Subjects

After obtaining IRB approval, we prospectively screened adult patients (≥ 18 years) with any intracerebral pathology admitted to the neurocritical care unit at a tertiary care academic medical center in Aug 2020 and then Nov 2020–March 2021 who were anticipated to stay in the intensive care unit for > 72 h. The study recruitment period reflects the impact of pandemic-imposed research enrollment restrictions. Consent was obtained in person from the legally authored representative if patients were unable to consent. Electronic consents were not available for research trials during this study recruitment period and visitor restrictions were in place due to COVID. Patients were included if they had cerebral imaging with CT or Magnetic Resonance Imaging (MRI) performed as part of standard of care and an investigatory ultrasound could be performed within 72 h of the CT or MRI scan. Patients were excluded if they had any penetrating head trauma or skull defects that could affect ultrasound insonation, if they had isolation precautions in effect for suspected or confirmed COVID infection and if images could not be obtained within 72 h of last CT or MRI imaging.

Design

All scans were performed by an investigator (AS) with experience in neuroimaging, diagnostic neurovascular sonography and point of care ultrasound. The investigator performed B mode imaging only when able to be completely blinded to the patient’s diagnosis and not participating in direct patient care. A report was logged by the investigator immediately after the ultrasound scan to report ultrasound findings, namely, if the patient had an intracerebral lesion suspicious for intracerebral hemorrhage based on the presence of hyperechoic signal at a location not corresponding to an expected intracerebral anatomical structure. Stored images were reviewed post-hoc by the same investigator with comparison to patient’s CT scans for elucidating B-mode pathology corresponding to signals seen on ultrasound. Knowledge gained from each review informed future imaging. For example, choroid plexus was deemed to cause false positive findings by mimicking possible hemorrhage. After comparison with CT scan, the location of hyperechoic signals in relation to each other, such as the midbrain, provided the expected visualization for future scans to note presence of choroid plexus. All images were serially reviewed with a faculty neuroradiologist (PB) to define different anatomical structures visualized on cranial ultrasound and explore the correlative anatomy and artifacts seen on a cranial ultrasound. We considered this step necessary, since there was no available resource on cranial topography as a reference beyond the basic anatomy of visualizing the midbrain, sphenoid wing and petrous part of the temporal bone commonly used for transcranial color-coded duplex imaging. Since we reviewed ultrasound images and corresponding CT scans after every image acquired in an iterative fashion, each scan served as a learning tool for follow-up scans helping readjust interpretation.

Ultrasound imaging

A low frequency 3–1 MHz phased array probe (echo probe) on ICU POCUS device (Fujifilm, Sonosite® Xporte) was used to perform B mode images of the brain using the temporal windows by the investigator blinded to patient’s diagnosis. Transcranial and abdominal presets were used to visualize the opposite skull visible as a hyperechoic shadow. A depth of 13–16 cm was used and adjusted to ensure the opposite skull could be visualized. Probe index marker was pointed toward the eyes with line of insonation aligned to get an axial section of the brain. Transcranial preset was used due to high mechanical and thermal index designed to facilitate imaging through the skull. Abdominal presets were investigated to allow future scalability of this feasibility study on hand-held machines, most of which don’t have transcranial presets. The midbrain was visualized as a butterfly shaped structure. The exam was repeated on each side to assess for presence of a temporal window. The patient was excluded if neither temporal windows were present. Once adequate windows were confirmed with both the skull and midbrain visualized, the probe was pointed above and below the level of midbrain and then positioned to look for any reproducible signals corresponding to expected anatomical structures or abnormal pathology. Signals corresponding to a lesion or anatomical structure were labelled as “hyperechoic” if the lesion/structure appeared white (or bright) with echogenicity similar to the echoic character of the skull visible on B mode.

Analysis

We present a descriptive analysis of patient demographics and ultrasound imaging characteristics visible on cranial ultrasound in relation to its ability to detect ICH. We evaluated images for presence of anatomical landmarks, brain topography and alternative etiologies that could potentially affect sensitivity and specificity of an ICH diagnosis using point-of-care cranial ultrasound. We compared the results of using static images on B mode ultrasound as well-recorded cine loops. Acquisition of images using the abdominal preset was compared to findings on the transcranial Doppler preset. Inferential statistics were used to generate exploratory data on sensitivity, specificity and accuracy of POCUS diagnosis of ICH in comparison with CT scan diagnosis of ICH. We calculated 95% exact confidence intervals using Jeffrey’s intervals due to the small sample size [15]. The exploratory nature of this study design and small sample size prohibited meaningful analysis of impact of age, time since onset, or comparison of CT Hounsfield units to ultrasound detection of hemorrhage. The STROBE checklist was used to report findings, since feasibility was the initial aim with secondary aim to describe imaging technique and topography relevant to ultrasound-based ICH detection.

