation). Furthermore, telestroke assessments while in transit supported by hospital-based specialists could improve triage of patients who would bene t from reperfusion therapy [16].
It is crucial to exclude intracranial haemorrhage as one of the absolute contraindications for recanalization therapies. However, scoring tools, such as the Siriraj score, have insu cient precision to predict the presence of haemorrhage reliably [17]. Therefore, brain imaging is currently essential to differentiate haemorrhagic from ischaemic events and to exclude stroke mimics such as tumours [17,18] that would not bene t from reperfusion therapy. However, this is not widely incorporated into the prehospital setting, and is usually performed once a patient arrives in hospital, which may result in treatment delay.
Portable transcranial ultrasound has emerged as a potential method for rapidly assessing the intracranial circulation [19]. Transcranial Doppler (TCD) ultrasound ndings, such as absent or diminished blood ow signal in a major intracranial artery in the Circle of Willis were shown to be suggestive of LVO [20][21][22].
To the best of our knowledge, to date no studies have been published concerning the performance of a combination of clinical assessment and brain imaging with transcranial ultrasonography for selecting subjects who would bene t from early initiation of reperfusion therapy and/or direct transfer for mechanical thrombectomy. We hypothesised that a novel theoretical treatment decision model can be designed to serve this purpose.
Research questions were as follows: 1. What transcranial grey-scale/TCD ultrasound and clinical assessment ndings should contribute to a triage system to correctly select subjects with acute ischaemic stroke who could bene t from early intervention, such as IV thrombolysis and/or mechanical thrombectomy?
2. In what proportion of patients could a correct treatment decision potentially be made using the proposed model?
3. Can the diagnostic accuracy of a treatment decision based exclusively on clinical assessment be improved by adding transcranial grey-scale/TCD ultrasound?
4. In what proportion of suspected stroke patients could transcranial grey-scale/TCD ultrasound not be used due to technical limitations, such as insu cient acoustic window?
As part of a larger study to evaluate the diagnostic precision of a treatment decision tool based on clinical data and transcranial grey-scale/TCD ultrasound, we planned an exploratory evaluation of data from the rst 50 eligible patients.

Subjects
An exploratory study was designed as a part of a larger project with the aim of determining which subjects would bene t from early IV thrombolysis and/or direct transfer to an endovascular centre on the basis of transcranial grey-scale/TCD ultrasound and clinical assessment ndings.
Data collection was performed from 28 July 2017 to 16 November 2018. Patients suspected of having suffered an acute stroke were recruited into the study upon their admission to Raigmore Hospital (Inverness, UK). Research team members identi ed eligible candidates on the basis of the following criteria: Inclusion criteria: -Suspected stroke patients admitted to Raigmore Hospital for whom standard brain imaging was requested -Patients with suspected subarachnoid haemorrhage for whom standard brain imaging was requested -Adults (16 years and over) -Onset of stroke symptoms not more than 72 hours previously -Informed consent from the participant, or if they were unable to give meaningful consent, proxy consent from a relative or legal representative in compliance with the Adults with Incapacity Act (Scotland, 2000).
Exclusion criterion: -Patients for whom no standard brain imaging was requested.

Data collection
Baseline demographic and clinical information, including electrocardiography ndings, past medical history, symptoms at onset, including cortical symptoms such as neglect, dysphasia or aphasia, and eye deviation, the NIHSS and modi ed Rankin Scale (mRS) scores were collected retrospectively. The NIHSS total score was calculated using standard procedure [23].

