Detection and Classification of Bronchiectasis Based on Improved Mask-RCNN.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Pulmonology

Background:

• Bronchiectasis, a chronic lung condition, is diagnosed via CT scans, but current methods lack standardized severity classification.
• Existing research often overlooks low-dose CT (LDCT) in favor of high-resolution CT (HRCT), limiting practical applications.
• Improved accuracy in bronchiectasis detection and classification is crucial for effective patient management.

Purpose of the Study:

• To develop an automated system for detecting and classifying bronchiectasis using LDCT images.
• To establish a reliable method for assigning bronchiectasis severity scores based on established criteria.
• To improve the efficiency and accuracy of bronchiectasis diagnosis in clinical practice.

Main Methods:

• Proposed ACER image enhancement, RDU-Net lung lobe segmentation, and HDC Mask R-CNN models for bronchiectasis analysis.
• Developed a Python-based system for automated LDCT image processing, detection, and classification.
• Integrated Reiff and BRICS scoring criteria for automatic patient bronchiectasis scoring.

Main Results:

• Achieved 91.4% accuracy in bronchiectasis detection and classification.
• Reported IOU, sensitivity, and specificity of 88.8%, 88.6%, and 85.4%, respectively.
• Demonstrated a rapid recognition speed of approximately 1 second per image.

Conclusions:

• The developed AI system accurately detects and classifies bronchiectasis from LDCT scans, matching human doctor accuracy.
• The system significantly reduces physician workload by efficiently processing large datasets and handling routine cases.
• This automated approach provides a valuable tool for clinical decision-making in bronchiectasis diagnosis and management.