REVIEW

Neuroimaging of Intracerebral Hemorrhage

∗
Rima S. Rindler, MD
Jason W. Allen, MD, PhD
Jack W. Barrow, BS§
Gustavo Pradilla, MD∗
Intracerebral hemorrhage (ICH) accounts for 10% to 20% of strokes worldwide and is
‡ associated with high morbidity and mortality rates. Neuroimaging is indispensable for
rapid diagnosis of ICH and identification of the underlying etiology, thus facilitating
triage and appropriate treatment of patients. The most common neuroimaging modal-

Downloaded from https://academic.oup.com/neurosurgery/article-abstract/86/5/E414/5766367 by guest on 12 July 2020

ities include noncontrast computed tomography (CT), CT angiography (CTA), digital
Daniel L. Barrow, MD∗
subtraction angiography, and magnetic resonance imaging (MRI). The strengths and disad-
∗
Department of Neurosurgery, Emory vantages of each modality will be reviewed. Novel technologies such as dual-energy
University Hospital, Atlanta, Georgia; CT/CTA, rapid MRI techniques, near-infrared spectroscopy, and automated ICH detection
‡
Department of Radiology and Imaging hold promise for faster pre- and in-hospital ICH diagnosis that may impact patient
Sciences, Emory University Hospital, Atla-
nta, Georgia; § Mercer University School management.
of Medicine, Savannah, Georgia
KEY WORDS: Computed tomography, Digital subtraction angiography, Intracerebral hemorrhage, Magnetic
resonance imaging, Neuroimaging
Correspondence:
Rima S. Rindler, MD,
Neurosurgery 86:E414–E423, 2020 DOI:10.1093/neuros/nyaa029 www.neurosurgery-online.com
Emory University Hospital,
1365 Clifton Road, Suite B6200,
Atlanta, GA 30322, USA.
ways.4 First, it is indispensable for diagnosing

Email: rrindle@emory.edu
Received, September 22, 2019.
Accepted, December 28, 2019.
Published Online, February 28, 2020.
Copyright 
C 2020 by the
Congress of Neurological Surgeons
S
troke is the second leading cause of
death worldwide. Intracerebral hemor-
rhage (ICH) accounts for 10% to 20% of
strokes, and is associated with a mortality rate
of approximately 40%.1,2 The absolute number
of hemorrhagic strokes has increased by approx-
imately 50% since 1990, with the majority
occurring in low-income populations.3 These
surprising numbers are attributed to an enlarging
and aging world population, as well as improve-
ments in recognition of hemorrhagic stroke.
As population growth has occurred dispro-
portionately in underdeveloped countries, and
management of stroke risk factors has improved
in developed countries, the burden of hemor-
rhagic stroke remains in lower socioeconomic
populations.
Given the high incidence and mortality of
ICH, rapid diagnosis and treatment of these
patients is imperative. Neuroimaging provides
the foundation for this process in 3 important
ICH. Second, neuroimaging can identify the
etiology of ICH, which is instrumental in
guiding treatment and minimizing recurrence.
Lastly, it identifies unique qualities of the
hemorrhage or pathology that may impact a
patient’s overall prognosis for recovery and
risk of mortality.5,6 The most relied upon
neuroimaging modalities, including noncontrast
computed tomography (NCCT), CT angiog-
raphy (CTA), digital subtraction angiography
(DSA), and magnetic resonance imaging (MRI),
will be reviewed.
ICH ETIOLOGY
Primary ICH accounts for 70% to 80%
of all ICH. The majority of these (40%-
60%) are caused by rupture of diseased small
penetrating arteries that have been weakened
by long-standing hypertension. The resulting

ABBREVIATIONS: AVM, arteriovenous malformation; CAA, cerebral amyloid angiopathy; CT, computed tomog-
raphy; CTA, CT angiography; CTV, CT venogram; DECT, dual-energy CT; DSA, digital subtraction angiography;
ENRICH, Early MiNimally-invasive Removal of IntraCerebral Hemorrhage; FDA, Food and Drug Administration;
GRE, gradient recalled echo; ICH, intracerebral hemorrhage; INVEST, Minimally Invasive Endoscopic Surgical
Treatment With Apollo/Artemis in Patients With Brain Hemorrhage; MRA, MR angiography; MRI, magnetic
resonance imaging; MISTIE, minimally invasive surgery plus alteplase in intracerebral haemorrhage evacu-
ation; NCCT, noncontrast computed tomography; NIRS, near-infrared spectroscopy; QSM, quantitative suscep-
tibility mapping; SWI, susceptibility-weighted imaging; TOF, time-of-flight; TR, time-resolved; VIPS, volumetric
impedance phase-shift spectroscopy

