1
|
Werdiger F, Parsons MW, Visser M, Levi C, Spratt N, Kleinig T, Lin L, Bivard A. Machine learning segmentation of core and penumbra from acute stroke CT perfusion data. Front Neurol 2023; 14:1098562. [PMID: 36908587 PMCID: PMC9995438 DOI: 10.3389/fneur.2023.1098562] [Citation(s) in RCA: 4] [Impact Index Per Article: 4.0] [Reference Citation Analysis] [Abstract] [Key Words] [Track Full Text] [Journal Information] [Subscribe] [Scholar Register] [Received: 11/20/2022] [Accepted: 02/02/2023] [Indexed: 02/25/2023] Open
Abstract
Introduction Computed tomography perfusion (CTP) imaging is widely used in cases of suspected acute ischemic stroke to positively identify ischemia and assess suitability for treatment through identification of reversible and irreversible tissue injury. Traditionally, this has been done via setting single perfusion thresholds on two or four CTP parameter maps. We present an alternative model for the estimation of tissue fate using multiple perfusion measures simultaneously. Methods We used machine learning (ML) models based on four different algorithms, combining four CTP measures (cerebral blood flow, cerebral blood volume, mean transit time and delay time) plus 3D-neighborhood (patch) analysis to predict the acute ischemic core and perfusion lesion volumes. The model was developed using 86 patient images, and then tested further on 22 images. Results XGBoost was the highest-performing algorithm. With standard threshold-based core and penumbra measures as the reference, the model demonstrated moderate agreement in segmenting core and penumbra on test images. Dice similarity coefficients for core and penumbra were 0.38 ± 0.26 and 0.50 ± 0.21, respectively, demonstrating moderate agreement. Skull-related image artefacts contributed to lower accuracy. Discussion Further development may enable us to move beyond the current overly simplistic core and penumbra definitions using single thresholds where a single error or artefact may lead to substantial error.
Collapse
Affiliation(s)
- Freda Werdiger
- Melbourne Brain Centre, Department of Neurology, The Royal Melbourne Hospital, Melbourne, VIC, Australia.,Department of Medicine, University of Melbourne, Melbourne, VIC, Australia
| | - Mark W Parsons
- Southwestern Sydney Clinical School, University of New South Wales, Sydney, NSW, Australia.,Department of Neurology, Liverpool Hospital, Liverpool, NSW, Australia.,Ingham Institute for Applied Medical Research, Liverpool, NSW, Australia
| | - Milanka Visser
- Melbourne Brain Centre, Department of Neurology, The Royal Melbourne Hospital, Melbourne, VIC, Australia.,Department of Medicine, University of Melbourne, Melbourne, VIC, Australia
| | - Christopher Levi
- Hunter Medical Research Institution, University of Newcastle, Newcastle, NSW, Australia.,Department of Neurology, John Hunter Hospital, University of Newcastle, Newcastle, NSW, Australia
| | - Neil Spratt
- Hunter Medical Research Institution, University of Newcastle, Newcastle, NSW, Australia.,Department of Neurology, John Hunter Hospital, University of Newcastle, Newcastle, NSW, Australia
| | - Tim Kleinig
- Department of Neurology, Royal Adelaide Hospital, Adelaide, SA, Australia
| | - Longting Lin
- Hunter Medical Research Institution, University of Newcastle, Newcastle, NSW, Australia.,Department of Neurology, John Hunter Hospital, University of Newcastle, Newcastle, NSW, Australia
| | - Andrew Bivard
- Melbourne Brain Centre, Department of Neurology, The Royal Melbourne Hospital, Melbourne, VIC, Australia.,Department of Medicine, University of Melbourne, Melbourne, VIC, Australia
| |
Collapse
|
2
|
Automated CT Perfusion Imaging to Aid in the Selection of Patients With Acute Ischemic Stroke for Mechanical Thrombectomy: A Health Technology Assessment. ONTARIO HEALTH TECHNOLOGY ASSESSMENT SERIES 2020; 20:1-87. [PMID: 33240454 PMCID: PMC7668535] [Citation(s) in RCA: 0] [Impact Index Per Article: 0] [Reference Citation Analysis] [Abstract] [MESH Headings] [Subscribe] [Scholar Register] [Indexed: 06/11/2023]
