Disease Recognition in X-ray Images with Doctor Consultation-Inspired Model.

Area of Science:

• Computer Vision
• Artificial Intelligence
• Medical Imaging

Background:

• Deep learning models show promise for chest X-ray analysis in early disease screening.
• Current deep learning models exhibit inconsistent performance across different datasets.
• There is a need for robust methods to improve diagnostic accuracy in medical image analysis.

Purpose of the Study:

• To propose a novel doctor consultation-inspired method for fusing multiple deep learning models for chest X-ray analysis.
• To investigate the effectiveness of early and late fusion mechanisms in this model-agnostic approach.
• To enhance the accuracy and consistency of disease detection in medical imaging.

Main Methods:

• A doctor consultation-inspired framework was developed, treating individual deep learning models as medical doctors.
• Two fusion strategies were explored: early fusion (combining deep learned features) and late fusion (combining confidence scores).
• The proposed method was evaluated on two distinct X-ray imaging datasets, including the UIT COVID-19 dataset.

Main Results:

• The proposed doctor consultation-inspired method significantly outperformed baseline individual models on both datasets.
• Early fusion consistently demonstrated superior performance over late fusion across both benchmark datasets.
• Accuracy improvements of 3.03% and 1.86% were observed with the early fusion model on the UIT COVID-19 and chest X-ray datasets, respectively.

Conclusions:

• The doctor consultation-inspired fusion method offers a superior approach to chest X-ray analysis compared to individual deep learning models.
• Early fusion of deep learned features is a more effective strategy than late fusion for this task.
• This novel method holds significant potential for improving early disease screening using medical imaging.