Available via license: CC BY 4.0
Content may be subject to copyright.
Muhannad Sabieleish*, Maximilian Thormann, Jonathan Metzler, Axel Boese, Michael
Friebe, Anastasios Mpotsaris, and Daniel Behme
Image processing-based mTICI grading after
endovascular treatment for acute ischemic
stroke
Abstract: Introduction: The grade of reperfusion after
endovascular treatment of ischemic stroke e.g. mechanical
thrombectomy is determined based on the mTICI score. The
mTICI score shows significant interrater variability; it is
usually biased towards better reperfusion results if self-
assessed by the operator. We therefore developed a semi-
automated image processing technique for assessing and
evaluating the degree of reperfusion independently, resulting
in a more objective mTICI score. Methods: Fifty angiography
datasets of patients who were treated with mechanical
thrombectomy for middle cerebral artery (MCA) occlusion
were selected from our database. Image datasets were
standardized by adjustment of field of view and orientation.
Based on pixel intensity features, the internal carotid artery
(ICA) curve was detected automatically and used as a starting
point for identifying the target downstream territory (TDT) of
the MCA on the DSA series. Furthermore, a grid with
predefined dimensions was used to divide the TDT into check-
zones and be classified as perfused or unperfused. Results:
The algorithm detected the TDT and classified each zone of
the grid as perfused or unperfused. Lastly, the percentage of
the perfused area in the TDT was calculated for each patient
and compared to the grading of experienced clinical users.
Conclusion: A semi-automatic image-processing workflow
was developed to evaluate perfusion rate based on
angiographic images. The approach can be used for the
objective calculation of the mTICI score. The semi-automatic
grading is currently feasible for MCA occlusion but can be
extended for other brain territories. The work shows a starting
point for a machine learning approach to achieve a fully
automated system that can evaluate and give an accurate
mTICI score to become a common AI-based grading standard
in the coming near future.
Keywords: Stroke, image processing, automatic mTICI
grading, endovascular treatment, perfusion.
https://doi.org/10.1515/cdbme-2021-2060
1 Introduction
Stroke is known as the second leading cause of death
worldwide [1, 2], 87% of the strokes are classified as ischemic
stroke, which are generated by the cerebral artery occlusion.
Moreover, 50% of the ischemic strokes are thrombotic
ischemic strokes [2]. Thrombotic ischemic strokes are more
common for elderly people, in which it happens for 80% of the
patients without any symptoms before they occur [2].
Nowadays, acute ischemic strokes are treated by endovascular
therapy [1, 3, 4, 9]. Various grading scales were invented to
evaluate the endovascular therapy outcomes, one of the most
used grading scales is the Thrombolysis in Cerebral Infarction
(TICI) score, which was modified (mTICI) and lastly extended
(eTICI). The TICI score is based on the final digital
subtraction angiography (DSA) images after endovascular
treatment. The score reflects the percentage of the perfused
treated area in the brain. The mTICI score is defined as:
mTICI=0; no perfusion in the TDT, mTICI=1; minimal brain
tissue perfusion in the TDT, mTICI=2a; lower than 50%
perfusion in the TDT, mTICI=2b; more than 50% perfusion in
the TDT, mTICI=2c; nearly complete perfusion in the TDT,
and mTICI=3; complete perfusion in the TDT [1, 3]. However,
______
*Corresponding author: Muhannad Sabieleish: University
Hospital Magdeburg, Department of Neuroradiology + INKA
HealthTec Innovation Laboratory, Otto-von-Guericke-University,
Leipziger Str. 44, Magdeburg, Germany, e-mail:
muhannad.sabieleish@gmail.com
2nd Author Maximilian Thormann, 3rd Author Jonathan
Metzler: University Hospital Magdeburg,, Otto-von-Guericke-
University , Leipziger Str. 44, Magdeburg, Germany, Otto-von-
Guericke-University, Leipziger Str. 44, Magdeburg, Germany
4th Author Axel Boese: INKA HealthTec Innovation Laboratory,
Otto-von-Guericke-University Magdeburg + MEDICS GmbH,
Magdeburg, Germany
5th Author Michael Friebe: INKA HealthTec Innovation
Laboratory, Otto-von-Guericke-
University, Magdeburg, Germany +
IDTM GmbH, Recklinghausen, Germany.
6th Author Anastasios Mpotsaris: University Hospital
Magdeburg, Department of Neuroradiology, Magdeburg, Germany
7th Author Daniel Behme: University Hospital Magdeburg,
Department of Neuroradiology, Magdeburg, Germany
DE GRUYTER Current Directions in Biomedical Engineering 2021;7(2): 235-238
Open Access. © 2021 The Author(s), published by De Gruyter. This work is licensed under the Creative Commons Attribution 4.0 International License.
235
the TICI score is facing several limitations; several studies
indicated the need of increasing the resolution for a more
precise treatment assessment [5-7], another limitation is
overestimation of TICI gradings by operators during
endovascular therapy when compared to core-lab raters [8].