E414 | VOLUME 86 | NUMBER 5 | MAY 2020 www.neurosurgery-online.com


hypertensive arteriopathy leads to tunica media hyperplasia in
the vessel wall.7 Rates of ICH in patients with hypertension are
roughly 3.5 times those of normotensive individuals.2 Primary
hemorrhages are less commonly caused by cerebral amyloid
angiopathy (CAA; 5%-10%),2 a condition created by abnormal
deposition of beta-amyloid protein in the tunica media. The
incidence of CAA increases with age. Secondary ICH is caused by
a variety of structural and physiological pathologies. Cerebrovas-
cular lesions, including arteriovenous malformations (AVMs),
cavernous malformations, and aneurysms, account for approx-
imately 10% to 20% of ICH. Less common etiologies include
neoplasms, alcohol abuse, use of sympathomimetic drugs, coagu-
lopathy, mycotic aneurysms, moyamoya disease, vasculitis, dural
arteriovenous fistula, and hemorrhagic conversion of ischemic
stroke.
IMAGING OF ICH
Computed Tomography
ICH is most often first detected on an NCCT study in an
emergency setting. NCCT rapidly and accurately identifies an
acute ICH as a hyperdense lesion within brain parenchyma.
NCCT is less expensive than other imaging modalities and
is widely available in U.S. emergency departments. For these
reasons, NCCT is the recommended imaging technique for
diagnosing ICH in patients presenting with an acute neurological
deficit.8 NCCT is also ideal for inpatient monitoring of ICHs,
either as routine surveillance or in response to a patient’s neuro-
logical deterioration.
NCCT defines the basic characteristics of an acute ICH, which
dictate medical or surgical management and overall prognosis.
Acute ICH usually measures 30 to 80 Hounsfield units, though
this can vary by hemorrhage protein concentration and serum
hematocrit level.9 Acute ICH is usually round/ellipsoid with
defined borders and minimal hypodense perihematomal edema.
Any clot heterogeneity, such as the “swirl sign,”10,11 portends
higher risk (>33%) of hematoma expansion and poor outcome
and mortality.12,13 Over time, clot density decreases by 2
Hounsfield units daily,14 and maintains a residual ring-like hyper-
density in the subacute phase before becoming iso- to hypodense
to gray matter in the chronic phase.15
NCCT defines the primary hematoma location, which holds
significance for a hemorrhage’s etiology and the patient’s
prognosis.4 Hypertensive ICH primarily occurs in the basal
ganglia (35%), followed by cerebral lobes (25%), thalamus
(20%), cerebellum (8%), and pons (7%).16 CAA is most
commonly found in the cerebral lobes.1 Location of secondary
ICHs depends entirely on the location of the underlying
pathology. Thirty-day mortality is highest for brainstem hemor-
rhages (60%), followed by deep (44%) and lobar (40%) hemor-
rhages, with lowest mortality for cerebellar hemorrhages (34%).17
Notably, ICH location on NCCT has been the basis for decision-
making in recent clinical trials evaluating the effect of ICH evacu-
NEUROSURGERY
NEUROIMAGING OF ICH
ation on patient recovery; pending results of those trials, NCCT
will likely continue to be a focal point for initial triage of patients
that might benefit from operative intervention.18-20
ICH volume on NCCT also holds prognostic significance.
Large clot size predicts hematoma expansion and those exceeding
30 cc are associated with high mortality.1,21 An accurate
pragmatic method of measuring ICH volume on NCCT that
clinicians often utilize is the ellipsoid method, as follows: (A ×
B × C)/2, where A is anteroposterior diameter, B is width, and C
is height.22,23 A CT scanner software is capable of automatically
Downloaded from https://academic.oup.com/neurosurgery/article-abstract/86/5/E414/5766367 by guest on 12 July 2020
calculating ICH volume using an algorithm and is more accurate
than manual calculation if precise measurement is required. In
most clinical scenarios, the ellipsoid method suffices.
Secondary findings detected on NCCT might also affect initial
patient stabilization and management.4 Presence of intraven-
tricular hemorrhage portends high risk of developing obstructive
hydrocephalus and decreased survival.24 Mass effect from
the hemorrhage or perihematomal edema may manifest as
midline shift or distortion of normal adjacent structures. In
the acute phase, progressive mass effect from edema expansion,
usually occurring within 2 d, may lead to elevated intracranial
pressure with life-threatening herniation syndromes such as
subfalcine herniation in the anterior fossa, downward or transten-
torial herniation in deep subcortical structures, uncal herni-
ation in the middle fossa, or upward herniation or downward
tonsillar herniation through the foramen magnum in the posterior
fossa.4,25,26 Any combination of these findings is cause for short-
interval repeat imaging to monitor progression, intervening with
a ventriculostomy or craniotomy, and considering more specific
imaging studies.
CT Angiography
Following ICH detection, most other neuroimaging modalities
are obtained in an effort to determine etiology, although there
is wide variability in the type, number, and order of imaging
techniques preferred by treating physicians.27 CTA is commonly
the next study of choice, as it rapidly and accurately identifies
several vascular abnormalities that require prompt intervention
such as AVMs or aneurysms. If indicated, the CTA can be
performed immediately after the NCCT without relocating the
patient. CTA provides high-resolution images of cerebral arteries
by using helical scanning techniques on multidetector scanners at
1.5 mm slice thickness. Using an iodine-based contrast admin-
istered intravenously, images are acquired as the contrast passes
through the intracranial arteries.4 Occasionally, CTA can detect
