Detecting Large Vessel Occlusion at Multiphase CT Angiography by Using a Deep Convolutional Neural Network

Abstract

A cute stroke care changed dramatically in 2015 with the publication of several randomized control trials demonstrating that endovascular therapy is more effective than alteplase alone for large vessel occlusion (LVO) stroke (1-7). Furthermore, the endovascular therapy treatment effect is profoundly time dependent, and every minute that we work faster to achieve vessel recanalization we can provide the gift of a week of disability-free life to patients (8). LVO stroke is a medical diagnosis that cannot be missed and must be made quickly. One of the commonly used methods to confirm or exclude the presence of a LVO quickly is with CT angiography, a 3-minute examination that can easily be performed following the noncontrast head CT that is standard of care for all acute stroke imaging (9-11). Multiphase CT angiography is a protocol recently introduced for acute stroke imaging that aims to both improve LVO detection and improve patient selection for endovascular therapy (12-14). Recent advances in deep learning, a class of machine learning, inspired new research on the uses of convolutional neural networks to perform at high levels in computer vision problems (15). These architectures hold great promise for enhancing the workflow in radiology. Specifically, in the neuroradiology domain, deep learning has been used to segment microhemorrhages at MRI with a sensitivity greater than 93% (16), to automate identification and to segment ischemic brain lesions on diffusion-weighted MRI scans (17), to detect early ischemic infarct at noncontrast CT with precision similar to diffusion-weighted MRI (18), and to detect intracranial hemorrhage at CT by reducing the time to diagnosis by 96% (19).

Diagnostic performance of an algorithm for automated large vessel occlusion detection on CT angiography

“…http://jnis.bmj.com/ pattern, and tortuosity of these vessels, making vascular segmentation more susceptible to errors. Especially for M2 occlusions, it is possible that other acquisition schemes such as multiphase CTA lead to better detection by automated algorithms, 10 as is seen for M2 occlusion detection by human readers. 29 The strengths of this study include the large sample size of LVOs, allowing us to assess diagnostic performance both for overall detection of an LVO and per individual occlusion location with sufficient precision.…”

“…These tools are aimed at screening CTAs for LVOs and, in case of a positive finding, notifying not only local radiologists but also the stroke team at the nearest EVT-capable stroke center. [10][11][12][13][14] Determining the performance of such algorithms is needed to estimate their potential clinical utility.…”

Emerging Detection Techniques for Large Vessel Occlusion Stroke: A Scoping Review

“…Nine studies evaluated the viability of using algorithmic and machine learning methods to predict LVO (57)(58)(59)(60)(61)(62)(63)(64)(65). Five of these studies used algorithms to accurately predict LVO from imaging methods, such as CT images (57,60,(62)(63)(64).…”

“…Nine studies evaluated the viability of using algorithmic and machine learning methods to predict LVO (57)(58)(59)(60)(61)(62)(63)(64)(65). Five of these studies used algorithms to accurately predict LVO from imaging methods, such as CT images (57,60,(62)(63)(64). One study established an automated evaluation system which contained three hierarchical models based on patients' demographic data (model 1), clinical data (model 2), and NCCT scans obtained from a deep learning model (model 3).…”

“…Due to the success of deep learning applications in other brain pathology, a deep convolutional neural network was trialled to assess the feasibility of detecting acute LVOs on multiphase CTA, with sensitivity of 96% and specificity of 81% (62).…”