From Binary to Multi-Class Classification: A Two-Step Hybrid CNN-ViT Model for Chest Disease Classification Based on X-Ray Images

Diagnostics, Journal Year: 2024, Volume and Issue: 14(23), P. 2754 - 2754

Published: Dec. 6, 2024

Background/Objectives: Chest disease identification for Tuberculosis and Pneumonia diseases presents diagnostic challenges due to overlapping radiographic features the limited availability of expert radiologists, especially in developing countries. The present study aims address these by a Computer-Aided Diagnosis (CAD) system provide consistent objective analyses chest X-ray images, thereby reducing potential human error. By leveraging complementary strengths convolutional neural networks (CNNs) vision transformers (ViTs), we propose hybrid model accurate detection distinguishing between Pneumonia. Methods: We designed two-step that integrates ResNet-50 CNN with ViT-b16 architecture. It uses transfer learning on datasets from Guangzhou Women’s Children’s Medical Center cases Qatar Dhaka (Bangladesh) universities cases. CNNs capture hierarchical structures while ViTs, their self-attention mechanisms, excel at identifying relationships features. Combining approaches enhances model’s performance binary multi-class classification tasks. Results: Our CNN-ViT achieved accuracy 98.97% detection. For classification, Tuberculosis, viral Pneumonia, bacterial an 96.18%. These results underscore improving reliability based images. Conclusions: proposed demonstrates substantial advancing robustness CAD systems diagnosis. integrating ViT architectures, our approach precision, which may help alleviate burden healthcare resource-limited settings improve patient outcomes

Language: Английский

DW-MLSR: Unsupervised Deformable Medical Image Registration Based on Dual-Window Attention and Multi-Latent Space

Electronics, Journal Year: 2024, Volume and Issue: 13(24), P. 4966 - 4966

Published: Dec. 17, 2024

(1) Background: In recent years, the application of Transformers and Vision (ViTs) in medical image registration has been constrained by sliding attention mechanisms, which struggle to effectively capture non-adjacent but critical structures, such as hippocampus ventricles brain. Additionally, lack labels unsupervised often leads overfitting. (2) To address these issues, we propose a novel method, DW-MLSR, based on dual-window multi-latent space. The mechanism enhances transmission information across while space improves model’s generalization learning latent representations. (3) Experimental results demonstrate that DW-MLSR outperforms mainstream models, showcasing significant potential registration. (4) method addresses limitations transmitting between windows, performance registration, demonstrates broad prospects

Language: Английский