presence of a tumor blush if the image acquisition occurs slightly
after the arterial contrast phase.
If CTA is obtained within hours of an acute ICH, it may be
useful for predicting hematoma expansion which is associated
with clinical deterioration, mortality, and worse functional
outcome28 making rapid identification critical. On CTA, the
“spot” sign is a hyperdense dot of contrast (greater than 120
Hounsfield units) within the nonenhancing hematoma that is not
VOLUME 86 | NUMBER 5 | MAY 2020 | E415
RINDLER ET AL
FIGURE 1. Axial CTA with contrast showing a right frontotemporal ICH
with 2 small foci of enhancement compatible with the “spot sign,” indicated
by the arrow, suggestive of active extravasation.
continuous with a vessel.24 Extravasation is suspected if this “spot”
enlarges on immediate postcontrast imaging (Figure 1). The spot
sign is an independent predictor of hematoma expansion (average
8.6-14.3 cc volume) with a 61% to 77% positive predictive
value.29,30 Higher spot sign density or more than 3 spot signs also
predict clot expansion.24 The spot sign is associated with longer
hospital stays,29 increased mortality rate, and worse functional
outcome.30 Consideration of short-interval repeat imaging or
operative intervention is prudent in these patients.
A CT venogram (CTV) acquires images during the venous
phase of contrast. Venous structures can often be seen on CTA
depending on the timing of image acquisition relative to the time
of contrast administration. A CTV is useful for detecting a venous
sinus thrombus that may result in a venous infarction with hemor-
rhagic conversion. Unfortunately, smaller cortical vessel throm-
bosis is likely underdiagnosed.
The disadvantages of CTA are the associated risks of
radiation, contrast-induced nephropathy, and life-threatening
allergic reactions. The potential benefits of an emergent CTA
must be weighed carefully.31
Digital Subtraction Angiography
DSA is the gold standard diagnostic modality preceding the
development of CTA for detecting macrovascular causes of ICH.4
In addition to its ability to detect more obvious structural vascular
abnormalities (eg, AVMs, aneurysms), DSA has the advantage
over CTA of time resolution that characterizes direction and rate
of blood flow within the cranial circulation. This allows detection
of arteriovenous shunting within a dural arteriovenous fistula that
might otherwise be missed on CTA. Furthermore, therapeutic
E416 | VOLUME 86 | NUMBER 5 | MAY 2020
intervention can be performed in sequence at the time of the study
with DSA if necessary.
The angiographic yield of DSA, alone, is highest for younger
patients (<45 yr) without a history of hypertension,32 with
absence of small vessel disease on NCCT,33 and with cerebellar or
lobar ICH locations.34 However, even in older adults with small
vessel disease and pre-existing hypertension33 or deep ICH,34
DSA still has a diagnostic rate of 1% to 2%. Notably, CTA
has replaced DSA in many institutions as the initial vascular
screening study for patients presenting with ICH due to its
Downloaded from https://academic.oup.com/neurosurgery/article-abstract/86/5/E414/5766367 by guest on 12 July 2020
noninvasiveness and reduced radiation doses while maintaining
excellent ability to detect macrovascular causes of ICH (95%
sensitivity and 99% specificity).35,36 However, a recent study
suggests that CTA, alone, obtained in the acute phase has a more
modest sensitivity of 74%, specificity of 91%, and a diagnostic
yield of 17%, even in patients with the highest pretest proba-
bility of harboring a vascular abnormality (age < 45 yr, no hyper-
tension).37 DSA is still very useful in the setting of a negative
CTA: the DSA yield for these patients without small vessel disease
or hypertension is 22.1%. These data suggest that DSA should
still be performed in populations at highest risk of macrovas-
cular pathology if CTA is negative or inconclusive and should
be considered for lower risk populations if no other etiology for
ICH is found. Additionally, if any odd features are present, such as
significant calcification, subarachnoid blood, or unusual hemor-
rhage locations, DSA should be performed.38
Other disadvantages of DSA include its associated procedural
risks of bleeding and stroke (<1%) and the requisite availability of
specialty-trained diagnosticians and interventionalists to perform
and interpret the study.
Magnetic Resonance Imaging
MRI is rarely used as a primary neuroimaging survey for
acute ICH although it is as accurate as NCCT in detecting
acute ICH.39 An MRI “spot sign” that corresponds to active
contrast extravasation has been described, and observation studies
suggest an association between hematoma enlargement and poor
clinical outcome, similar to the CT spot sign.40,41 Unfortunately,
utilizing MRI as a primary survey is currently not practical due
to cost, availability, and potentially increased risk of prolonged
imaging time to the patient. As a result, MRI techniques are most
useful for deciphering the etiology of ICH in order to stratify risk
of recurrence and to guide secondary prevention strategies.
An MRI of the brain with and without gadolinium contrast
is usually obtained if CTA is negative for vascular abnormal-
ities. The imaging sequences included allow the broadest evalu-
ation of etiology over other modalities. MRI provides excellent
detection of markers of small vessel disease, including white
matter changes, enlarged perivascular spaces, lacunar infarc-
tions, and deep microhemorrhages that point to hypertensive
arteriopathy as the culprit. Furthermore, MRI is accurately
able to identify lobar microhemorrhages, cortical superficial
siderosis, and cortical atrophy, which are used in the modified
www.neurosurgery-online.com

