Classification of radiographic lung pattern based on texture analysis and machine learning.

Area of Science:

• Veterinary Radiology
• Medical Imaging Analysis
• Artificial Intelligence in Diagnostics

Background:

• Radiographic interpretation of lung patterns is crucial for diagnosing respiratory diseases in animals.
• Objective and quantitative methods are needed to improve the accuracy and consistency of radiographic assessments.

Purpose of the Study:

• To assess the feasibility of using texture analysis and machine learning (ML) to differentiate between normal and abnormal radiographic lung patterns in companion animals.
• To evaluate the performance of ML models in classifying four distinct lung patterns.

Main Methods:

• Extraction of 44 texture parameters from 1200 regions of interest across four lung patterns (normal, alveolar, bronchial, interstitial) from 512 thoracic radiographs of dogs and cats.
• Utilizing eight distinct texture analysis methods, including spatial gray-level-dependence matrices and fractal dimension analysis.
• Training and testing artificial neural networks with extracted texture parameters, evaluating classification performance using accuracy and area under the receiver operating characteristic curve (AUC).

Main Results:

• Forty texture parameters demonstrated significant differences among the lung patterns.
• High classification accuracy achieved: 99.1% in the training dataset and 91.9% in the testing dataset.
• Area under the receiver operating characteristic curve (AUC) values exceeded 0.98 (training) and 0.92 (testing), indicating robust model performance.

Conclusions:

• Texture analysis combined with machine learning provides a highly accurate and feasible method for distinguishing radiographic lung patterns.
• This AI-driven approach holds significant potential to enhance the evaluation of veterinary medical images.
• Further development could lead to improved diagnostic tools for animal respiratory conditions.