Brain imaging
According to the National Institute for Health and Care Excellence guideline NG128 (May 2019), brain imaging with a non-enhanced CT is recommended for the early assessment of people with suspected acute stroke. CT was performed as per standard hospital protocol upon patient presentation using a General Electric 750HD G64-slice Discovery or General Electric 64-slice light speed scanner. Diagnosis of acute ischaemic stroke due to LVO was made on the basis of ASPECTS -Alberta stroke programme early CT score.
Transcranial grey-scale/TCD ultrasound was performed within 24 hours of the CT scan using a SonoSite M-Turbo® Point-of-Care ultrasound machine; Philips Sparq or Philips CX50 ultrasound. Since there is no dedicated probe for transcranial scanning available, a 1-5 MHz low-frequency phased array transducer with small footprint was used to allow penetration through the skull. All transcranial grey-scale/TCD scans were obtained by the same medically trained research team member after completing brief training that would re ect the type and volume of training that prehospital clinicians might receive. The researcher performing transcranial ultrasound scanning was blinded to the CT ndings but presenting complaints and/or examination ndings were available in all cases. Index test (transcranial grey-scale/TCD ultrasound) results were not available to the assessors of the reference standard (CT), although clinical information was provided.
Participants were asked to remain in a supine or sitting position throughout the ultrasound scanning.
Scanning was done through temporal bone windows on either side of the head using grey-scale mode to visualise brain structures and TCD ultrasound for assessment of blood ow in the major intracranial arteries. The operator followed a standard protocol to explore for: 1. Brain structures in the coronal and transverse planes, aiming to visualise any acute intracranial haemorrhage. Acute intracerebral haemorrhage can be detected on ultrasound images as homogenous hyperechogenic structures which are well distinguished from surrounding tissues. Spontaneous subarachnoid haemorrhage can be detected as hyperechogenic signal in the basal cisterns [24].
2. Midline shift by taking three consecutive measurements of the distance between the ultrasound probe and the centre of the third ventricle measured along a line perpendicular to the walls of the third ventricle from the ipsilateral side. Midline shift was calculated using the equation provided elsewhere [25].

Visualisation of the blood ow in the major intracranial arteries forming the Circle of Willis bilaterally
with TCD sonography -middle cerebral artery (MCA), anterior cerebral artery (ACA) and posterior cerebral artery (PCA). MCA was identi ed as orthograde blood ow at insonation depth of 40-65mm and 30-40mm for proximal (M1) and distal (M2) segment, respectively. ACA was detected as retrograde ow at the 60-75mm depth of insonation. PCA was visualised as orthograde (P1 segment) or retrograde (P2 segment) at 55-75mm depth [26].
LVO was de ned as occlusion of one of the major intracranial arteries forming the Circle of Willis, namely M1, M2 segments of the MCA, ACA and PCA. TCD ultrasound diagnosis of the MCA mainstem occlusion was considered when the ow was absent or minimal, blunted, or damped throughout the MCA with increased ow velocity in the anterior or posterior cerebral arteries in comparison to contralateral side ( ow diversion) [20][21][22][27][28][29]. ACA and PCA occlusion were de ned as the absence of the ow or the presence of minimal, blunted, or dampened ow signals throughout these vessels [22,30].
An insu cient temporal bone acoustic window was de ned as the inability to visualise the heart-shaped cerebral peduncles as a hypoechoic structure on grey-scale transcranial ultrasound images and/or the major intracranial arteries.
The patient's nal discharge diagnosis was referred to as the "gold" standard.

Statistical analysis
All patients were divided into four groups for the purposes of the comparative analysis as follows: 1acute ischaemic stroke patients due to LVO; 2 -acute ischaemic stroke patients with no evidence of LVO; 3 -acute haemorrhagic stroke; 4 -stroke mimicking conditions and TIA. The following transcranial greyscale/TCD ultrasound and clinical assessment ndings were incorporated into comparative analysis with the aim to identify difference in the outcome between the four groups: presence of haemorrhagic focus on transcranial grey-scale ultrasonography, midline shift (cm), signs of LVO on TCD ultrasound; plus the following clinical data: age (years), presence of cortical signs, NIHSS total score, mRS total score, history or ECG ndings of atrial brillation, systolic blood pressure levels (mmHg), current anticoagulant or antiplatelet therapy. Statistical analysis was performed using IBM SPSS Statistics 25 software. Distribution shape of the numerical data was analysed to determine the appropriate statistical test for two groups of independent data. Unpaired t-tests were chosen for normally distributed data and Mann-Whitney U tests were employed to analyse skewed data sets with mean and median summary measures reported, respectively. Fisher's exact test was used for analysis of the two groups of independent categorical data. Statistically signi cant differences between the groups were documented if the probability value (p) was equal or below 0.05 Designing the treatment decision model A tentative treatment decision model was designed on the basis of the following data: 1. Statistically signi cant (p≤05) transcranial grey-scale/TCD ultrasound and clinical assessment ndings based on the results of the comparative analysis 2. Evidence of any contraindications for IV thrombolysis [31] 3. Clinically important parameters based on previously published data, such as NIHSS³10 [9,11,12] and presence of cortical symptoms [10] for selecting subjects with LVO.
The aim was to determine whether the proposed pathway could be used to select subjects that: (1) would potentially bene t from early IV thrombolysis (acute ischaemic stroke group with no signs of LVO); (2) would bene t from a direct transfer to an endovascular centre for mechanical thrombectomy preceded by early IV thrombolysis (acute ischaemic stroke due to LVO group); (3) would not bene t from reperfusion therapy (patients with a TIA, intracranial haemorrhage, and stroke mimics).

