Detection and vascular territorial classification of stroke on diffusion-weighted MRI by deep learning

PURPOSE

Rapid detection and vascular territorial classification of stroke enable the determination of the most appropriate treatment. In this study, we aimed to investigate the performance of convolutional neural network (CNN) models in the detection and vascular territorial classification of stroke on diffusion-weighted images (DWI).

METHODS

DWI of 421 cases (271 acute ischemic stroke patients and 150 cases without any ischemia findings on DWI) obtained between January 2017 to April 2020 were reviewed. We created two custom datasets. A stroke detection dataset was created with 1800 slices (900 S and 900 normal) consisting of 1400 for training, 200 for validation, 200 for test. A vascular territorial type dataset was created with 1717 slices (883 middle cerebral artery stroke, 416 posterior circulatory stroke, and 418 watershed stroke) consisting of 1117 slices for training, 300 for validation, 300 for test. A transfer learning approach based on MobileNetV2 and EfficientNet-B0 CNN architecture was used. The performance of the models was evaluated.

RESULTS

Modified MobileNetV2 and EfficientNet-B0 models achieved 96% (κ: 0.92) and 93% (κ: 0.86) accuracy in stroke detection, respectively. In vascular territorial classification of stroke as middle cerebral artery, posterior circulation, or watershed infarction, an accuracy of 93% (κ: 0.895) was achieved with modified MobileNetV2 model and 87% (κ: 0.805) with modified EfficientNet-B0 CNN model.

CONCLUSION

Transfer learning approach with custom top CNN models achieve sufficiently high performance for both the detection of ischemic stroke and the classification of its vascular territorial type on DWI.

Relationship of the vascular territory affected by delayed cerebral ischemia and the location of the ruptured aneurysm in patients with aneurysmal subarachnoid hemorrhage

OBJECTIVE

To determine the area most at risk of delayed cerebral ischemia (DCI) in relation to the location of the ruptured aneurysm in patients with aneurysmal subarachnoid hemorrhage (aSAH) and, therefore, help to choose the site for focal multimodal neuromonitoring.

METHODS

We retrospectively analyzed angiographic findings, CCT scans, and patient charts of patients who were admitted with aSAH to our neurosurgical intensive care unit between 2009 and 2017. DCI was defined as infarction on CCT 2-6 weeks after aSAH.

RESULTS

DCI occurred in 17.9% out of 357 included patients. A DCI occurring in the vascular territory of the artery carrying the ruptured aneurysm was found in 81.0% of patients with anterior circulation aneurysms but only in 16.7% with posterior circulation aneurysms (Fisher's exact, p=0.003). The vascular territory most frequently showing a DCI was the ipsilateral MCA territory (86.7%) in ICA aneurysms, the contra- (71.4%) and the ipsilateral (64.3%) ACA territory in ACA aneurysms, the right (93.8%) and the left (81.3%) ACA territory in AcomA aneurysms, and the ipsilateral MCA territory in MCA aneurysms (69.2%) as well as in VA/PICA/SCA aneurysms (100.0%). DCI after the rupture of a BA aneurysm occurred with 33.3% in 6 out of 8 vascular territories, respectively. DCI of multiple vascular territories occurred in 100.0% of BA aneurysms, 87.5% of AcomA aneurysms, 71.4% of ACA aneurysms, 40.0% of ICA aneurysms, 38.5% of MCA aneurysms, and 33.3% of VA/PICA/SCA aneurysms.

DISCUSSION

Few studies exist that could determine the area most at risk of a DCI after an aSAH. Our data could identify the territory most at risk for DCI with a probability of > 60% except for BA aneurysms, which showed DCI in various areas and patients suffering from multiple DCIs. Either the ipsilateral ACA or MCA were affected by the DCI in about 80% of ACA and more than 90% of AcomA, ICA, MCA, and VA/PICA/SCA aneurysms. Therefore, local intraparenchymal neuromonitoring in the ACA/MCA watershed area might detect the vast majority of DCIs for all aneurysm locations, except for BA aneurysms. In ACA and AcomA aneurysms, bilateral DCI of the ACA territory was common, and bilateral probe positioning might be considered for monitoring high-risk patients. Non-focal monitoring methods might be preferably used after BA aneurysm rupture.

Magnetic resonance imaging characteristics of ischemic stroke in an Afro-Caribbean population: A 1-year prospective MRI study on 534 consecutive patients

PURPOSES

Few population-based MRI studies on stroke, particularly in African-descent populations, are available. Based on a 1-year Afro-Caribbean population-based study MRI, ischemic stroke characteristics were extensively analyzed.

METHODS

All strokes occurring in Martinique (390,371 inhabitants) were prospectively included. Ascertainment was based, whenever possible, on MRI. All patients were categorized as single- (subclassified as cortical, cortical-subcortical, subcortical, lacunar) or multiple-lesion pattern, and vascular (single, multiple or junctional) territory. Brain parenchyma was evaluated, based on visualization of macrobleeds, microbleeds, white-matter hyperintensities or stroke sequelae. Etiology was classified according to TOAST criteria.

RESULTS

Among 596 ischemic stroke patients included, 534 (295 men, 239 women; mean age, 71 [range 23-110] years) underwent MRI (median delay 1 day). Four hundred and eighty-eight had single-type lesion (14.8% cortical, 42.4% cortical-subcortical, 14.5% subcortical, 16.6% lacunar), involving anterior cerebral (4%), middle cerebral (63.7%), posterior cerebral artery (10.4%) or basilar trunk (11.7%) territories, with 10.3% simultaneously involving multiple territories and 4.9% junctional infarction. Etiologies were LAA (11.2%), SVD (10.7%), CE (29.6%), rare (4.5%) or undetermined (44.1%).

CONCLUSION

Our prospective, consecutive, ischemic stroke series gives a comprehensive description of ischemic stroke imaging patterns and etiologic distributions in an Afro-Caribbean population with high socio-economic status. Our patients' stroke characteristics are close to those of European-descent populations.