Artificial Intelligence-Based Methods for Integrating Local and Global Features for Brain Cancer Imaging: Scoping
Hazrat Ali1, Rizwan Qureshi2, Zubair Shah1
1College of Science and Engineering, Hamad Bin Khalifa University, Doha, Qatar.
JMIR Medical Informatics
|November 17, 2023
Summary
Vision transformers (ViTs) show promise in brain cancer imaging, particularly for tumor segmentation. However, their computational complexity presents a challenge for widespread clinical adoption.
Area of Science:
- Artificial Intelligence
- Medical Imaging
- Neuroscience
Background:
- Transformer-based models are increasingly utilized in medical and cancer imaging.
- Recent studies highlight their application in brain cancer diagnosis and tumor segmentation.
Purpose of the Study:
- To review the contribution of various vision transformers (ViTs) to brain cancer diagnosis and segmentation.
- To examine ViT architectures for enhancing brain tumor segmentation.
- To explore how ViT models improve convolutional neural network performance in brain cancer imaging.
Main Methods:
- Systematic review following PRISMA-ScR guidelines.
- Searches conducted on PubMed, Scopus, IEEE Xplore, and Google Scholar.
- Independent reviewer selection and data extraction, with narrative synthesis.
Main Results:
- 22 studies (2021-2022) were included, focusing primarily on tumor segmentation using ViTs.
- Shifted Window transformer architectures are currently most popular.
- UNet transformer and TransUNet architectures require significant computational resources (8 GPUs).
- ViTs are combined with CNNs to capture both global and local image contexts.
Conclusions:
- Computational complexity of transformer architectures is a significant bottleneck for clinical translation.
- This review offers valuable insights for researchers in medical AI and brain cancer applications.
