BO-CLAHE enhancing neonatal chest X-ray image quality for improved lesion classification

Area of Science:

• Medical Imaging
• Artificial Intelligence in Healthcare
• Neonatal Care

Background:

• Portable X-rays are vital for diagnosing neonates, particularly preterm and high-risk infants.
• Image quality of portable X-rays is often suboptimal, hindering accurate abnormality detection and AI model training.
• Enhancing image quality is critical, but increasing radiation dose is not feasible due to neonates' sensitivity.

Purpose of the Study:

• To develop an automated method for optimizing image enhancement techniques for neonatal portable X-rays.
• To improve the quality of X-ray images for better diagnostic accuracy in preterm and high-risk neonates.
• To facilitate the development of reliable AI-based diagnostic tools for neonatal lung diseases.

Main Methods:

• Proposed Bayesian Optimization CLAHE (BO-CLAHE) for automatic hyperparameter tuning of Contrast Limited Adaptive Histogram Equalization (CLAHE).
• Applied BO-CLAHE to enhance portable X-ray images of neonates.
• Evaluated the performance of enhanced images using various deep learning classification models.

Main Results:

• BO-CLAHE significantly improved the performance of AI models in classifying neonatal lung diseases.
• Notable improvements were observed in the diagnosis of Transient Tachypnea of the Newborn (TTN).
• Enhanced images led to superior classification accuracy compared to standard enhancement methods.

Conclusions:

• BO-CLAHE offers an efficient and effective solution for enhancing neonatal portable X-ray images.
• This method reduces the need for manual hyperparameter tuning, saving time and resources.
• The approach supports reduced radiation exposure and advances AI-driven diagnostics for improved neonatal outcomes.