Aneurysm Detection Using Ternary View Densenet and Shifted Window Vision Transformer on Cta Images

Abstract

Detection and management of intracranial aneurysms (IAs) are essential to prevent deadly complexities. Artificial Intelligence (AI) can investigate medical images, like Computed Tomographic Angiography (CTA) or Magnetic Resonance Angiography (MRA), distinguishing traces or implications probably ignored by humans. Early detection smoothen timely interferences and enhances results. In addition, AI algorithms provide quantitative data on aneurysm attributes, assisting in long-term monitoring. Essentially, recent years have noticed the growth of sophisticated computational methods using deep learning algorithms. However, convolutional neural network (CNN)-based methods have an intrinsic drawback in capturing long-range spatial dependencies that accidentally trade-offs the holding of global features crucial for segmentation. In response to this challenge, we introduce an artificial intelligence (AI) application to screen aneurysm detection using Ternary View DenseNet and Shifted Window Vision Transformer (TVDN-SWViT) which incorporates three different vessel-enhanced scales into vision transformer using shifted windows to perform aneurysm detection on CTA images. Initially Lightweight Statistical Hessian-based Pre-processing incorporating skull stripping and vessel enhancement are applied to the sample images to improve noise ratio. Second the pre-processed samples are subjected to an AI-based model called, Ternary View Partial Attention DenseNet to extract relevant features. Finally the noise improved pre-processed samples and relevant features extracted forms as input to the Dual Position Shifted Window-based Vision Transformer for Aneurysm Detection. We trained TVDN-SWViT using the CMHA dataset and noticed significant performance improvements compared to conventional method. The TVDN-SWViT method reached an accuracy of 28% showing its robustness in ascertaining the detection of aneurysm. The promising results demonstrate that TVDN-SWViT can play an important role in clearing the way for accurate diagnosis, while also exhibiting the prospective of vision transformer-based architectures in medical imaging with improved noise ratio.