Abstract
BACKGROUND Stroke is a sudden interruption in the blood supply to a part of the brain, causing loss of neurological function. It is the third leading cause of death in Canada and affects mainly older people. In the acute setting, neuroimaging is integral to stroke evaluation and decision-making. The neuroimaging results guide patient selection for mechanical thrombectomy. Using automated image processing techniques facilitates efficient review of this information and communication between centres. We conducted a health technology assessment of automated CT perfusion imaging as a tool for selecting stroke patients with anterior circulation occlusion for mechanical thrombectomy. This assessment included an evaluation of clinical effectiveness, cost-effectiveness, and the budget impact of publicly funding automated CT perfusion imaging. METHODS We performed a systematic literature search of the clinical evidence. We assessed the risk of bias of each study using QUADAS-2 or the Cochrane risk-of-bias tool, and the quality of the body of evidence according to the Grading of Recommendations Assessment, Development, and Evaluation (GRADE) Working Group criteria. We performed a systematic economic literature search and approximated cost-effectiveness based on previous analyses. We also analyzed the budget impact of publicly funding automated CT perfusion imaging to evaluate people with acute ischemic stroke in Ontario. RESULTS Automated CT perfusion imaging had a sensitivity of 84% for identifying the infarct core (dead tissue that does not recover despite restoring blood flow with mechanical thrombectomy), compared with diffusion-weighted MRI imaging at 24 hours. One study reported that 7% of patients were misclassified with respect to eligibility for mechanical thrombectomy (either erroneously classified as eligible or erroneously classified non-eligible). Two randomized controlled trials (DEFUSE 3 and DAWN) demonstrated the efficacy of mechanical thrombectomy up to 24 hours after stroke onset, with patient selection guided by automated CT perfusion imaging. These data showed that a significantly higher proportion of patients in the mechanical thrombectomy group achieved functional independence compared with the standard care group (DEFUSE 3: risk ratio: 2.67 [95% confidence interval 1.60-4.48]; DAWN: adjusted rate difference: 33% [95% credible interval 21%-44%]; GRADE: Moderate).A previous health technology assessment in stroke patients presenting at 0 to 6 hours after stroke symptom onset and the results from recent randomized controlled trials for patients presenting at 6 to 24 hours informed the evaluation of cost-effectiveness. Mechanical thrombectomy informed by automated CT perfusion imaging to assess eligibility is likely to be cost-effective for patients presenting at 6 to 24 hours after stroke symptom onset. The annual budget impact of publicly funding automated CT perfusion imaging in Ontario over the next 5 years would be $1.3 million in year 1 and $0.9 million each year thereafter. Some of the costs of automated CT perfusion imaging could be offset by avoiding unnecessary patient transfers between hospitals. CONCLUSIONS Automated CT perfusion imaging has an acceptable sensitivity and specificity for detecting brain areas that have been affected by stroke. In patients selected for mechanical thrombectomy using automated CT perfusion imaging, there was significant improvement in functional independence. Mechanical thrombectomy informed by automated CT perfusion imaging is likely to be cost-effective. We estimate that publicly funding automated CT perfusion imaging in Ontario would result in additional costs of $1.3 million in year 1 and $0.9 million per year thereafter.
