Automated evaluation of probe-based confocal laser endomicroscopy in the lung

• Published in: PLOS ONE Vol. 15; no. 5; p. e0232847
• Main Authors: Bondesson, David, Schneider, Moritz J., Silbernagel, Edith, Behr, Jürgen, Reichenberger, Frank, Dinkel, Julien
• Format: Journal Article
• Language: English
• Published: United States Public Library of Science (PLoS) 06.05.2020; Public Library of Science
• Subjects: Aged; Algorithms; Alveoli; Automation; Biology and Life Sciences; Biopsy; Bronchoscopy; Bronchoscopy - methods; Computer aided medical diagnosis; Computer and Information Sciences; Computer Systems; Data mining; Diagnosis; Diagnostic systems; Diseases; Elastin; Elastin - analysis; Elastin - ultrastructure; Equipment Design; Evaluation; Female; Hospitals; Humans; Image Processing, Computer-Assisted; Image segmentation; Interstitial lung diseases; Laser endoscopy; Laser scanning microscopy; Lasers; Learning algorithms; Lung diseases; Lung Diseases, Interstitial; Lung Diseases, Interstitial - pathology; Machine learning; Male; Medical diagnosis; Medicine; Medicine and Health Sciences; Methods; Microscopy, Confocal; Microscopy, Confocal - instrumentation; Microscopy, Confocal - methods; Microscopy, Video; Microscopy, Video - instrumentation; Microscopy, Video - methods; Middle Aged; Non-Smokers; Patients; Physical Sciences; Pneumonia; Pulmonary Alveoli; Pulmonary Alveoli - chemistry; Pulmonary Alveoli - ultrastructure; Q; R; Radiology; Research and Analysis Methods; Research Article; Respiratory tract diseases; Science; Smoking Cessation; Statistical analysis; Supervised Machine Learning; Thickness; Time; Germany; France; Mauna Kea
• ISSN: 1932-6203; 1932-6203
• DOI: 10.1371/journal.pone.0232847

Probe-based confocal endomicroscopy provides real time videos of autoflourescent elastin structures within the alveoli. With it, multiple changes in the elastin structure due to different diffuse parenchymal lung diseases have previously been described. However, these evaluations have mainly relied on qualitative evaluation by the examiner and manually selected parts post-examination. To develop a fully automatic method for quantifying structural properties of the imaged alveoli elastin and to perform a preliminary assessment of their diagnostic potential. 46 patients underwent probe-based confocal endomicroscopy, of which 38 were divided into 4 groups categorizing different diffuse parenchymal lung diseases. 8 patients were imaged in representative healthy lung areas and used as control group. Alveolar elastin structures were automatically segmented with a trained machine learning algorithm and subsequently evaluated with two methods developed for quantifying the local thickness and structural connectivity. The automatic segmentation algorithm performed generally well and all 4 patient groups showed statistically significant differences with median elastin thickness, standard deviation of thickness and connectivity compared to the control group. Alveoli elastin structures can be quantified based on their structural connectivity and thickness statistics with a fully-automated algorithm and initial results highlight its potential for distinguishing parenchymal lung diseases from normal alveoli.