Multimodal image fusion for ich detection and classification using parallel Dl models
Intracranial haemorrhage (ICH) can be a potential consequence of traumatic brain injuries. Single-modality images provide limited information due to different imaging principles and organ structures. Multimodality image fusion techniques can help incorporate different types of images into a single,...
Saved in:
| Main Authors: | , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Taylor & Francis Group
2025-12-01
|
| Series: | Computer Methods in Biomechanics and Biomedical Engineering: Imaging & Visualization |
| Subjects: | |
| Online Access: | https://www.tandfonline.com/doi/10.1080/21681163.2025.2468436 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Intracranial haemorrhage (ICH) can be a potential consequence of traumatic brain injuries. Single-modality images provide limited information due to different imaging principles and organ structures. Multimodality image fusion techniques can help incorporate different types of images into a single, high-quality image, but current methods face challenges like fusion artefacts, complex design, and high computational costs. This paper introduces an innovative multimodal medical image fusion method using parallel Deep Learning (DL) models. The research uses a carefully chosen multimodal dataset, including CT and MRI, and a novel segmentation algorithm called DFA-UNet. The model employs an encoder-decoder structure and uses a DFA attention module for efficient image feature extraction and easy channel feature weight integration. The method uses Graph Neural Networks (GNNs) for spatial features and Recurrent Neural Networks (RNNs) for temporal complexities. The architecture starts with pre-processing to standardize and enhance multimodal data, and simultaneous DL models are included to analyze fused input for a comprehensive understanding of ICH patterns. Attention mechanisms are carefully included to improve accessibility. The proposed parallel DL architecture demonstrates high efficiency in classifying images as either haemorrhage or non-haemorrhage. |
|---|---|
| ISSN: | 2168-1163 2168-1171 |