Lastly, when self-assessed by the operator, the mTICI score
shows high interrater variability; it is frequently biased
towards better reperfusion results.
This work aims to develop a semi-automated image
processing technique that evaluates the percentage of
reperfusion in the TDT after an endovascular therapy
independently, resulting in a more proper mTICI score.
2 Methods
The following flow chart (Figure 1) represents the main
steps used in the semi-automated image processing technique
for the reperfusion assessment.
2.1 Dataset preparation
For this work, a dataset of fifty DSA series of treated
patients with mechanical thrombectomy for MCA occlusion
were collected from our database. The datasets were converted
to binary format by automatic thresholding, in which all values
that are equal or greater than a threshold are set to one and
indicate that it belongs to the contrast agent, in other words it
means that the area is perfused. The threshold was selected
above the maximum grey value of the DSA background. Zero
values of the binary format belong to the unperfused. An
example of a binary format image can be seen in figure 2.
Furthermore, the field of view was reduced to the brain area
only in order to have comparable ROI of the lateral image.
Lastly, two frames from each DSA series were automatically
detected to be used for the following steps: the first frame is
the first fill image which shows the frame with internal carotid
artery only (illustrated in Figure 2a), the second detected frame
is capillary fill phase frame which shows the spread of contrast
in soft tissue (shown in Figure 2b).
2.2 Reference point detection
A reference point is needed to be used as a starting point for
the ROI which represents the TDT. For that, the decision was
to use the ICA curve as a reference point. The ICA curve was
automatically identified by filtering the first fill image using a
median filter, then skeletonization and pruning were applied
and the most far point representing the middle of the ICA
curve was detected and chosen to be the reference point. The
red spot in Figure 2a shows the detected reference point.
2.3 Region of interest definition and
classification
Starting from the detected reference point, the ROI was
identified with a 45 degrees angle until the last intensity values
belonging to the brain, as illustrated in Figure 3. Furthermore,
grid based check zones with a size of 20*20 pixels were
applied to the ROI. The grid was used to classify each zone as
perfused or unperfused. A zone is considered as perfused if the
number of ones in the zone is more than 25% of its total
number of pixels. We chose this 25% after analysis of the
arterial phase of several DSA images. The average of pixels
belonging to arteries in the brain compared to pixels belonging
to Background is less than 25%. After classification the
percentage of perfusion was calculated as in the equation
below:
Figure 1: Flowchart of the semi-
automatic image processing
technique.
Figure 2: Binary format images of a)
The first “fill” frame and the
detected reference point (in red), b) The capillary fill phase image.
Figure 3:
Grid for “perfused” (blue) and “unperfused” (red)
classification.
236
𝑷𝒆𝒓𝒄 = # 𝒐𝒇 𝒑𝒆𝒓𝒇𝒖𝒔𝒆𝒅 𝒛𝒐𝒏𝒆𝒔
𝑻𝒐𝒕𝒂𝒍 # 𝒐𝒇 𝒛𝒐𝒏𝒆𝒔 (1)
To be able to evaluate our image processing algorithm, two
experienced neuroradiologists graded the dataset blindly
(Without knowing the algorithm score) and independently
(Without knowing the score of each other). Lastly, the semi-
automatic image processing technique mTICI scores were
compared to the neuroradiologists scores.
3 Results and discussion
The results showed the ability of the algorithm to detect the
TDT and to classify the zones as perfused and unperfused.
Table 1 below shows the results of the evaluation of fifty cases
using the algorithm in comparison to the evaluation of two
experienced neuroradiologists.
Table 1: The image processing algorithm results for the mTICI
score in comparison to two experienced neuroradiologists scores.
Table 1 shows, that the neuroradiologists gave the same
score for only 66% of the cases, 75% of them received the
same score from the developed algorithm. Moreover, the
algorithm's score matched with at least one neuroradiologist in
82% of the total cases. However, 18% of the patient data
received a different score from the algorithm than the
neuroradiologists given scores. This can be due to different
factors such as; high noisy images (as shown in Figure 4),
wrong reference point detection (as illustrated in Figure 5), and
large ROI detection (as shown in Figure 6).
Noisy images are generated mostly when a motion occurs
during the acquisition, the reference point can be wrongly
detected when the DSA series frame rate is slow, in which an
image without the capillary fill cannot be detected.
Nevertheless, large ROI can be detected when the image is not
standardized perfectly because of big orientation.
% of cases received the same score from
both neuroradiologists
33/50 (66 %
% of cases received different scores from
the neuroradiologists
17/50 (34%
% of the algorithm scores matched to both
neuroradiologists score
25/50 (50%
% of the algorithm scores matched to at
least one neuroradiologist
41/50 (82%
% of the algorithm scores which did not
match to the neuroradiologists score
9/50 (18%
Figure 4: Wrong detection for the ICA curve (Reference point).
Figure 5:
Motion during acquiring the DSA series leading to a noisy
image.
Figure 6:
Large detected TDT; the detected ROI is the MCA
and the PCA regions.
