Evaluation of image quality in pediatric portable chest radiographs using AI-based noise reduction and edge enhancement.

PURPOSE: To evaluate the image quality of pediatric portable chest radiographs processed using a deep learning-based noise reduction (NR) algorithm implemented in clinical radiography systems, which is designed to reduce image noise without altering radiation dose, both alone and with edge enhancement.

MATERIALS AND

METHODS

This retrospective visual grading analysis included 101 pediatric patients (median age: 33 days; median weight: 2844 g) who underwent portable chest radiography. Each image was processed using four techniques: (1) standard (no processing), (2) edge enhancement only, (3) NR only, and (4) NR with edge enhancement. Image quality was assessed using five criteria: visualization of proximal bronchi, small peripheral airways, vertebrae, image noise, and overall image quality. In an anonymous, randomized review, two pediatric radiologists rated each criterion using a 5-point Likert scale. Statistical comparisons were conducted between processing methods.

RESULTS

Images processed with NR and edge enhancement (NR + /Filter +) achieved the highest mean scores across all criteria. Structural visibility-particularly of small peripheral airways, proximal bronchi, and vertebrae-showed significant improvement with edge enhancement (p < 0.0001). No significant difference in image noise was observed between NR-only and NR + /Filter + groups (p = 0.482).

CONCLUSION

AI-based noise reduction significantly improves image quality by reducing noise. Although edge enhancement does not further suppress noise, it improves the visibility of delicate anatomical structures. This combined approach may enhance diagnostic confidence in neonatal chest radiography, particularly under low-dose conditions.