Collapse
|
3
|
Bathla G, Ortega-Gutierrez S, Klotz E, Juergens M, Zevallos CB, Ansari S, Ward CE, Policeni B, Samaniego E, Derdeyn C. Comparing the outcomes of two independent computed tomography perfusion softwares and their impact on therapeutic decisions in acute ischemic stroke. J Neurointerv Surg 2020; 12:1028-1032. [PMID: 32424007 DOI: 10.1136/neurintsurg-2020-015827] [Citation(s) in RCA: 8] [Impact Index Per Article: 2.0] [Reference Citation Analysis] [Abstract] [Key Words] [MESH Headings] [Track Full Text] [Journal Information] [Subscribe] [Scholar Register] [Received: 01/21/2020] [Revised: 04/07/2020] [Accepted: 04/08/2020] [Indexed: 11/04/2022]
Abstract
BACKGROUND To compare the computed tomography perfusion (CTP) outcomes derived from two commercial CTP processing software and evaluate their concordance in terms of eligibility for mechanical thrombectomy (MT) in acute ischemic stroke (AIS), based on DEFUSE III criteria. METHODS A total of 118 patients (62 patients in the MT group and 56 patients in the non-MT (NMT) group) were included. Volumetric perfusion outputs were compared between Syngo.via (package A) and RAPID (package B). Influence on proceeding or not-proceeding with MT was based on DEFUSE III imaging eligibility criteria. RESULTS Median core infarct/hypoperfusion volumes were 12.3/126 mL in the MT group and 7.7/29.3 ml in the NMT group with package A and 10.5/138 mL and 1.9/24.5 mL with package B, respectively. In the MT group (n=62), concordant perfusion results in terms of patient triage were noted in all but two cases. Of these, one patient would not have qualified (low ASPECTS), while the other qualified based on package A results. For the NMT group (n=56), there was discordance in terms of MT eligibility in seven cases. However, none of these patients qualified for MT based on DEFUSE III criteria. CONCLUSIONS Both perfusion softwares showed high concordance in correctly triaging patients in the MT versus NMT groups (110/118, 93.2%), which further improved when all DEFUSE III imaging criteria were considered (117/118, 99.1%). The core/hypoperfusion volumes in the NMT group and core infarct volumes in the MT groups were comparable. The hypoperfusion volumes in the MT group varied slightly but did not affect triage between groups.
Collapse
Affiliation(s)
- Girish Bathla
- Radiology, University of Iowa Hospitals and Clinics, Iowa City, Iowa, USA
| | | | - Ernst Klotz
- SHS DI CT R&D CTC SA, Siemens Healthineers, Forchheim, Germany
| | - Markus Juergens
- SHS DI CT R&D CTC SA, Siemens Healthineers, Forchheim, Germany
| | | | | | - Caitlin E Ward
- Department of Biostatistics, University of Iowa, Iowa City, Iowa, USA
| | - Bruno Policeni
- Radiology, University of Iowa Hospitals and Clinics, Iowa City, Iowa, USA
| | - Edgar Samaniego
- Neurology, Division of Neurointerventional Surgery, University of Iowa, Iowa City, Iowa, USA
| | - Colin Derdeyn
- Radiology, Division of Neurointerventional Radiology, University of Iowa Hospitals and Clinics, Iowa City, Iowa, USA
| |
Collapse
|
4
|
Siegler JE, Messé SR, Sucharew H, Kasner SE, Mehta T, Arora N, Starosciak AK, De Los Rios La Rosa F, Barnhill NR, Mistry AM, Patel K, Assad S, Tarboosh A, Dakay K, Wagner J, Bennett A, Jagadeesan B, Streib C, Weber SA, Chitale R, Volpi JJ, Mayer SA, Yaghi S, Jayaraman MV, Khatri P, Mistry EA. Noncontrast CT versus Perfusion-Based Core Estimation in Large Vessel Occlusion: The Blood Pressure after Endovascular Stroke Therapy Study. J Neuroimaging 2019; 30:219-226. [PMID: 31762108 DOI: 10.1111/jon.12682] [Citation(s) in RCA: 11] [Impact Index Per Article: 2.2] [Reference Citation Analysis] [Abstract] [Key Words] [Track Full Text] [Journal Information] [Subscribe] [Scholar Register] [Received: 09/22/2019] [Revised: 11/08/2019] [Accepted: 11/11/2019] [Indexed: 01/01/2023] Open
Abstract