Results
A total of 50 participants aged 33-92 years old (mean ± standard deviation, 70.7±13) meeting the inclusion criteria were selected consecutively upon their admission to the hospital and recruited into the current study ( Figure 1). Figure 1. Flow-diagram. The recruitment of participants into the current study and nal analysis.
Among the included participants, 21 were males and 29 were females.
Characteristics of participants are presented in Additional le 1.
All participants were of white ethnic origin. Ischaemic stroke was diagnosed in 27 cases including ten participants with LVO (LVO in the anterior circulation n=8, LVO in the posterior circulation n=2). Acute intracranial haemorrhage was diagnosed in 10 cases, including 1 participant with subarachnoid haemorrhage. Eight patients had a nal diagnosis of a stroke mimicking condition, such as multiple sclerosis, benign paroxysmal positional vertigo, functional neurological disorder, neuropathic pain, Bell's palsy, migraine, hypertension induced paraesthesia, and angioma. Five people had a nal diagnosis of a TIA. NIHSS total score varied from 0 to 28 points (mean ± standard deviation, 5.72±7). All included participants underwent CT brain imaging.

Comparative analysis
Statistically signi cant outcomes of the comparative analysis are presented in Additional le 2.
Patients with a con rmed diagnosis of acute stroke were older (median age 74 vs. 58 years, p=0.03), had a higher total NIHSS score (median score 4 vs. 2, p=0.032), and were more likely to have cortical signs (p=0.039) compared with subjects with a stroke mimicking condition. NIHSS cut-off ³9 was a signi cant predictor of a stroke diagnosis (p=0.021).
Systolic blood pressure was signi cantly higher in patients with intracranial haemorrhage (median level 171 vs.153 mmHg, p=0.135). It was also shown that normal blood ow on TCD ultrasound (p=0.009) ( Figure 2) and presence of haemorrhagic foci on grey-scale (p=0.001) ( Figure 3) were signi cantly associated with the diagnosis of intracranial haemorrhage.  Patients with LVO were more likely to have a high NIHSS score (median score 14 vs. 2, p<0.001) with optimal cut-off of 4 (p=0.032). Presence of cortical signs (p=0.005) and atrial brillation (from history or ECG) (p=0.033) were also signi cant predictors of LVO.

Treatment decision pathway
On the basis of the results of the comparative analysis, indications for IV thrombolysis [31] and previously published data [9,11,12], a tentative treatment decision model was designed (Figure 4). On the basis of the proposed model, 7/10 (70%) of the sample's patients with LVO would be identi ed and correctly selected for direct transfer to endovascular centre. All (16/16, 100%) cases of ischaemic stroke not associated with LVO would have been correctly assigned to early thrombolysis. In 20/23 (87%) cases of patients who would not bene t from IV thrombolysis or endovascular treatment (intracranial haemorrhage, TIA and stroke mimicking conditions), it was correctly decided not to initiate reperfusion therapy.
One case of LVO in the posterior circulation would incorrectly not have received reperfusion therapy due to a misleading mild neurological de cit. In the remaining two cases of LVO that would not have received treatment the decision could not be made due to insu cient temporal bone acoustic window and therefore the inability to exclude intracranial haemorrhage on transcranial grey-scale/TCD ultrasound.
2/10 (20%) subjects with intracranial haemorrhage had an insu cient temporal bone acoustic window, which still led to a treatment decision of not initiating reperfusion therapy, which would have been the correct decision. In 3/8 (38%) patients who were diagnosed with a stroke mimicking condition the decision would have been made incorrectly to initiate reperfusion therapy, mainly because of the high NIHSS total score and false signs of absent blood ow in MCA on TCD ultrasound. The treatment decision for each case is presented in Additional le 2.
Overall, an insu cient temporal bone acoustic window was documented in 20/50 (40%) cases. TP, TN, FP, FN rates of the proposed model are presented in Table 1. The proposed model showed 77% sensitivity and 92% speci city with 77% PPV, 92% NPV.
Performance of the model based only on clinical assessment ndings, without transcranial greyscale/TCD ultrasound examination, is presented in Table 2. As demonstrated in Table 2, performance of the model based on clinical ndings alone was poorer than the proposed model with a combination of transcranial grey-scale/TCD ultrasound and clinical assessment. It showed a lower sensitivity of 50%, speci city of 89%, PPV of 64% and NPV of 82%.

