Machine Learning Application for Rupture Risk Assessment in Small-Sized Intracranial Aneurysm

Abstract

The assessment of rupture probability is crucial to identifying at risk intracranial aneurysms (IA) in patients harboring multiple aneurysms. We aimed to develop a computer-assisted detection system for small-sized aneurysm ruptures using a convolutional neural network (CNN) based on images of three-dimensional digital subtraction angiography. A retrospective data set, including 368 patients, was used as a training cohort for the CNN using the TensorFlow platform. Aneurysm images in six directions were obtained from each patient and the region-of-interest in each image was extracted. The resulting CNN was prospectively tested in 272 patients and the sensitivity, specificity, overall accuracy, and receiver operating characteristics (ROC) were compared to a human evaluator. Our system showed a sensitivity of 78.76% (95% CI: 72.30%–84.30%), a specificity of 72.15% (95% CI: 60.93%–81.65%), and an overall diagnostic accuracy of 76.84% (95% CI: 71.36%–81.72%) in aneurysm rupture predictions. The area under the ROC (AUROC) in the CNN was 0.755 (95% CI: 0.699%–0.805%), better than that obtained from a human evaluator (AUROC: 0.537; p < 0.001). The CNN-based prediction system was feasible to assess rupture risk in small-sized aneurysms with diagnostic accuracy superior to human evaluators. Additional studies based on a large data set are necessary to enhance diagnostic accuracy and to facilitate clinical application.

Rupture Risk Assessment for Cerebral Aneurysm Using Interpretable Machine Learning on Multidimensional Data

“…Although such models are simple and robust, they are limited to the use of a relatively small number of features and assume linear relationships between each feature and the risk of rupture. (19). Silva et al also developed a random forest model and achieved an area under the ROC curve of 0.81 (40).…”

“…Moreover, Liu et al developed a prediction model using Lasso regression based on radiomics features derived from angiographic images ( 18 ) and achieved area under the ROC curve of 0.853. Kim et al applied deep convolutional neural network to classify the rupture risk of small aneurysms based on angiographic images and achieved an area under the ROC curve of 0.755 ( 19 ). Silva et al also developed a random forest model and achieved an area under the ROC curve of 0.81 ( 40 ).…”

“…There has been growing interest in the use of machine learning to predict aneurysm rupture. Some of these prediction models have been developed based on morphological features (17)(18)(19), whereas others have been based on hemodynamic features (20). As the rupture of the aneurysm is clearly secondary to multifactorial causes, the use of only morphological and hemodynamic features may result in missing important information from other risk factors.…”

Easily created prediction model using deep learning software (Prediction One, Sony Network Communications Inc.) for subarachnoid hemorrhage outcomes from small dataset at admission

“…Machine learning has been used in neurosurgical situations,[ 27 , 33 , 39 , 42 , 47 ] but gradually DL is starting to be used as well in decision making for spinal canal stenosis,[ 2 ] predicting outcomes after stroke,[ 6 ] detecting seizure in intracranial electroencephalography recordings (UPenn and Mayo Clinic’s Seizure Detection Challenge),[ 44 , 48 ] pathological diagnosis,[ 29 ] or radiomics studies of brain tumors. [ 7 , 30 ] However, regarding predicting outcomes of SAH, prediction models only using random forests, categorized as machine learning, were made[ 39 , 47 ] with an accuracy of 70.9% from the 147 patients[ 39 ] or AUC of 0.837 from the 441 patients,[ 47 ] and there were no reports on the prediction model for SAH outcomes using DL.…”

Machine Learning Application With Quantitative Digital Subtraction Angiography for Detection of Hemorrhagic Brain Arteriovenous Malformations

“…In the clinical setting, it is challenging for radiologists to assess multiple clinical features and hemodynamic QDSA features simultaneously. Machine learning is a powerful approach for managing massive data with high-dimensional features; it has been used to predict the outcomes of several cerebrovascular diseases [9], [10]. To our knowledge, no researcher has used the temporal features of DSA to build machine learning models for the assisted diagnosis of hemorrhagic BAVMs and compared its performance with that of clinical radiologists.…”