Deep learning for quantitative motion tracking based on optical coherence tomography

Optical coherence tomography (OCT) is a cross-sectional imaging modality based on low coherence light interferometry. OCT has been widely used in diagnostic ophthalmology and has found applications in other biomedical fields such as cancer detection and surgical guidance. In the Laboratory of Biophotonics Imaging and Sensing at New Jersey Institute of Technology, we developed a unique needle OCT imager based on a single fiber probe for breast cancer imaging. The needle OCT imager with sub-millimeter diameter can be inserted into tissue for minimally invasive in situ breast imaging. OCT imaging provides spatial resolution similar to histology and has the potential to become a device to perform virtual biopsy to fast and accurate breast cancer diagnosis, because abnormal breast tissue and normal breast tissue have different characteristics in OCT image. The morphological features of OCT image are related to the microscopic structure of the tissue and the speckle pattern in OCT image is related to cellular/subcellular optical properties of the tissue. In addition, depth attenuation of OCT signal depends on the scattering and absorption properties of the tissue. However, the above described OCT image features are at different spatial scales and it is challenging for human visualization to effectively recognize these features for tissue classification. Particularly, our needle OCT imager, given its simplicity and small form factor, does not have a mechanical scanner for beam steering and relies on manual scan to generate 2D images. The nonconstant translation speed of the probe in manual scanning inevitably introduces distortion artifacts in OCT imaging, which further complicates the tissue characterization task.] OCT images of tissue samples provide comprehensive information about the morphology of normal and unhealthy tissue. Image analysis of tissue morphology can help cancer researchers develop a better understanding of cancer biology. Classification of tissue images and recovering distorted OCT images are two common tasks in tissue image analysis. In this master thesis project, a novel deep learning approach is investigated to extract beam scanning speed from different samples. Furthermore, a novel technique is investigated and tested to recover distorted OCT images. The long-term goal of this study is to achieve robust tissue classification for breast cancer diagnosis, based on a simple single fiber OCT instrument. The deep learning network utilized in this study depends on Convolutional Neural Network (CNN) and Naïve Bayes Classifier. For image retrieval, we used algorithms that extract, represent and match common features between images. The CNN network achieved accuracy of 97% in tissue type and scanning speed classification, while the image retrieval algorithms achieved very high-quality recovered image compared to the reference image