BACKGROUND AND PURPOSE The 2018 AHA guidelines recommend perfusion imaging to select patients with acute large vessel occlusion (LVO) for thrombectomy in the extended window. However, the relationship between noncontrast CT and CT perfusion imaging has not been sufficiently characterized >6 hours after last known normal (LKN). METHODS From a multicenter prospective cohort of consecutive adults who underwent thrombectomy for anterior LVO 0-24 hours after LKN, we correlated baseline core volume (rCBF < 30%) and the Alberta Stroke Program Early CT Scale (ASPECTS) score. We compared perfusion findings between patients with an unfavorable ASPECTS (<6) against those with a favorable ASPECTS (≥6), and assessed findings over time. RESULTS Of 485 enrolled patients, 177 met inclusion criteria (median age: 69 years, interquartile range [IQR: 57-81], 49% female, median ASPECTS 8 [IQR: 6-9], median core 10 cc [IQR: 0-30]). ASPECTS and core volume moderately correlated (r = -.37). A 0 cc core was observed in 54 (31%) patients, 70% of whom had ASPECTS <10. Of the 28 patients with ASPECTS <6, 3 (11%) had a 0 cc core. After adjustment for age and stroke severity, there was a lower ASPECTS for every 1 hour delay from LKN (cOR: 0.95, 95% confidence of interval [CI]: 0.91-1.00, P = .04). There was no difference in core (P = .51) or penumbra volumes (P = .87) across patients over time. CONCLUSIONS In this multicenter prospective cohort of patients who underwent thrombectomy, one-third of patients had normal CTP core volumes despite nearly three quarters of patients showing ischemic changes on CT. This finding emphasizes the need to carefully assess both noncontrast and perfusion imaging when considering thrombectomy eligibility.
Collapse
Affiliation(s)
- James E Siegler
- Department of Neurology, Hospital of the University of Pennsylvania, Philadelphia, PA
| | - Steven R Messé
- Department of Neurology, Hospital of the University of Pennsylvania, Philadelphia, PA
| | - Heidi Sucharew
- Cincinnati Children's Hospital Medical Center, University of Cincinnati, Cincinnati, OH
| | - Scott E Kasner
- Department of Neurology, Hospital of the University of Pennsylvania, Philadelphia, PA
| | - Tapan Mehta
- Department of Neurology, University of Minnesota Medical Center, Minneapolis, MN.,Department of Neurology, Fairview Southdale Hospital, Minneapolis, MN.,Department of Neurology, Hennepin County Medical Center, Minneapolis, MN
| | - Niraj Arora
- Department of Neurology, Jackson Memorial Hospital, Miami, FL
| | | | | | - Natasha R Barnhill
- Department of Neurology, Oregon Health and Science University, Portland, OR
| | | | - Kishan Patel
- Department of Neurology, Houston Methodist Medical Center, Houston, TX
| | - Salman Assad
- Department of Neurology, Henry Ford Health System, Detroit, MI
| | - Amjad Tarboosh
- Department of Neurology, Henry Ford Health System, Detroit, MI
| | - Katarina Dakay
- Department of Neurology, Brown University, Providence, RI
| | - Jeff Wagner
- Department of Neurology, Blue Sky Neurology, Englewood, CO
| | - Alicia Bennett
- Department of Neurology, Blue Sky Neurology, Englewood, CO
| | - Bharathi Jagadeesan
- Department of Radiology, University of Minnesota Medical Center, Minneapolis, MN
| | - Christopher Streib
- Department of Neurology, University of Minnesota Medical Center, Minneapolis, MN.,Department of Neurology, Fairview Southdale Hospital, Minneapolis, MN
| | - Stewart A Weber
- Department of Neurology, Oregon Health and Science University, Portland, OR
| | - Rohan Chitale
- Department of Neurosurgery, Vanderbilt University Medical Center, Nashville, TN
| | - John J Volpi
- Department of Neurology, Houston Methodist Medical Center, Houston, TX
| | - Stephan A Mayer
- Department of Neurology, Henry Ford Health System, Detroit, MI
| | - Shadi Yaghi
- Department of Neurology, Brown University, Providence, RI
| | - Mahesh V Jayaraman
- Department of Neurology, Brown University, Providence, RI.,Department of Diagnostic Imaging, Brown University, Providence, RI.,Department of Neurosurgery, Brown University, Providence, RI
| | - Pooja Khatri
- Department of Neurology, University of Cincinnati, Cincinnati, OH
| | - Eva A Mistry
- Department of Neurosurgery, Vanderbilt University Medical Center, Nashville, TN
| |
Collapse
|