Abstract
State of the art to specify the entity of solitary pulmonary nodules (SPN) is TTNA or TBNA. Optical coherence tomography (OCT) has the capability to close the gap between confocal microscopy and ultrasound.
To investigate OCT imaging in the diagnosis of SPN′s we developed a forward imaging needle probe. Single A-Scan images were taken from different tissue sides manually and consecutive A-Scans while puncturing robotically guided.
Histopathology defined the tissue probe as malignant or nonmalignant and characterized the architecture of the lung and SPN along the robotically punctured channel. The A-Scans were classified by a bidirectional long short term memory network. From 15 different tissue specimens 75 datasets of lung tissue and 22 datasets of tumor tissue have been acquired. Results are shown in Figure 1 [1 and 2: A-Scans from the same patient contribute to test set or trainings set. All tumor samples (PatAll), adeno carcinoma (PatAd). 3 and 4: A-Scans from the same record contribute to test set or training set. All carcinoma (RecAll), adeno carcinoma (RecAd)]. Further analysis of the individual patient data showed improved sensitivity, specificity and accuracy.
Our results demonstrate that individual OCT A-Scans can provide information for differentiation of malignant and non-malignant lung tissue. Further were we able to align histopathologic features and OCT features of the consecutive data sets.
- © 2014 ERS
