Multi-axis transformer based U-Net with class balanced ensemble model for lung disease classification using X-ray...

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computer Vision

Background:

• Chest X-rays are crucial for diagnosing lung disorders due to high sensitivity.
• Conventional Convolutional Neural Networks (CNNs) face limitations from localization bias in classification tasks.

Purpose of the Study:

• To introduce a novel Multi-Axis Transformer based U-Net with Class Balanced Ensemble (MaxTU-CBE) for enhanced multi-label classification of chest X-rays.
• To address the localization bias inherent in traditional CNN models.

Main Methods:

• Integration of hierarchical Multi-Axis Transformer into the U-Net architecture's encoder and decoder.
• Utilizing a Contextual Fusion Engine (CFE) with self-attention for multi-scale feature interdependence.
• Employing deep CNN with ensemble learning to manage class imbalance.

Main Results:

• The MaxTU-CBE model demonstrates improved semantic segmentation of lung images.
• The Contextual Fusion Engine (CFE) effectively captures global feature interdependencies across scales.
• The ensemble learning approach mitigates challenges associated with unbalanced class learning.

Conclusions:

• The proposed MaxTU-CBE significantly outperforms BiDLSTM classifiers by 1.42% and CBIR-CSNN techniques by 5.2% in multi-label classification accuracy.
• MaxTU-CBE represents a significant advancement in AI-driven chest X-ray analysis for improved diagnostic performance.