237
4 Conclusion
A semi-automatic image-processing technique was developed
for the evaluation of reperfusion rate based on DSA series. The
technique is meant to be used for the mTICI scoring. For that,
fifty cases suffering from MCA occlusion were used and
evaluated by the developed algorithm and two experienced
neuroradiologists after EVT. The resulting scores from the
algorithm were compared to the given scores from both
neuroradiologists. The results show a 75 % agreement between
both neuroradiologists and image-processing-based algorithm
and 82% agreement between the algorithm and at least one
neuroradiologist. Lastly, in 18% of the cases the algorithm’s
score did not match with any of the neuroradiologists' scores.
The promising results show that the semi-automatic grading is
feasible for the MCA occlusions and can be applied for other
brain territories. Moreover, the work can be extended and used
as a starting point for a machine learning approach for a fully
automated approach with more accurate mTICI score. The
approach can become a common AI-based grading routine in
the near future.
Author Statement
Research funding: The author state no funding involved.
Conflict of interest: Authors state no conflict of interest.
Informed consent: Informed consent has been obtained from
all individuals included in this study. Ethical approval: The
research related to human use complies with all the relevant
national regulations, institutional policies and was performed
in accordance with the tenets of the Helsinki Declaration, and
has been approved by the authors’ institutional review board
or equivalent committee.
References
[1] Su, Ruisheng & Cornelissen, Sandra & Sluijs, Matthijs & Van
Es, Adriaan & Zwam, Wim & Dippel, Diederik & Lycklama,
Geert & Doormaal, Pieter & Niessen, W.J. & Lugt, Aad &
Walsum, Theo. (2020. autoTICI: Automatic Brain Tissue
Reperfusion Scoring on 2D DSA Images of Acute Ischemic
Stroke Patients.
[2] Chugh C. Acute Ischemic Stroke: Management Approach.
Indian J Crit Care Med. 2019;23(Suppl 2:S140-S146.
doi:10.5005/jp-journals-10071-23192.
[3] Prasetya, Haryadi & Ramos, Lucas & Epema, Thabiso &
Treurniet, Kilian & Emmer, Bart & Wijngaard, Ido & Zhang,
Guang & Kappelhof, Manon & Berkhemer, Olvert & Yoo,
Albert & Roos, Yvo & Oostenbrugge, Robert & Dippel,
Diederik & Zwam, Willem & Lugt, Aad & Mol, Bas & Majoie,
Charles & Vanbavel, Ed & Marquering, Henk. (2020. qTICI:
Quantitative assessment of brain tissue reperfusion on digital
subtraction angiograms of acute ischemic stroke patients.
International Journal of Stroke. 16.
10.1177/1747493020909632.
[4] Musuka, Tapuwa & Wilton, Stephen & Traboulsi, Mouhieddin
& Hill, Michael. (2015. Diagnosis and management of acute
ischemic stroke: Speed is critical. Canadian Medical
Association Journal. 187. 10.1503/cmaj.140355.
[5] M. G. Lansberg, M. Straka, S. Kemp, M. Mlynash, L. R.
Wechsler, T. G. Jovin, M. J. Wilder, H. L. Lutsep, T. J.
Czartoski, R. A. Bernstein et al., “Mri profile and response to
endovascular reperfusion after stroke (defuse 2: a
prospective cohort study,” The Lancet Neurology, vol. 11, no.
10, pp. 860–867, 2012.
[6] T. Tomsick, J. Broderick, J. Carrozella, P. Khatri, M. Hill, Y.
Palesch, J. Khoury et al., “Revascularization results in the
interventional management of stroke ii trial,” American
Journal of Neuroradiology, vol. 29, no. 3, pp. 582–587, 2008.
[7] S. H. Suh, H. J. Cloft, J. E. Fugate, A. A. Rabinstein, D. S.
Liebeskind, and D. F. Kallmes, “Clarifying differences among
thrombolysis in cerebral infarction scale variants: is the artery
half open or half closed?” Stroke, vol. 44, no. 4, pp. 1166–
1168, 2013.
[8] G. Zhang, K. M. Treurniet, I. G. Jansen, B. J. Emmer, R. van
den Berg, H. A. Marquering, M. Uyttenboogaart, S. F.
Jenniskens, Y. B. Roos, P. J. van Doormaal et al., “Operator
versus core lab adjudication of reperfusion after
endovascular treatment of acute ischemic stroke,” Stroke,
vol. 49, no. 10, pp. 2376–2382, 2018.
[9] Shiraz Bhurwani, Mohammad Mahdi & Snyder, Kenneth &
Waqas, Muhammad & Mokin, Maxim & Rava, Ryan &
Podgorsak, Alexander & Chin, Felix & Davies, Jason & Levy,
Elad & Siddiqui, Adnan & Ionita, Ciprian. (2021. Use of
quantitative angiographic methods with a data-driven model
to evaluate reperfusion status (mTICI during thrombectomy.
Neuroradiology. 10.1007/s00234-020-02598-3.
238