Discussion
Early recognition and initiation of reperfusion treatment is associated with more favourable outcome [32].
However, the diagnostic accuracy of currently available tools does not allow reliable selection of patients who would bene t from a direct transfer to regional endovascular centres for mechanical thrombectomy, potentially leading to inappropriate and costly transfers [9].
Previous research has demonstrated sensitivity of TCD ultrasound ranging from 68% to 100% and speci city 78% to 99% for detecting acute vessel occlusion and stenosis [20][21][22]30,[33][34][35]. It is a simple, non-invasive and affordable diagnostic tool that takes <15 minutes for a complete assessment of cerebral vessels [27,36] and can be performed in space-restricted environments, such as ambulances [16,37]. As demonstrated previously [16,38,39], transcranial grey-scale/TCD ultrasound images can be transferred from remotely supported ambulances in remote and rural areas for hospital-based expert interpretation, which reduces the requirement for skill acquisition and maintenance on the part of the onscene assessor.
In the current exploratory study, we propose that the designed novel treatment decision tool based on transcranial grey-scale/TCD ultrasound together with clinical assessment ndings could potentially improve the selection of patients with acute ischaemic stroke who would bene t from early initiation of reperfusion therapy, though further con rmatory work is required. It was demonstrated that the presence of cortical signs was signi cantly associated with the diagnosis of a stroke in general, and ischaemic stroke in particular. In agreement with previously published data, higher NIHSS total score [11,12], presence of cortical signs [10] and atrial brillation were signi cant predictors of a diagnosis of LVO.
Additional features, such as more advanced age, can point towards a more likely diagnosis of acute stroke rather than a stroke mimicking condition.
Results of the diagnostic accuracy analysis of the proposed model show that a correct treatment decision would have been made in 70% of LVO cases and 100% of patients with other types of acute ischaemic stroke that were not caused by LVO. Subjects with a LVO in the posterior circulation are likely to be missed due to the poorer sensitivity of TCD ultrasound and clinical assessment tools in this group [22,35,40].
A haemorrhagic focus was detected on transcranial grey-scale/TCD ultrasound only in 50% of con rmed cases, mainly in central structures, such as basal ganglia. In agreement with previously reported studies [41], haemorrhages were likely to be missed if located in the cortical areas, posterior fossa, or in other cerebral structures that cannot be reached by ultrasound through the transtemporal window (for example, the thalamic lesions in cases 15 and 47, Additional le 1). Clinical assessment can therefore provide additional valuable information, for example, history of anticoagulant therapy, active brain tumour, and high systolic blood pressure [42] that would point towards suspected intracranial haemorrhage even in the absence of haemorrhagic foci on transcranial grey-scale ultrasound. A decision not to initiate reperfusion therapy would have been made based on transcranial grey-scale/TCD ultrasound and clinical assessment ndings in all cases of intracranial haemorrhage, including those with insu cient acoustic window.
By making an effort to shorten the time interval from stroke symptom onset to delivery of IV thrombolysis, it is highly important to avoid providing this therapy to patients with conditions other than stroke. As demonstrated by Tsivgoulis et al (2015), about 15% of all patients treated with IV thrombolysis actually have stroke mimics. An inappropriate administration of reperfusion therapy could potentially lead to serious complications, such as intracranial and extracranial bleeding [43] but the risks are relatively low except when haemorrhage is present. In the present study, the highest rate of incorrect treatment decisions was in the stroke mimic group (38%). These were the subjects with a functional neurological disorder, neuropathic pain with associated weakness, and Bell's palsy. It should be acknowledged that patients with stroke mimicking conditions might be given inappropriate treatment even in the settings of a tertiary care stroke centre [43]. A correct treatment decision was made in all cases of TIA.
Overall performance of the proposed model based on a combination of transcranial grey-scale/TCD ultrasound and clinical assessment was better when compared with a similar system based on clinical ndings alone. Interestingly, in one case a stroke mimicking condition was suspected on the basis of clinical assessment and transcranial grey-scale ndings, in contrast with the CT ndings which reported a bifrontal subarachnoid haemorrhage. A review of the original CT scan revised the nal diagnosis to a probable bifrontal angioma.
Further work is needed to re ne the model to decrease false positive rates, particularly for stroke mimicking cases, and false negative rates to ensure LVO in posterior circulation are accurately detected.
As demonstrated in the theoretical modelling study by Holodinsky et al (2018), in remote and rural areas with increased transport times, patients with ischaemic stroke due to LVO might bene t from early IV thrombolysis followed by streamlined transfer to an endovascular centre ("drip and ship") [8]. Furthermore, the minimum 15 minute decrease in time to treatment can be achieved by offering thrombolysis in specialised stroke ambulances without increasing the risk of intracerebral haemorrhage or mortality [7]. The current exploratory work is part of a larger ongoing research study aimed at validating a novel triage system to select subjects who would bene t from early reperfusion therapy and direct transfer to a specialised centre. Adequate sensitivity and speci city in such a triage tool could potentially justify a randomised controlled trial of the "drip and ship" model with remotely supported specialist interpretation of the data and decision making.

