研究目的
To investigate whether individual A-scans obtained in needle direction can contribute to the identification of pulmonary nodules using Optical Coherence Tomography (OCT) and Bidirectional Long Short Term Memory networks (BLSTMs).
研究成果
The results indicate that an automatic classification based on A-scans is feasible, although the method seems not suitable to provide a solitary diagnosis. The results indicated a substantial influence of patient specific properties on the OCT signal. Important signal features were located in the lower and upper part of the spatial frequency spectrum.
研究不足
The classification accuracies are not sufficient to form the sole basis for diagnosis, particularly for the initial experiments and without a priori knowledge about the tumor entity. The limited insight into the data processing is one major drawback of neuronal networks.
1：Experimental Design and Method Selection:
OCT A-scans from freshly resected human lung tissue specimen were recorded through a customized needle with an embedded optical fiber. BLSTMs were trained on randomly distributed training and test sets of the acquired A-scans.
2：Sample Selection and Data Sources:
Freshly resected and non-fixated human lung-tissue-specimens were obtained from lobectomies. Tumor suspicious areas were manually identified by palpation.
3：List of Experimental Equipment and Materials:
Spectral Domain OCT device 'Callisto' (Thorlabs), customized common path OCT probe based on an 18 gauge brachytherapy needle with embedded optical fiber.
4：Experimental Procedures and Operational Workflow:
The needle probe was manually inserted in the target area and fixated to prevent movement artifacts. Both, tumor suspicious and non-suspicious areas were punctured twice.
5：Data Analysis Methods:
BLSTM nets with a common net architecture consisting of two hidden layers and 8 LSTM blocks were trained on the training data and applied to test data.
独家科研数据包，助您复现前沿成果，加速创新突破
获取完整内容
