Scholarlink

Abstract <p>In this study, the authors have developed a prototype of a three-dimensional convolutional neural network with an attention block that detects the probability of intracranial cerebral aneurysms in a single contrast computed tomography angiography study. DICOM contrast computed tomography angiography data with and without intracranial aneurysms were used to train the network. Metadata from the studies were not used. The data were divided into training and validation subsets in the proportion of 65 and 35%, respectively. Using Keras and TensorFlow libraries in the Python programming environment, a 192 × 192 × 128 three-dimensional convolutional neural network model with four convolutional layers, a kernel of dimension 3 and self-attention block was developed. The classification accuracy, precision, and recall on test samples reached 96, 99, and 93% respectively. This exceeded the performance of previously known neural networks.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Development of a Three-Dimensional Convolutional Neural Network with Attention for Aneurysm Detection

  • 期刊论文

  • 发表日期：

    2025年12月1日

查看全文
  • S. G. Sinitsa,
  • E. I. Zyablova,
  • D. O. Kardailskaya,
  • I. A. Zayats,
  • A. A. Khalafyan,
  • A. V. Ishchenko

摘要

Abstract

In this study, the authors have developed a prototype of a three-dimensional convolutional neural network with an attention block that detects the probability of intracranial cerebral aneurysms in a single contrast computed tomography angiography study. DICOM contrast computed tomography angiography data with and without intracranial aneurysms were used to train the network. Metadata from the studies were not used. The data were divided into training and validation subsets in the proportion of 65 and 35%, respectively. Using Keras and TensorFlow libraries in the Python programming environment, a 192 × 192 × 128 three-dimensional convolutional neural network model with four convolutional layers, a kernel of dimension 3 and self-attention block was developed. The classification accuracy, precision, and recall on test samples reached 96, 99, and 93% respectively. This exceeded the performance of previously known neural networks.