FIGURE 2. MRI of 4 patients with ICH of different ages demonstrating the
appearance on T1, T2, and GRE sequences. Note that unlike acute, early
subacute, and chronic ICH, late subacute ICH are associated with minimal,
generally peripheral GRE susceptibility.
Boston criteria to diagnose CAA (sensitivity: 94.7%, specificity:
81.2%).42-44 This is prognostically significant, as CAA-related
hemorrhage carries greater than 7% annual45 risk of recurrence
vs a 1.1% risk with non-CAA-related hemorrhages. Nodular or
rim enhancement that does not correspond to a hemosiderin ring
is highly suggestive of an underlying tumor.46 Cavernous malfor-
mations, which are otherwise difficult to characterize on CT and
are typically invisible on DSA, have a characteristic appearance
on MRI as a hypointense ring around a heterogeneous core on
gradient recalled echo (GRE)/T2∗ sequences. Hemorrhagic trans-
formation of an ischemic infarct reveals a small area of abnormal
GRE within a larger area of diffusion restriction on diffusion-
weighted imaging/apparent diffusion coefficient.
MRI is superior to CT modalities for dating hemorrhage age,
and especially for visualizing chronic hemorrhages (Figure 2).39
Hematoma appearance is dependent on the paramagnetic effects
of hematoma components related to the timing of the hemor-
rhage, magnet strength, and pulse sequence (Figure 3). Histor-
ically, T1 and T2-weighted images have been utilized to date
hematomas, though GRE sequences are also useful, especially
for identifying microhemorrhages or older hemorrhages. On T1
sequences, hemorrhages transition from isointense in the hyper-
acute/acute phases, to hyperintense in the subacute phase, and
to hypointense in the chronic phase. On T2 sequences, hemor-
NEUROSURGERY
NEUROIMAGING OF ICH
Downloaded from https://academic.oup.com/neurosurgery/article-abstract/86/5/E414/5766367 by guest on 12 July 2020
FIGURE 3. Schematic diagram of ICH evolution on T1 and T2-
weighted images. Oxygenated hemoglobin appears as increased T2 and
intermediate T1 signal (1. Hyperacute). The paramagnetic effects of
deoxygenated hemoglobin and intracellular methemoglobin produce T2
hypointensity, while T1 signal is slightly hypointense and then becomes
hyperintense (2. Acute, 3. Early subacute). Extracellular methemoglobin
does not have paramagnetic effects and has hyperintense signal on both T1
and T2-weighted images (4. Late subacute). Finally, hemosiderin again
has paramagnetic effects and demonstrates both T1 and T2 hypointensity
(5. Chronic). The length of time in each phase is approximated in paren-
theses. Hb, hemoglobin; met-Hb, methemoglobin.
rhages are hyperintense in the hyperacute phase, hypointense in
acute and early subacute phase, hyperintense in late subacute
phase, and hypointense in the chronic phase. Fluid-attenuated
inversion-recovery sequences typically parallel changes in T2
sequences. GRE/T2∗ sequences depict hyperacute hemorrhages
with a hypointense rim and isointense core, with increasing core
hypointensity in the acute and subacute phases followed by a slit-
like hyperintensity in the chronic phase.15,47
In recent years, MR angiography (MRA) has been evaluated
as an alternative to CTA and DSA for detecting vascular abnor-
malities. Two or three-dimensional MRA time-of-flight (TOF)
sequences depict images of intracranial arteries based on direction
of blood flow without the need for contrast administration.
Administration of intravenous gadolinium contrast significantly
shortens acquisition time. Time-resolved (TR) images add
temporal resolution to MRA: a series of sequential images are
obtained in a region of the brain following a contrast bolus. MRA
TOF-TR images, together, allow for dynamic imaging of blood
flow through the brain and can identify feeding arterial vessels and
draining veins in AVMs and dural AV fistulas.48,49 The spatial
resolution of 1.5 Tesla MRA is not as robust as CTA and DSA,
although this is becoming less of a concern as higher magnetic
field strengths are introduced for clinical use that substantially
improve contrast-to-noise ratios (3 and 7 Tesla).50 A meta-analysis
determined that MRA detection of macrovascular abnormal-
ities has a 98% sensitivity and 99% specificity compared with
DSA, and is not statistically different from CTA.35 Furthermore,
VOLUME 86 | NUMBER 5 | MAY 2020 | E417
RINDLER ET AL
delayed MRI/MRA presents an additional 1% diagnostic yield
when compared to a negative CTA alone (18%).37 In other
words, MRI/MRA rarely diagnoses macrovascular causes of ICH
over CTA but can lead to diagnosis of other etiologies. This
is especially relevant for patients with allergies to iodine-based
contrast or severe renal disease that may preclude safe CTA or
DSA. In general, patients who present with spontaneous ICH in
whom a CTA and/or DSA is negative should undergo multimodal
contrasted MRI to investigate the underlying etiology. If this is
initially unrevealing, an MRI should be repeated several weeks
later, as an underlying cavernous malformation or tumor may be
obscured by the acute hematoma. As MRI technologies become
faster, of higher quality, and more readily available, this imaging
modality may become a strong alternative to CTA and or DSA as
a secondary neuroimaging study in ICH.
Imaging Predictors of Clot Consistency
Clinical trials in the last decade have evaluated the efficacy
of various techniques for ICH evacuation to improve morbidity
or mortality. Completeness of clot removal for each technique
is appropriately scrutinized in these trials. Investigators note
that the intraoperative clot density varies among patients, from
tough and fibrous to soft and easily suctioned, thereby poten-
tially impacting the effectiveness of one surgical technique over
another for a particular clot type. For instance, low-density,
soft clots may best be evacuated by mechanical aspiration via a
stereotactically placed catheter and/or thrombolysis (minimally
invasive surgery plus alteplase in intracerebral haemorrhage
evacuation, MISTIE trial)19 or with endoscopic evacuation
(Minimally Invasive Endoscopic Surgical Treatment With
Apollo/Artemis in Patients With Brain Hemorrhage, INVEST).18
Higher density clots might be more effectively evacuated with
direct visualization through a minimally invasive parafascicular
approach,51 as is being investigated in the Early MiNimally-
invasive Removal of IntraCerebral Hemorrhage (ENRICH)
trial.20
Intraoperative Imaging
Intraoperative neuroimaging is especially useful in patients who
require emergent operative procedures that preclude diagnostic
imaging before interventions. Such imaging can assist with ICH
localization or characterization of vascular etiologies. Direct ultra-
sonography of brain parenchyma can reliably locate an ICH for
direct aspiration.52,53 Some institutions have utilized integrated
intraoperative CT scanners for this purpose, and C-arm cone-
beam CT is being investigated as a faster alternative.54 DSA using
C-arm or biplane fluoroscopy can identify an underlying AVM
or aneurysm that can be addressed at the time of decompressive
craniotomy. If DSA is unavailable, intraoperative 3D rotational
fluoroscopy after contrast administration can be employed to
create an angiographic image. Finally, the advantage of a low-
field intraoperative MRI over other modalities is likely limited
to resection of cavernous malformation-related ICH. Although
hybrid suites that combine such sophisticated imaging capabil-
E418 | VOLUME 86 | NUMBER 5 | MAY 2020
ities within an operating room are available in some institutions,
most others are limited to ultrasonography, which suffices for
basic management of ICH.55
UPDATES IN ICH NEUROIMAGING
Dual-Energy CT/CTA
Dual-energy CT (DECT) utilizes the simultaneous
measurement of 2 different energy spectra, allowing the compo-
Downloaded from https://academic.oup.com/neurosurgery/article-abstract/86/5/E414/5766367 by guest on 12 July 2020
sition of a voxel to be estimated through the energy-dependent
differential attenuation of the component elements. In particular,
DECT provides a means to differentiate between hemor-
rhage and iodinated contrast or calcification, which may have
overlapping attenuation coefficients and appear similar on NCCT
(Figures 4 and 5).56 DECT has been reported to have an up
to 99% accuracy rate in differentiating between ICH and
calcium.57 In addition, DECT can differentiate between ICH
and contrast staining. In a recent study, DECT revealed that
contrast staining mimicked ICH in 59% of patients imaged
within 18 h endovascular thrombectomy, which has important
implications for therapeutic decisions.58 DECT angiography can
also better detect lesions near osseous structures over single-source
CTA, especially dural arteriovenous fistulas and aneurysms near
the skull base.59,60 Recent studies show that contrast-enhanced
DECT can better predict ICH expansion than the spot sign on
single-source CT.61
There are several limitations of using DECT for detecting
ICH. These include increased noise related to the decomposition
algorithm, difficulty in correcting for beam hardening artifact,
and increased radiation dose.56,62 DECT is more expensive than
single-source CT, is less available than standard CT or MRI,
and requires longer processing and interpretation times often
without additional reimbursement.62,63 Some of these drawbacks
have been recently addressed as vendors are decreasing the
cost of DECT scanners and are evaluating new dose-reduction
techniques.64 Furthermore, virtual noncontrast images may be
generated from postcontrast DECT, removing the need for
additional scans and decreasing the patient’s radiation exposure.60
Magnetic Resonance Imaging
Improvements in MRI are occurring rapidly to improve spatial
resolution and decrease acquisition time. For example, significant
acceleration has been achieved with techniques such as SENSE,
GRAPPA, and compressed sensing.65-67 Implementation of these
techniques has the dual advantage of decreasing the length of time
a patient must be motionless during the scan—which is difficult
for many people—and allowing for increased throughput which
minimizes the need to interrupt more routine scans for emergent
cases.
MRI of ICH has also been improved by advances in
sequence development. Susceptibility-weighted imaging (SWI)
is a high-spatial resolution, 3-dimensional GRE MR sequence
that is more sensitive to magnetic field inhomogeneity than
www.neurosurgery-online.com
 NEUROIMAGING OF ICH

