Optical coherence tomography for identification of malignant pulmonary nodules based on random forest machine...

Area of Science:

• Pulmonology
• Medical Imaging
• Machine Learning

Background:

• Peripheral pulmonary nodules require accurate assessment to differentiate between benign and malignant conditions.
• Early and precise diagnosis is crucial for effective lung cancer treatment.

Purpose of the Study:

• To evaluate the feasibility of employing a random forest (RF) machine learning algorithm for classifying pulmonary nodules using in vivo endobronchial optical coherence tomography (EB-OCT).

Main Methods:

• A cohort of 31 patients with pulmonary nodules underwent EB-OCT imaging.
• Quantitative features, including attenuation coefficient and 29 image characteristics, were extracted from 1703 EB-OCT images.
• A RF classifier was trained on 70% of the data and tested on the remaining 30%, with validation against pathological results.

Main Results:

• Significant differences in attenuation coefficient and 29 image features were identified between normal and malignant nodules.
• The RF algorithm achieved a classification accuracy of 83.51% for malignant pulmonary nodules, with 90.41% sensitivity and 77.87% specificity.

Conclusions:

• Integrating EB-OCT imaging with machine learning offers a clinically viable method for distinguishing the nature of pulmonary nodules.
• Automated analysis of EB-OCT quantitative features presents a promising approach for the early detection of lung cancer.