Results

A total of 30 patients were eligible during the screening period. Consent could not be obtained in 10 patients despite documented temporal windows due to inability to locate or contact legally authorized representative to obtain consent. Three patients had no temporal windows. Of the 17 consented with temporal windows, 4 patients could not receive an ultrasound during their hospital stay due to death, discharge or sonographer unavailability, and 2 patients had expected concurrent CT/MRI imaging delayed, thus the ultrasound scans occurred ≥ 72 h from last CT/MRI. Ultrasound images were obtained and saved successfully for review in 11 patients (Table 1). These patients had a mean age of 57.45 years (11 patients, 28–77 years, 5 males). The mean time between CT/MRI and ultrasound was 13.3 h (21 min–39 h). 7 patients had some form of intracranial hemorrhage reported on CT brain (one with hemorrhagic conversion within ischemic stroke one with SAH but interhemispheric bleed), 3 patients had ischemic stroke without (Table 2) any hemorrhagic conversion, and 1 patient had a thalamic tumor.

Table 1 Flow diagram of study enrolmentTable 2 Patient and imaging characteristics from CT scan and post hoc ultrasound analysis of enrolled patients who had temporal windowsImaging technique

We used the low frequency echo probe placed on the patient's temple with index marker of the probe pointed toward the patient’s eyes (Additional file 1: Figure S1). The probe was positioned with the line of insonation aligned with an imaginary line from lateral canthus of the eye to the tragus. Depth of the image was adjusted to allow insonation of the opposite skull typically requiring 13–16 cm depth. Once the skull was visualized, the probe was slowly rocked cranially caudally until the butterfly shaped midbrain was visualized which allowed distinction of ipsilateral and contralateral cerebral hemisphere. We then rocked the probe further cranially and caudally followed by anteriorly and posteriorly to look for other hyperechoic signals. In general, visualization of brain parenchyma was more feasible contralateral to the window insonated (Fig. 1). Hence, left temporal windows were best suited to assess for right-sided parenchymal pathology and vice versa. The sector nature of the probe and artifact produced by ipsilateral sphenoid wing made it difficult to visualize the anatomy ipsilateral to the insonated window. Abdominal presets with high gain settings provided comparable resolution for imaging brain to elucidate known anatomical and abnormal pathological structures when compared to transcranial preset. Transcranial preset tended to produce some artifacts that were not visible in abdominal presets. We did not find any challenges with imaging associated with placement of electroencephalography electrodes or invasive monitoring devices. Rocking the probe from one location did not enable visualization of the whole anterior–posterior span of the skull and we found it useful to move the probe across the temporal bone for maximum visualization, which sometimes led to the loss of temporal window. Realigning the probe to visualize the opposite skull was then re-attempted and the process repeated. The duration of each US was 5–10 min to image both the right- and left-side windows. Sonographer position at the side of the bed was used for all scanning and it was feasible to scan both sides while standing in one location. No patient positioning was necessary for the study scans. Though none of the study patients had significant head movement or agitation during scanning, we did need to adjust the probe for head turning or cough.

Fig. 1

Comparison of cranial B mode images on abdominal and transcranial preset with anatomical landmarks visible. Right Image Panel A, C are transcranial presets and Right Image Panel B, D are abdominal presets on same image. Marking on these images are labelled as followed: Blue line—opposite skull, orange line sphenoid wing and petrous part temporal bone, green midbrain, orange dot inside the midbrain—cerebral aqueduct. Left image shows unlabeled images of of both the transcranial presets (A, C) and abdominal presets (B, D)

Topography on cranial ultrasound imaging

The most discernible anatomical structures visible on B mode imaging in patients with windows have been outlined in Figs. 1 and 2 and included:

a)

Opposite skull visible as a convex hyperechoic signal farthest from the probe. Depth was adjusted to visualized this at the bottom of the image

b)

Midbrain, visible as a hypoechoic butterfly-shaped structure in the center of the ultrasound image. Peduncles and colliculi were generally visible to allow identification of midbrain.

c)

Cerebral aqueduct visible as a hyperechoic shadow in the posterior part of the midbrain.

d)

Sphenoid wings and petrous part of the temporal bone visible as hyperechoic signal, both ipsilateral and contralateral.

e)

Falx cerebri visible as a linear hyperechoic signal originating from frontal end of the ultrasound images.

f)

Calcified choroid plexus visible in midline and in the temporal horns of lateral ventricles.