Downloaded from https://academic.oup.com/neurosurgery/article-abstract/86/5/E414/5766367 by guest on 12 July 2020

FIGURE 4. DECT images showing calcification of the posterior left thalamus. There is hyperdensity on noncontrast CT A, with a differential of calcium vs ICH.
Hyperdensity persists on CT with calcium overlay B, but not on virtual noncalcium image C, confirming that this represents calcium and not ICH.

FIGURE 5. DECT images showing lymphoma of the left thalamus. On the initial contrast-enhanced images, there is a hyperdense lesion within the left thalamus A,
with a differential of an enhancing mass vs ICH. There is persistent hyperdensity on the iodine overlay image B and only mild hyperdensity on the virtual noncontrast
image C indicating that this represents a hypercellular, enhancing tumor (lymphoma) and not ICH.

traditional GRE techniques.68 SWI has been shown to be
detect significantly more ICH than standard GRE in patients
with a variety of disorders such as diffuse axonal injury,
CAA, and cerebral cavernous malformations (Figure 6).69,70 By
exploiting the differences between paramagnetic substances—
such as deoxygenated hemoglobin, intracellular methemoglobin,
and hemosiderin—and diamagnetic effects of certain forms of
calcium, SWI-filtered phase images allow the differentiation
between these compounds.70 Unfortunately, the differentiation
between ICH and calcium using SWI is not always successful.
For example, oxygenated hemoglobin has diamagnetic effects and
ferrocalcinosis, such as found in the basal ganglia, has paramag-
netic properties, which may lead to erroneous classification.70,71
Furthermore, the “blooming” effect that occurs with SWI may
obscure adjacent anatomy and render volumetric measurements
inaccurate. Quantitative susceptibility mapping (QSM) is an
extension of SWI that allows the quantification of magnetic
susceptibility sources.70,72 Similar to DECT and SWI, QSM
allows for the differentiation between hemorrhage and calcifi-
cation (Figure 7).73 Importantly, unlike GRE techniques that

NEUROSURGERY VOLUME 86 | NUMBER 5 | MAY 2020 | E419

RINDLER ET AL
FIGURE 6. SWI is more sensitive for ICH than traditional GRE sequences. A, Standard GRE image shows several small ICH in a patient with diffuse
axonal injury. B, Two additional hemorrhages are identified on SWI sequence at the same slice in this patient (arrows).
are hampered by blooming artifacts, accurate measurement of
hematoma volume is possible with QSM.74 Therefore, QSM
may be used both in the initial diagnosis of ICH and follow-up
imaging, removing the significant radiation exposure related to
frequent repeat CT.73
Finally, over the past few decades, new low-field (0.25-1 T)
MR scanners have been developed that take advantage of recent
advancements in software and hardware.75 In comparison to
conventional modern scanners, their relatively small footprint
allows the versatility of site locations such as in the emergency
department or intensive care unit which increases accessibility
for patients with acute ICH that may need close monitoring.
In addition, as a large portion of the scanner cost is related
to the superconducting magnet, these low-field scanners are
significantly less expensive to purchase, do not require expensive
helium-based cooling system maintenance, and consume less
power.75 As low-field MR become more available, studies of their
impact on ICH are warranted.
Near-Infrared Spectroscopy
Near-infrared spectroscopy (NIRS) exploits differences in
light absorption and scattering to characterize tissues. NIRS
may be used for ICH detection as the greater concentration
of hemoglobin within an acute hematoma in comparison to
intravascular blood leads to greater light absorption.76 Several
NIRS devices have been developed and approved by the Food and
E420 | VOLUME 86 | NUMBER 5 | MAY 2020
Downloaded from https://academic.oup.com/neurosurgery/article-abstract/86/5/E414/5766367 by guest on 12 July 2020
Drug Administration (FDA) including portable sensors that may
be conveniently used in the field. A list of FDA-approved devices
is provided in the Table.
There are limitations to ICH detection with the NIRS devices
currently available.76,77 Although a recent multisite trial reported
a high sensitivity and specificity of up to 93% and 87%, respec-
tively, in the detection of ICH >3.5 mL and <2.5 cm from the
brain surface,76 other studies have reported much lower rates.77
The depth of near-infrared light penetration limits detection of
deep hemorrhages, and the size, type, and location of intracranial
hemorrhages cannot be determined with accuracy. Bilateral ICH
may be missed given that NIRS depends upon the differential
light absorbance between contralateral head locations. Patients
with traumatic brain injury may also have scalp hematomas
that produce false-positive results. Finally, variations in hair,
scalp, and skull thickness introduce additional barriers to ICH
detection.
Other approaches have been suggested for portable, prehos-
pital detection of stroke and ICH, including accelerom-
eters, electroencephalogram, microwaves, radiofrequency, and
volumetric impedance phase-shift spectroscopy (VIPS).78 A
wearable VIPS device demonstrated high sensitivity (93%) and
specificity (87%) for detecting severe ischemic and hemorrhagic
strokes.79 Unfortunately, these approaches cannot distinguish
ICH from ischemic strokes, and cannot detect bilateral ICH due
to reliance on asymmetric pathology.
www.neurosurgery-online.com
 NEUROIMAGING OF ICH

Downloaded from https://academic.oup.com/neurosurgery/article-abstract/86/5/E414/5766367 by guest on 12 July 2020

FIGURE 7. QSM is able to differentiate between ICH and calcification. A, Axial GRE image shows a small hypointense lesion within the left frontal lobe
subcortical white matter (white arrow). B, Coronal QSM image acquired through this region during the same MRI session shows a corresponding hyperintense
lesion confirming that this is an ICH with paramagnetic properties (white arrow).

TABLE. Food and Drug Administration-approved products for automated intracerebral hemorrhage detection

Product Company Overview

InfrascannerR
2000 InfraScan Handheld device that detects ICH67
VisorTM System Cerebrotech Wearable device that detects potential severe strokes (both ischemic and hemorrhagic)70
ACCIPIOTM Ix MaxQ AI (formerly MedyMatch) Program that highlights potential ICH and prioritizes cases for review
Aidoc brain package Aidoc Program that highlights potential ICH for review75
HealthICH Zebra Medical Vision Program that detects potential ICH and triages/highlights case for review

Automated ICH Detection
Rapid advancements in machine learning techniques have
prompted a number of studies to evaluate automated ICH
detection algorithms for identifying both intra- and extra-axial
ICH with varying sensitivities (81%,80 area under the curve
0.84681 to 0.9082 ) and specificities (92%).83 FDA-approved
programs are listed in the Table (A Bar, MS et al, unpublished
data, September 2018).84
Automated algorithms that detect critical findings would
facilitate triage of cases awaiting interpretation, especially
in underserved areas, thereby improving workflow and
patient outcomes.82 Utilizing a machine learning algorithm
to detect ICH reduces the time to diagnosis by 96%.81
However, barriers have prevented widespread adoption of these
techniques, including limited interinstitutional generalizability
of algorithms that were trained on limited, occasionally single-
site datasets. Furthermore, ultimate accountability for errors
generated using a machine learning algorithm remains to be
determined.
CONCLUSION
NCCT remains the mainstay for detection of acute ICH,
with MRI and DSA used as follow-up studies to identify ICH
etiology. Advancements in pre- and in-hospital imaging acqui-
sition techniques aim to provide faster ICH detection and more

