Radiologist-Level Stroke Classification on Non-contrast CT Scans with Deep U-Net

Avetisian, Manvel; Kokh, Vladimir; Tuzhilin, Alexander; Umerenkov, Dmitry

doi:10.1007/978-3-030-32248-9_91

Cited by 17 publications

(5 citation statements)

References 18 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Based on deep learning methods, with the development of deep learning in image segmentation [ 22 ], many studies have also applied it to the segmentation task of intracranial hemorrhage. Manvel et al [ 23 ] combined improved U-Net [ 24 ] and a dual-path network [ 25 ], which segmented the intracranial hemorrhage area from three perspectives and finally merged multiple separate training networks to predict the intracranial hemorrhage area. Islam et al [ 26 ] extract output feature maps of VGG network and combine them into supercolumn features for intracranial hemorrhage segmentation.…”

Section: Related Workmentioning

confidence: 99%

Deep Learning-Based Detection and Diagnosis of Subarachnoid Hemorrhage

Gou

2021

Journal of Healthcare Engineering

View full text Add to dashboard Cite

Subarachnoid hemorrhage (SAH) is one of the critical and severe neurological diseases with high morbidity and mortality. Head computed tomography (CT) is among the preferred methods for the diagnosis of SAH, which is confirmed by CT showing high-density shadow in the subarachnoid space. Analysis of these images through a deep learning-based subarachnoid hemorrhage will reduce the approximate rate of misdiagnosis in general and missed diagnosis by clinicians in particular. Deep learning-based detection of subarachnoid hemorrhage mainly includes two tasks, i.e., subarachnoid hemorrhage classification and subarachnoid hemorrhage region segmentation. However, it is difficult to effectively judge reliability of the model and classify bleeding which is based on limited predictive probability of convolutional neural network output. Moreover, deep learning-based bleeding area segmentation requires a large amount of training data to be marked in advance and the large number of network parameters makes the model training unable to reach the optimal. To resolve these problems associated with existing models, Bayesian deep learning and neural network-based hybrid model is presented in this paper to estimate uncertainty and efficiently classify subarachnoid hemorrhage. Uncertainty estimation of the proposed model helps in judging whether the model’s prediction is reliable or not. Additionally, it is used to guide clinicians to find the neglected subarachnoid hemorrhage area. In addition, a teacher-student mechanism deep learning model was designed to introduce observational uncertainty estimation for semisupervised learning of subarachnoid hemorrhage. Observation uncertainty estimation detects the uncertain bleeding areas in CT images and then selects areas with high reliability. Finally, it uses these unlabeled data for model training purposes as well.

show abstract

Section: Related Workmentioning

confidence: 99%

Deep Learning-Based Detection and Diagnosis of Subarachnoid Hemorrhage

Gou

2021

Journal of Healthcare Engineering

View full text Add to dashboard Cite

show abstract

“…Other studies that aimed to develop automated infarct segmentation on NCCT have used deep learning models trained on manual NCCT segmentations by neuroradiologists. 16–20 As shown in this study, there is substantial variation in manual segmentations even among experts. Similarly to algorithms trained on DWI, algorithms trained on manual segmentations will also consider other characteristics, such as tissue swelling or loss of differentiation between gray and white matter.…”

Section: Discussionmentioning

confidence: 70%

Semiautomated Detection of Early Infarct Signs on Noncontrast CT Improves Interrater Agreement

