Abstract:
The lack of sensing feedback information (information that can be used to guide surgical operations, including but not limited to force and visual information) has been a widely recognized issue in minimally invasive surgery (MIS). Without sufficient information about tissue types at the target area of surgical instruments, precise operations cannot be guaranteed. Cancer tissue cannot be fully excised without accurate and precise information on cancer locations and cancer margins. One approach to improve these situations is to apply fiber-optic Raman sensing technique in MIS or tumor resection procedures. In this research, a low-resolution fiber-optic Raman sensing system is designed to solve the problems of tissue type identification and bladder cancer diagnosis. Corresponding data analysis modules including data pre-processing algorithms, multivariate analysis, classification models, and data visualization module are developed and incorporated into a graphical user interface (GUI). Raman spectroscopy, a spectroscopic technique that is able to provide chemical compositional information of tissues at a molecular level, has been one of the most powerful and promising analytical tools in the last few decades and has been applied to various fields including biomedical fields, analytical chemistry, and biology, etc. It is comparatively straightforward, reproducible, non-invasive to cells or tissues under investigation. Furthermore, it is rapid and does not need any sample preparation, which makes it possible to perform in vivo inspections. Thus, it is able to fulfill the tissue type identification and bladder cancer diagnosis tasks of this research. However, the Raman signal is intrinsically weak. It can be easily contaminated by random noise and fluorescence background noise. To overcome all these, a denoising algorithm based on the lifting wavelet transform and a fluorescence correction algorithm based on mathematical morphology are developed in this project. Ex vivo tissue identification is done on porcine backbone samples bought from supermarkets using the low-resolution fiber-optic Raman sensing system and a Raman microscope. The results show that using a principal component analysis fed linear discriminant analysis model (PCA-LDA model), the Raman sensing system is able to classify bone, fatty tissue, muscle, intervertebral disc, and spinal cord with an overall accuracy of 93.1%. And the bone, fat and spinal cord samples are correctly identified with 100% specificity and sensitivity. A miniature fiber-optic Raman sensor, which can meet the requirement of MIS and be applied through endoscopes, is also designed and manufactured. Ex vivo bladder cancer diagnostic experiment is also done using the proposed Raman sensing system. High accuracy in discriminating normal bladder tissues, low-grade bladder tumors and high-grade bladder tumors is achieved which demonstrates the potential of using the low-resolution fiber-optic Raman system for in vivo bladder cancer diagnosis. A comparison experiment of the Raman sensing system on bladder cancer tissues is also carried out before forwarding this technique to the clinical environment. This experiment studies the influence of varying incident laser power, different integration time, and different classification algorithms. This project provides a comprehensive understanding of how low-resolution Raman technique can be applied to assist tissue type identification or bladder cancer diagnosis in the operating theater. A whole set of data analysis algorithms from raw Raman data acquirement to the construction of classification models is developed. Databases of porcine backbone and bladder cancer are established and will be continuously supplemented as more data is collected and analyzed.