NEUROSURGERY VOLUME 86 | NUMBER 5 | MAY 2020 | E421

RINDLER ET AL
accurate images that may soon change the current protocols of
ICH workup.
Disclosures
The authors have no personal, financial, or institutional interest in any of the
drugs, materials, or devices described in this article.
REFERENCES
1. Gross BA, Jankowitz BT, Friedlander RM. Cerebral intraparenchymal hemorrhage:
a review. JAMA. 2019;321(13):1295-1303.
2. Ikram MA, Wieberdink RG, Koudstaal PJ. International epidemiology of intrac-
erebral hemorrhage. Curr Atheroscler Rep. 2012;14(4):300-306.
3. Feigin VL, Krishnamurthi RV, Parmar P, et al. Update on the global burden of
ischemic and hemorrhagic stroke in 1990-2013: the GBD 2013 study. Neuroepi-
demiology. 2015;45(3):161-176.
4. Kranz PG, Malinzak MD, Amrhein TJ. Approach to imaging in patients with
spontaneous intracranial hemorrhage. Neuroimaging Clin N Am. 2018;28(3):353-
374.
5. Manno EM, Atkinson JL, Fulgham JR, Wijdicks EF. Emerging medical and
surgical management strategies in the evaluation and treatment of intracerebral
hemorrhage. Mayo Clin Proc. 2005;80(3):420-433.
6. Garibi J, Bilbao G, Pomposo I, Hostalot C. Prognostic factors in a series of 185
consecutive spontaneous supratentorial intracerebral haematomas. Br J Neurosurg.
2002;16(4):355-361.
7. Dastur CK, Yu W. Current management of spontaneous intracerebral haemor-
rhage. Stroke Vasc Neurol. 2017;2(1):21-29.
8. Salmela MB, Mortazavi S, Jagadeesan BD, et al. ACR appropriateness criteria((R))
cerebrovascular disease. J Am Coll Radiol. 2017;14(5):S34-s61.
9. New PF, Aronow S. Attenuation measurements of whole blood and blood fractions
in computed tomography. Radiology. 1976;121(3 pt 1):635-640.
10. Selariu E, Zia E, Brizzi M, Abul-Kasim K. Swirl sign in intracerebral haemorrhage:
definition, prevalence, reliability and prognostic value. BMC Neurol. 2012;12:109.
11. Xiong X, Li Q, Yang WS, et al. Comparison of swirl sign and black hole sign
in predicting early hematoma growth in patients with spontaneous intracerebral
hemorrhage. Med Sci Monit. 2018;24:567-573.
12. Barras CD, Tress BM, Christensen S, et al. Density and shape as CT predictors of
intracerebral hemorrhage growth. Stroke. 2009;40(4):1325-1331.
13. Zhang D, Chen J, Xue Q, et al. Heterogeneity signs on noncontrast computed
tomography predict hematoma expansion after intracerebral hemorrhage: a meta-
analysis. Biomed Res Int. 2018;2018:6038193.
14. Herold S, von Kummer R, Jaeger C. Follow-up of spontaneous intracerebral
haemorrhage by computed tomography. J Neurol. 1982;228(4):267-276.
15. Kidwell CS, Wintermark M. Imaging of intracranial haemorrhage. Lancet Neurol.
2008;7(3):256-267.
16. Mayer SA, Rincon F. Treatment of intracerebral haemorrhage. Lancet Neurol.
2005;4(10):662-672.
17. Flaherty ML, Haverbusch M, Sekar P, et al. Long-term mortality after intracerebral
hemorrhage. Neurology. 2006;66(8):1182-1186.
18. Fiorella D, Arthur A, Mocco J. 305 The INVEST trial: a randomized, controlled
trial to investigate the safety and efficacy of image-guided minimally invasive
endoscopic surgery with Apollo vs best medical management for supratentorial
intracerebral hemorrhage. Neurosurgery. 2016;63(suppl 1):187.
19. Hanley DF, Thompson RE, Rosenblum M, et al. Efficacy and safety of minimally
invasive surgery with thrombolysis in intracerebral haemorrhage evacuation
(MISTIE III): a randomised, controlled, open-label, blinded endpoint phase 3 trial.
Lancet. 2019;393(10175):1021-1032.
20. Labib MA, Shah M, Kassam AB, et al. The safety and feasibility of image-guided
BrainPath-mediated transsulcul hematoma evacuation: a multicenter study. Neuro-
surgery. 2017;80(4):515-524.
21. Broderick JP, Brott TG, Duldner JE, Tomsick T, Huster G. Volume of intrac-
erebral hemorrhage. A powerful and easy-to-use predictor of 30-day mortality.
Stroke. 1993;24(7):987-993.
22. Stocchetti N, Croci M, Spagnoli D, Gilardoni F, Resta F, Colombo A. Mass volume
measurement in severe head injury: accuracy and feasibility of two pragmatic
methods. J Neurol Neurosurg Psychiatry. 2000;68(1):14-17.
E422 | VOLUME 86 | NUMBER 5 | MAY 2020
23. Kothari RU, Brott T, Broderick JP, et al. The ABCs of measuring intracerebral
hemorrhage volumes. Stroke. 1996;27(8):1304-1305.
24. Hanley DF. Intraventricular hemorrhage: severity factor and treatment target in
spontaneous intracerebral hemorrhage. Stroke. 2009;40(4):1533-1538.
