Investigating recurrent neural networks for OCT A-scan based tissue analysis.

Area of Science:

• Pulmonary Medicine
• Medical Imaging
• Biotechnology

Background:

• Optical Coherence Tomography (OCT) offers high-resolution imaging for guiding medical procedures.
• Transbronchial biopsies are crucial for diagnosing lung conditions, but navigation can be challenging.

Purpose of the Study:

• To investigate the utility of individual OCT A-scans in identifying pulmonary nodules.
• To assess the contribution of OCT A-scans acquired in the needle's direction for diagnostic purposes.

Main Methods:

• OCT A-scans were acquired from human lung tissue using a custom needle with an embedded optical fiber.
• Bidirectional Long Short Term Memory (BLSTM) networks were trained on OCT A-scan datasets.
• The impact of patient-specific training and various pre-processing techniques was evaluated.

Main Results:

• Classification rates ranged from 67.5% to 76% across different training scenarios.
• Patient-specific training yielded the highest sensitivity (0.87) and specificity (0.85).
• Low-pass filtering negatively affected accuracy, reducing it to 62.2% and 56% at different cutoff frequencies.

Conclusions:

• Grey value-based classification using OCT A-scans is feasible for pulmonary nodule identification.
• OCT A-scans can potentially provide supplementary diagnostic and navigational information.
• Patient-specific signal characteristics and frequency spectrum components are important for classification.