Limitations
The current study had several limitations. Firstly, the data in this exploratory study were collected from a sample size of only 50 participants, ten of which were con rmed as having LVO. Some patients with very mild or resolved symptoms might have been missed out from the present study due to early discharge.
Equally, patients with a severe stroke may have been not included in the current study due to early mortality. These could potentially result in the studied cohort being unrepresentative of the wider population.
In most cases, the time interval between symptom onset and TCD ultrasound exceeded 24 hours. This could have potentially resulted in some early dynamic changes of blood ow being missed [44]. Future work is required to recruit a larger sample with suspected acute stroke as soon as possible after symptom onset, ideally at the pre-hospital stage, to validate the ndings of the current exploratory work.
A proportion of patients with acute ischaemic stroke due to LVO can present with mild neurological de cit [45]. As demonstrated previously, 9-13% of patients with NIHSS ≤4 had a LVO detected on angiography [46,47] which can be successfully treated with endovascular therapy [48]. The proposed triaging model has been designed to identify cases with a moderate to severe de cit that would bene t from reperfusion therapy. If based purely on clinical assessment, LVO cases with milder symptoms could be missed. We suggest that positive TCD ndings might potentially decrease the rate of false negatives.
Angiography was not used as a reference test for detecting LVO as it is not routinely performed in this study centre at present. Therefore, further assessment with CT or magnetic resonance angiography would be recommended to con rm the diagnosis of a major intracranial vessel occlusion. MRI diffusionweighted imaging would be required to con rm the diagnosis of an AIS and to exclude a stroke mimicking condition in those cases where CT imaging was negative.
Multivariate analysis has not been performed as part of the current interim analysis due to the low sample size, however further data analysis on a larger population would be required using a multivariate model which may allow more accurate predictions. Validation in an external cohort is advisable to test the generalisability of the proposed model.
An important and well recognised technical limitation of transcranial grey-scale/TCD ultrasound is the insu cient temporal acoustic window, which has been shown to occur in around 8% of the European general population [49] to 29% in the Oriental general population [50]. It is more commonly seen in elderly subjects, especially in women over 50 years old [49]. In our study, an insu cient window was seen in a slightly higher proportion of participants -about 40%. This may be explained by the fact that stroke patients who were generally older were recruited. However, an insu cient window only precluded a treatment decision in 4/50 (8%) patients. Further work to optimise ultrasound probe design and lowfrequency insonation may be justi ed.

Conclusions
A novel treatment decision model for patients with suspected acute stroke based on transcranial ultrasound ndings and clinical assessment has been proposed and tested in a small group of 50 subjects. It showed a better diagnostic performance when compared to a similar system using clinical assessment alone. A correct treatment decision was made in around 70% of patients with LVO and all subjects with other types of ischaemic stroke. Further work is needed to re ne the proposed model in a larger sample with acute stroke. If further validated, this system could potentially be implemented in the pre-hospital stroke care setting in remote and rural areas.

Consent for publication
Consent for publication was obtained from participants or their relative / legal representative according to the Adults with Incapacity Act (Scotland, 2000).

Availability of data and materials
The collected and analysed data are available from the corresponding author upon request. Table 1. Difference between the two groups of patients with a stroke and stroke mimics. Table 2. Difference between the two groups based on transcranial ultrasonography ndings. Table 3. Difference between the two groups on the basis of clinical features.   TCD ultrasound ndings in a patient with a massive right MCA territory infarct. Absent ow in the middle cerebral artery and reversed ow in the ipsilateral anterior cerebral artery (A) when compared to the vessels on the unaffected side (B).

Figure 3
Brain imaging ndings in a patient with acute intracranial haemorrhage. CT brain scan (A) and transcranial grey-scale ultrasonography (B) showing a 12 mm haemorrhagic lesion in the right basal ganglia.