25. Gebel JM, Jr, Jauch EC, Brott TG, et al. Natural history of perihematomal
edema in patients with hyperacute spontaneous intracerebral hemorrhage. Stroke.
2002;33(11):2631-2635.
26. Balami JS, Buchan AM. Complications of intracerebral haemorrhage. Lancet
Neurol. 2012;11(1):101-118.
27. Cordonnier C, Klijn CJ, van Beijnum J, Al-Shahi Salman R. Radiological investi-
gation of spontaneous intracerebral hemorrhage: systematic review and trinational
survey. Stroke. 2010;41(4):685-690.
Downloaded from https://academic.oup.com/neurosurgery/article-abstract/86/5/E414/5766367 by guest on 12 July 2020
28. Davis SM, Broderick J, Hennerici M, et al. Hematoma growth is a deter-
minant of mortality and poor outcome after intracerebral hemorrhage. Neurology.
2006;66(8):1175-1181.
29. Wada R, Aviv RI, Fox AJ, et al. CT angiography “spot sign” predicts hematoma
expansion in acute intracerebral hemorrhage. Stroke. 2007;38(4):1257-1262.
30. Demchuk AM, Dowlatshahi D, Rodriguez-Luna D, et al. Prediction of haematoma
growth and outcome in patients with intracerebral haemorrhage using the CT-
angiography spot sign (PREDICT): a prospective observational study. Lancet
Neurol. 2012;11(4):307-314.
31. Macellari F, Paciaroni M, Agnelli G, Caso V. Neuroimaging in intracerebral
hemorrhage. Stroke. 2014;45(3):903-908.
32. Zhu XL, Chan MS, Poon WS. Spontaneous intracranial hemorrhage: which
patients need diagnostic cerebral angiography? A prospective study of 206 cases
and review of the literature. Stroke. 1997;28(7):1406-1409.
33. Wilson D, Ogungbemi A, Ambler G, Jones I, Werring DJ, Jager HR. Developing
an algorithm to identify patients with intracerebral haemorrhage secondary to a
macrovascular cause. Eur Stroke J. 2017;2(4):369-376.
34. Hilkens NA, van Asch CJJ, Werring DJ, et al. Predicting the presence of
macrovascular causes in non-traumatic intracerebral haemorrhage: the DIAGRAM
prediction score. J Neurol Neurosurg Psychiatry. 2018;89(7):674-679.
35. Josephson CB, White PM, Krishan A, Al-Shahi Salman R. Computed tomog-
raphy angiography or magnetic resonance angiography for detection of intracranial
vascular malformations in patients with intracerebral haemorrhage. Cochrane
Database Syst Rev. 2014;9:Cd009372.
36. Manninen AL, Isokangas JM, Karttunen A, Siniluoto T, Nieminen MT. A
comparison of radiation exposure between diagnostic CTA and DSA examinations
of cerebral and cervicocerebral vessels. AJNR Am J Neuroradiol. 2012;33(11):2038-
2042.
37. van Asch CJ, Velthuis BK, Rinkel GJ, et al. Diagnostic yield and accuracy of CT
angiography, MR angiography, and digital subtraction angiography for detection of
macrovascular causes of intracerebral haemorrhage: prospective, multicentre cohort
study. BMJ. 2015;351:h5762.
38. Broderick J, Connolly S, Feldmann E, et al. Guidelines for the management
of spontaneous intracerebral hemorrhage in adults: 2007 update: a guideline
from the American Heart Association/American Stroke Association Stroke
Council, High Blood Pressure Research Council, and the Quality of Care and
Outcomes in Research Interdisciplinary Working Group. Stroke. 2007;38(6):2001-
2023.
39. Kidwell CS, Chalela JA, Saver JL, et al. Comparison of MRI and CT for detection
of acute intracerebral hemorrhage. JAMA. 2004;292(15):1823-1830.
40. Schindlbeck KA, Santaella A, Galinovic I, et al. Spot sign in acute intrac-
erebral hemorrhage in dynamic T1-weighted magnetic resonance imaging. Stroke.
2016;47(2):417-423.
41. Khan Z, Nattanmai P, George P, Newey CR. Spot sign in acute intracerebral
hemorrhage in magnetic resonance imaging: a case report and review of the liter-
ature. Neurologist. 2018;23(3):104-107.
42. Hostettler IC, Seiffge DJ, Werring DJ. Intracerebral hemorrhage: an update on
diagnosis and treatment. Expert Rev Neurother. 2019;19(7):679-694.
43. Wardlaw JM, Smith EE, Biessels GJ, et al. Neuroimaging standards for research into
small vessel disease and its contribution to ageing and neurodegeneration. Lancet
Neurol. 2013;12(8):822-838.
44. Wahlund LO, Barkhof F, Fazekas F, et al. A new rating scale for age-related white
matter changes applicable to MRI and CT. Stroke. 2001;32(6):1318-1322.
45. Charidimou A, Imaizumi T, Moulin S, et al. Brain hemorrhage recurrence,
small vessel disease type, and cerebral microbleeds: a meta-analysis. Neurology.
2017;89(8):820-829.
www.neurosurgery-online.com

