Convolutional Neural Network-Processed MRI Images in the Diagnosis of Plastic Bronchitis in Children

Area of Science:

• Medical Imaging
• Artificial Intelligence in Medicine
• Pediatric Pulmonology

Background:

• Plastic bronchitis (PB) is a rare but serious condition in children.
• Accurate imaging is crucial for diagnosing and managing PB.
• Current segmentation methods for PB on MRI may have limitations.

Purpose of the Study:

• To evaluate the effectiveness of a convolutional neural network (CNN) algorithm for segmenting magnetic resonance imaging (MRI) in pediatric patients with plastic bronchitis (PB).
• To compare the performance of the CNN algorithm against the Active Appearance Model (AAM) algorithm for MRI image segmentation in PB.
• To identify imaging features and complications associated with PB in children.

Main Methods:

• A CNN-based algorithm was developed and applied to chest MRI scans of 30 children diagnosed with PB.
• The segmentation performance of the CNN algorithm was compared with the Active Appearance Model (AAM) algorithm using Dice and Jaccard coefficients.
• Clinical data and imaging findings, including pulmonary inflammation, atelectasis, pleural effusion, and pneumothorax, were analyzed.

Main Results:

• The CNN algorithm achieved a higher maximum Dice coefficient (0.946) compared to AAM (0.843).
• The CNN algorithm also demonstrated a superior Jaccard coefficient (0.894) versus AAM (0.758), with P < 0.05.
• Pulmonary inflammation was present in all subjects; complications included atelectasis (46.66%), pleural effusion (30%), pneumothorax (10%), mediastinal emphysema (6.67%), and pneumopericardium (6.67%).

Conclusions:

• CNN-based algorithms significantly enhance the accuracy of MRI image segmentation for pediatric plastic bronchitis.
• Pleural effusion is identified as a critical factor influencing the occurrence and progression of PB.
• The findings support the use of advanced AI techniques for improved diagnosis and management of PB in children.