Quantitative evaluation model of variable diagnosis for chest X-ray images using deep learning

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Radiology

Background:

• Chest X-rays are vital for disease detection but physician interpretation of subtle findings can vary.
• Quantifying subjective clinical evaluations is challenging and labor-intensive.
• Variability in interpreting normal-range findings necessitates objective evaluation methods.

Purpose of the Study:

• To develop and demonstrate a deep learning model for quantitatively evaluating physician assessments of chest X-ray findings.
• To establish a novel method for quantifying variability in the interpretation of pleural thickening and scoliosis.
• To assess the model's performance using binary annotation data.

Main Methods:

• Utilized a dataset of 83,005 chest X-ray images.
• Trained a deep learning model using only binary annotation data.
• Applied transfer learning with convolutional neural networks and a vector quantization variational autoencoder model.

Main Results:

• Successfully quantified variations in physician interpretations of chest X-ray findings.
• Achieved high correlation coefficients ranging from 0.89 to 0.97.
• Demonstrated the model's applicability to different deep learning architectures.

Conclusions:

• The developed deep learning method provides an objective approach to quantifying subjective clinical evaluations in medical imaging.
• This approach can reduce diagnostic variability and improve the consistency of interpretations for findings like pleural thickening and scoliosis.
• The model's effectiveness highlights the potential of AI in standardizing medical image analysis.