Christensen,

Demeestere,

Verhaaren

et al. 2023

Stroke

View full text Add to dashboard Cite

Background: Acute ischemic infarct identification on noncontrast computed tomography (NCCT) is highly variable between raters. A semiautomated method for segmentation of acute ischemic lesions on NCCT may improve interrater reliability. Methods: Patients with successful endovascular reperfusion from the DEFUSE 3 trial (Endovascular Therapy Following Imaging Evaluation for Ischemic Stroke) were included. We created relative NCCT (rNCCT) color-gradient overlays by comparing the density of a voxel on NCCT to the homologous region in the contralateral hemisphere. Regions with a relative hypodensity of at least 5% were visualized. We coregistered baseline and follow-up images. Two neuroradiologists and 6 nonradiologists segmented the acute ischemic lesion on the baseline scans with 2 methods: (1) manually outlining hypodense regions on the NCCT (unassisted segmentation) and (2) manually excluding areas deemed outside of the ischemic lesion on the rNCCT color map (rNCCT-assisted segmentation). Voxelwise interrater agreement was quantified using the Dice similarity coefficient and volumetric agreement between raters with the detection index (DI), defined as the true positive volume minus the false positive volume. Results: From a total of 92, we included 61 patients. Median age was 59 (64–77), and 57% were female. Stroke onset was known in 39%. Onset to NCCT was median, 8.5 hours (7–11) and median 10 hours (8.4–11.5) in patients with known and unknown onset, respectively. Compared with unassisted NCCT segmentation, rNCCT-assisted segmentation increased the Dice similarity coefficient by >50% for neuroradiologists (Dice similarity coefficient, 0.38 versus 0.83; P <0.001) and nonradiologists (Dice similarity coefficient, 0.14 versus 0.84; P <0.001), and improved the DI among nonradiologists (mean improvement, 5.8 mL [95% CI, 3.1–8.5] mL, P <0.001) but not among neuroradiologists. Conclusions: The high variability of manual segmentations of the acute ischemic lesion on NCCT is greatly reduced using semiautomated rNCCT. The rNCCT map may therefore aid acute infarct detection and provide more reliable infarct estimates for clinicians with less experience.

show abstract

“…These can occur as a result of a rupture in the blood-brain barrier, extravasation (enhancement) of contrast medium, or hemorrhage. 22 Based on study findings, it may be preferable to use MRI in the acute evaluation of patients with TIAMS to identify those with the highest risk of recurrent stroke. Currently, the use of MRI for TIAMS evaluation is highly variable, and clinicians have limited ability to predict which patients will have infarct based on imaging.…”

Section: Discussionmentioning

confidence: 99%

Post Ischemic Stroke Imaging and Its Clinical Relevance: A Systematic Review

Amin

2023

JARMHS

View full text Add to dashboard Cite

Stroke is a focal or global functional brain disorder that starts suddenly and lasts longer than 24 hours and is caused by changes in cerebral blood flow. It is not caused by transient cerebral circulatory abnormalities, brain tumors, or secondary strokes as a result of trauma or infection. Strokes can develop in any part of the brain. Ischemia or brain hemorrhage are also potential causes of a stroke that occurs suddenly and leads in neurologic impairments. Strokes are classified into two types: bleeding and non-bleeding. Both are potentially fatal. Several studies have found that imaging is required before treating an acute ischemic stroke. The importance of imaging after stroke treatment, on the other hand, is less obvious. Bleeding, particularly cerebral haemorrhage, is the most feared complication in people who have had acute ischemic stroke treatment. As a result, the AHA/ASA IV-tPA guidelines recommend that a non-contrast computed tomography (NCCT) brain scan be performed at least 24 hours after therapy. Non-contrast CT (NCCT) is still the cheapest, quickest, most generally available, and simplest method for diagnosing intracerebral bleeding after an ischemic stroke. However, MRI with the right sequences may be able to detect the same thing. The NCCT may frequently detect hyperdense regions following treatment for acute ischemic stroke. Magnetic resonance imaging (MRI) is one of the most commonly used imaging methods both before and after stroke treatment due to the wealth of information it gives. Finally, post-ischemic stroke imaging is becoming increasingly essential because it can provide both specific clinical guidance and a better knowledge of the processes that occur after such a severe impact on the brain. It may aid in the prediction of long-term outcomes and, in the future, may assist clinicians in tailoring and optimizing rehabilitation efforts for specific patients.

show abstract

Radiologist-Level Stroke Classification on Non-contrast CT Scans with Deep U-Net

Cited by 17 publications

References 18 publications

Deep Learning-Based Detection and Diagnosis of Subarachnoid Hemorrhage

Deep Learning-Based Detection and Diagnosis of Subarachnoid Hemorrhage

Semiautomated Detection of Early Infarct Signs on Noncontrast CT Improves Interrater Agreement

Post Ischemic Stroke Imaging and Its Clinical Relevance: A Systematic Review

Contact Info

Product

Resources

About