Inhalation Injury Grading Using Transfer Learning Based on Bronchoscopy Images and Mechanical Ventilation Period.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Pulmonology

Background:

• The Abbreviated Injury Score (AIS) is a subjective clinical grading system for inhalation injuries.
• Current AIS grading shows inconsistent correlation with mechanical ventilation duration and no correlation with patient outcomes.
• Limitations stem from reliance on clinician expertise and inherent subjectivity.

Purpose of the Study:

• To develop and validate a novel, objective deep learning-based method for grading inhalation injuries.
• To utilize bronchoscopy images for automated injury assessment, focusing on carbonaceous deposits, blistering, and fibrin casts.
• To improve the accuracy and reliability of inhalation injury grading compared to traditional methods.

Main Methods:

• A deep learning model was developed using transfer learning and data augmentation techniques.
• The model analyzes bronchoscopy images to identify and grade inhalation injury severity.
• The model was tested on images from 18 patients with inhalation injuries (severity grades 1-6).

Main Results:

• The proposed deep learning method achieved an overall accuracy of 86.11%.
• The model demonstrated superior performance when incorporating both transfer learning and data augmentation.
• Key performance metrics included accuracy, sensitivity, specificity, F-1 score, and precision.

Conclusions:

• Deep learning analysis of bronchoscopy images offers a promising, objective approach to inhalation injury grading.
• The developed method, enhanced by transfer learning and data augmentation, significantly improves grading accuracy.
• This AI-driven approach has the potential to enhance clinical decision-making and patient outcome prediction for inhalation injuries.