Adapted generative latent diffusion models for accurate pathological analysis in chest X-ray images

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Public Health

Background:

• Respiratory diseases pose a significant global health burden, necessitating accurate diagnostic tools.
• Chest X-rays are vital for diagnosing respiratory conditions, but expert interpretation can be challenging.
• Deep learning models require substantial data, which is often limited in medical imaging.

Purpose of the Study:

• To propose a novel data augmentation methodology using adapted generative latent diffusion models.
• To enhance the performance of automatic pathological screening for tuberculosis and lung nodules.
• To address the data scarcity issue in medical imaging for deep learning applications.

Main Methods:

• Development of a new data augmentation technique based on generative latent diffusion models.
• Evaluation of the methodology on three publicly available chest X-ray datasets.
• An ablation study to determine the optimal configuration for image generation models.

Main Results:

• The proposed method significantly improved the performance of automatic screening for tuberculosis and lung nodules.
• Achieved high accuracy rates: 97.09% and 92.14% for tuberculosis screening, and 82.19% for lung nodule screening.
• Demonstrated superior performance compared to previous image generation methods for data augmentation.

Conclusions:

• The novel data augmentation methodology effectively enhances AI-based screening for critical respiratory diseases.
• The approach offers a promising solution to data scarcity challenges in medical AI.
• This contribution is vital for improving public health outcomes in diagnosing tuberculosis and lung nodules.