46. Atlas SW, Grossman RI, Gomori JM, et al. Hemorrhagic intracranial malignant
neoplasms: spin-echo MR imaging. Radiology. 1987;164(1):71-77.
47. Parizel PM, Makkat S, Van Miert E, Van Goethem JW, van den Hauwe L, De
Schepper AM. Intracranial hemorrhage: principles of CT and MRI interpretation.
Eur Radiol. 2001;11(9):1770-1783.
48. Blackham KA, Passalacqua MA, Sandhu GS, Gilkeson RC, Griswold MA,
Gulani V. Applications of time-resolved MR angiography. Am J Roentgenol.
2011;196(5):W613-W620.
49. Klisch J, Strecker R, Hennig J, Schumacher M. Time-resolved projection
MRA: clinical application in intracranial vascular malformations. Neuroradiology.
2000;42(2):104-107.
50. Nowinski WL, Puspitasaari F, Volkau I, Marchenko Y, Knopp MV. Comparison
of magnetic resonance angiography scans on 1.5, 3, and 7 Tesla units: a quantitative
study of 3-dimensional cerebrovasculature. J Neuroimaging. 2013;23(1):86-95.
51. Griessenauer C, Medin C, Goren O, Schirmer CM. Image-guided, minimally
invasive evacuation of intracerebral hematoma: a matched cohort study comparing
the endoscopic and tubular exoscopic systems. Cureus. 2018;10(11):e3569.
52. Hyodo A, Mizukami M, Tazawa T, Togashi O. Intraoperative use of real time ultra-
sonography applied to aneurysm surgery. Neurosurgery. 1983;13(6):642-645.
53. Imberti R, Pietrobono L, Klersy C, Gamba G, Iotti G, Cornara G. Intraoperative
intravenous administration of rFVIIa and hematoma volume after early surgery for
spontaneous intracerebral hemorrhage: a randomized prospective phase II study.
Minerva Anestesiol. 2012;78(2):168-175.
54. Schafer S, Wang A, Otake Y, et al. Intraoperative Imaging for Patient Safety and
QA: Detection of Intracranial Hemorrhage Using C-arm Cone-beam CT. Vol 8671.
Bellingham, Washington: Proc. SPIE; 2013.
55. Goren O, Monteith SJ, Hadani M, Bakon M, Harnof S. Modern intraoperative
imaging modalities for the vascular neurosurgeon treating intracerebral hemor-
rhage. Neurosurg Focus. 2013;34(5):E2.
56. Hu R, Padole A, Gupta R. Dual-energy computed tomographic applications for
differentiation of intracranial hemorrhage, calcium, and iodine. Neuroimaging Clin
N Am. 2017;27(3):401-409.
57. Hu R, Daftari Besheli L, Young J, et al. Dual-energy head CT enables accurate
distinction of intraparenchymal hemorrhage from calcification in emergency
department patients. Radiology. 2016;280(1):177-183.
58. Almqvist H, Holmin S, Mazya MV. Dual energy CT after stroke thrombectomy
alters assessment of hemorrhagic complications. Neurology. 2019;93(11):e1068-
e1075.
59. Guo Y, Ou SX, Qian M, Zeng X, Li B. Dual-energy CT angiography for
the diagnosis of intracranial dural arteriovenous fistula. Int J Clin Exp Med.
2015;8(5):7802-7808.
60. Ni QQ, Tang CX, Zhao YE, et al. Single phase dual-energy CT angiography:
one-stop-shop tool for evaluating aneurysmal subarachnoid hemorrhage. Sci Rep.
2016;6:26704.
61. Tan CO, Lam S, Kuppens D, et al. Spot and diffuse signs: quantitative markers of
intracranial hematoma expansion at dual-energy CT. Radiology. 2019;290(1):179-
186.
62. Yedavalli V, Sammet S. Contrast extravasation versus hemorrhage after
thrombectomy in patients with acute stroke. J Neuroimaging. 2017;27(6):570-576.
63. Goo HW, Goo JM. Dual-energy CT: new horizon in medical imaging. Korean J
Radiol. 2017;18(4):555-569.
64. Gottumukkala RV, Kalra MK, Tabari A, Otrakji A, Gee MS. Advanced CT
techniques for decreasing radiation dose, reducing sedation requirements, and
optimizing image quality in children. Radiographics. 2019;39(3):709-726.
65. Pruessmann KP, Weiger M, Scheidegger MB, Boesiger P. SENSE: sensitivity
encoding for fast MRI. Magn Reson Med. 1999;42(5):952-962.
66. Griswold MA, Jakob PM, Heidemann RM, et al. Generalized autocalibrating
partially parallel acquisitions (GRAPPA). Magn Reson Med. 2002;47(6):1202-
1210.
NEUROSURGERY
NEUROIMAGING OF ICH
67. Lustig M, Donoho D, Pauly JM. Sparse MRI: The application of compressed
sensing for rapid MR imaging. Magn Reson Med. 2007;58(6):1182-1195.
68. Haacke EM, Mittal S, Wu Z, Neelavalli J, Cheng YC. Susceptibility-weighted
imaging: technical aspects and clinical applications, part 1. AJNR Am J Neuroradiol.
2009;30(1):19-30.
69. Mittal S, Wu Z, Neelavalli J, Haacke EM. Susceptibility-weighted imaging:
technical aspects and clinical applications, part 2. AJNR Am J Neuroradiol.
2009;30(2):232-252.
70. Halefoglu AM, Yousem DM. Susceptibility weighted imaging: clinical applications
and future directions. World J Radiol. 2018;10(4):30-45.
71. Yamada N, Imakita S, Sakuma T, Takamiya M. Intracranial calcification on
gradient-echo phase image: depiction of diamagnetic susceptibility. Radiology.
Downloaded from https://academic.oup.com/neurosurgery/article-abstract/86/5/E414/5766367 by guest on 12 July 2020
1996;198(1):171-178.
72. de Rochefort L, Liu T, Kressler B, et al. Quantitative susceptibility map recon-
struction from MR phase data using bayesian regularization: validation and appli-
cation to brain imaging. Magn Reson Med. 2010;63(1):194-206.
73. Eskreis-Winkler S, Zhang Y, Zhang J, et al. The clinical utility of QSM:
disease diagnosis, medical management, and surgical planning. NMR Biomed.
2017;30(4):e3668.
74. Wang S, Lou M, Liu T, Cui D, Chen X, Wang Y. Hematoma volume
measurement in gradient echo MRI using quantitative susceptibility mapping.
Stroke. 2013;44(8):2315-2317.
75. Marques JP, Simonis FFJ, Webb AG. Low-field MRI: an MR physics perspective.
J Magn Reson Imaging. 2019;49(6):1528-1542.
76. Ayaz H, Izzetoglu M, Izzetoglu K, Onaral B, Ben Dor B. Early diagnosis of
traumatic intracranial hematomas. J Biomed Opt. 2019;24(5):1-10.
77. Trehan V, Maheshwari V, Kulkarni SV, Kapoor S, Gupta A. Evaluation of near
infrared spectroscopy as screening tool for detecting intracranial hematomas in
patients with traumatic brain injury. Med J Armed Forces India. 2018;74(2):139-
142.
78. Walsh KB. Non-invasive sensor technology for prehospital stroke diagnosis: current
status and future directions. Int J Stroke. 2019;14(6):592-602.
79. Kellner CP, Sauvageau E, Snyder KV, et al. The VITAL study and overall pooled
analysis with the VIPS non-invasive stroke detection device. J Neurointerv Surg.
2018;10(11):1079-1084.
80. Majumdar A, Brattain L, Telfer B, Farris C, Scalera J. Detecting intracranial hemor-
rhage with deep learning. Conf Proc IEEE Eng Med Biol Soc. 2018;2018:583-
587.
81. Arbabshirani MR, Fornwalt BK, Mongelluzzo GJ, et al. Advanced machine
learning in action: identification of intracranial hemorrhage on computed tomog-
raphy scans of the head with clinical workflow integration. NPJ Digital Med.
2018;1:9.
82. Chilamkurthy S, Ghosh R, Tanamala S, et al. Deep learning algorithms for
detection of critical findings in head CT scans: a retrospective study. Lancet.
2018;392(10162):2388-2396.
83. Ye H, Gao F, Yin Y, et al. Precise diagnosis of intracranial hemorrhage and subtypes
using a three-dimensional joint convolutional and recurrent neural network. Eur
Radiol. 2019 29;(11):6191-6201.
84. Ojeda P, Zawaideh M, Mossa-Basha M, Haynor D. The utility of deep learning:
evaluation of a convolutional neural network for detection of intracranial bleeds
on non-contrast head computed tomography studies. Paper presented at: Medical
Imaging 2019: Image Processing; Bellingham, Washington: Proc. SPIE.
Acknowledgments
The authors would like to acknowledge Susan D. O’Brien, MS, RPh, for her
assistance in literature review and manuscript preparation, and Ranliang Hu, MD,
for supplying the cases used in Figures 4, 5, and 7.
VOLUME 86 | NUMBER 5 | MAY 2020 | E423