An anatomically enhanced and clinically validated framework for lung abnormality classification using deep features...

Area of Science:

• Medical Imaging
• Radiology
• Computer-Aided Diagnosis

Background:

• Chest X-rays are crucial for diagnosing lung abnormalities, but subtle variations can be missed by traditional methods.
• Image augmentation techniques can distort anatomical features, leading to misdiagnosis.
• Accurate and efficient classification of lung conditions is essential for timely treatment.

Purpose of the Study:

• To propose a novel two-stage framework, Anatomical Segmentation and Color-Based Enhancement (ASCE), for precise lung abnormality classification.
• To preserve anatomical integrity during the detection process.
• To improve the efficiency and interpretability of lung abnormality diagnosis.

Main Methods:

• A two-stage framework integrating Anatomy-Preserved Segmentation and Color-Based Enhancement.
• Utilizing Kullback-Leibler (KL) divergence to quantify deviations from healthy lung regions.
• A lightweight pipeline for computational efficiency (approximately 0.06s/image).

Main Results:

• Stage 1 achieved 95% accuracy, 0.98 AUC, and 0.92 F1-score for Normal vs. Pneumonia classification.
• Stage 2 achieved 100% accuracy and F1-score for distinguishing Viral and Bacterial pneumonia subtypes.
• The method preserves diagnostic features and enhances visibility of abnormalities.

Conclusions:

• The ASCE framework offers precise and efficient classification of lung abnormalities.
• The approach enhances diagnostic visibility and enables quantitative analysis.
• This method ensures clinical interpretability by preserving critical anatomical structures.