Fig. 2

Abdominal presets showing the cerebral aqueduct and choroid plexus on cranial B mode imaging. Upper panel shows abdominal presets with markings labelled as followed: Blue line—opposite skull, green—midbrain, orange dot inside the midbrain—cerebral aqueduct, yellow line—falx cerebri, orange rectangles—choroid plexus calcification in lateral ventricles. Lower panel shows unlabeled images of abdominal presets

Suspected hemorrhage could be visualized as a hyperechoic signal reproducible in multiple planes by rocking the probe in anteroposterior and cranial caudal directions at a location, where no other hyperechoicity was expected (Fig. 3). In addition to above anatomical landmarks, the following reproducible hyperechoic artifacts were identified as possible hemorrhage mimics: (Fig. 4).

a)

Hyperechoic signals in cisterns around the midbrain with signal intensity lower than bone (opposite skull).

b)

Hyperechoic shadows inferior to the midbrain (midbrain acoustic shadow) could appear similar to hyperechoic signals produced by a hemorrhage. This contributed to one of the earlier false positive findings.

c)

Hyperechoic signals parallel to the occipital bone causes by thick skull ridges in the posterior fossa.

Fig. 3

Intracerebral hemorrhage visible as a hyperechoic signal best visualized contralateral to the insonated window. A—Transcranial preset, B—abdominal preset, C—COMPUTED tomography brain (CT) scan. Blue line—opposite skull, yellow line—falx cerebri, blue shape outlines the hyperechoic signal corresponding to hemorrhage on CT scan

Fig. 4

Artifact created by acoustic shadow of the midbrain causing a false positive finding of hemorrhage More visible on transcranial preset A but less enhanced on abdominal preset B. Blue line—opposite skull, green—midbrain

Blinded imaging results

Temporal windows were found in all but 3 patients (18.75%, including two that were excluded from post hoc analysis) which is comparable to published literature [16, 17]. The blinded investigator made a point of care diagnosis of ICH with 100% sensitivity and 50% specificity in comparison with a head CT (Table 3). The hyperechoic signals created by the artifact produced by midbrain acoustic shadows (Fig. 4) and a thalamic tumor (Fig. 5) contributed to the two false positive findings. Ischemic stroke did not create any discernible signals on ultrasound (Fig. 6).

Table 3 Results of blinded investigator-based diagnosis using POCUS compared to patient diagnosisFig. 5

Thalamic tumor creating a hyperechoic signal similar to intracerebral haemorrhage. A—Transcranial preset, B—abdominal preset, C—computed tomography (CT) brain scan. Blue line—opposite skull, green—midbrain. Blue shape outlines the hyperechoic signal corresponding to the tumor on CT scan

Fig. 6

Acute ischemic stroke does not produce a characteristic appearance of ultrasound that allows ultrasound-based diagnosis. A—Transcranial preset, B—abdominal preset, C—computed tomography brain scan. Blue line—opposite skull, green midbrain, orange dot inside the midbrain—cerebral aqueduct, yellow line—falx cerebri

Post hoc analysis

During post hoc image analysis, static snapshots of B mode images were compared to recorded cine loops by the principal investigator and neuroradiologist. We observed that review of cine loops revealed better delineation of relevant anatomical landmarks as well-reproducible hyperechoic lesions with appearance of a hemorrhage. None of the hyperechoic lesions identified as ICH were well-defined enough to reliably measure dimensions on cranial ultrasound.

Sample size estimation for future study to assess accuracy

Based on this exploratory study and the published literature in this area, (Additional file 2: Table S1) we determined that a total sample size of 310 patients (with 31 patients with ICH assuming the prevalence of ICH is 0.1 in this population) will achieve 81% power to detect a change in sensitivity from 0.7 to 0.9 using a two-sided binomial test, in other words ruling out a sensitivity of 70% (Null hypothesis) in favor of a sensitivity of 90% (or higher). In addition, there is 90% power to detect a change in specificity from 0.83 to 0.9 using a two-sided binomial test, in other words ruling out a specificity of 83% (Null Hypothesis) in favor of a specificity of 90% (or higher). The target significance level for these tests are 0.05 (two-sided). The actual significance level achieved by the sensitivity test is 0.0478 and achieved by the specificity test is 0.0378. (Additional file 3: Table S2).