Feature Of The Week 5/16/10: Scatterer Size-Based Analysis of Optical Coherence Tomography Images Using Spectral Estimation Techniques
Feature Of The Week 5/16/10: A great challenge in the fight against cancer is the timely diagnosis of the disease which greatly enhances the chance of survival. Optical imaging can play an important role in the process providing powerful, non-invasive, diagnostic tools. However, in order to reach their full potential, optical techniques must be able to delineate the changes associated with the early stages of cancer (known as dysplasia and carcinoma in situ.) These include sub-micron variations such as disparities in nuclear number and size, sub-nuclear modifications, and variations in other cellular organelles. Currently, this level of detail is only available by histopathology or, non-invasively, by confocal or multi-photon microscopy. Unfortunately, neither of the two modalities has been extensively implemented due their complexity and their limited penetration in tissue. Optical Coherence Tomography (OCT) is a non-invasive tissue imaging technique that generates in vivo cross-sectional images of tissue microstructure. OCT provides a resolution of 1-10 μm with a penetration of a few mm. However, some dysplastic changes are not clearly discernible even at that resolution. One possible remedy for this limitation may be the detection of such changes indirectly, based on the size-dependent spectral variations they induce on the backscattered light.
Dr. Costas Pitris, MD, PhD, from the University of Cyprus and colleagues describe here a novel technique for the analysis of the spectral contents of OCT signals which, in combination with advanced statistical methods, can extract information about scatterer size, a diagnostically relevant tissue feature which yet remains below the resolution of OCT. In the imaging volume of an OCT voxel many individual scattering sides, such as organelles, nuclei, nucleoli, etc, can exist simultaneously. These multiple scatterers, which are spaced less than a coherence length apart, interfere coherently in a stochastic manner. The statistical properties of this signal depend on scatterer size and concentration. Using Spectroscopic OCT, this information can be retrieved by analyzing the statistical properties of the backscattered spectra. The spectral content of the backscattered light can be obtained by time-frequency analysis of the interferometric OCT signal. Previous research relied on the Short-Time Fourier transform (STFT) and the continuous wavelet transform (Morlet transform) to obtain the spectrum. This paper introduces a novel technique for calculating and processing the spectral content of OCT images based on autoregressive techniques and Principal Component Analysis (PCA) with their application demonstrated on appropriate phantoms. The particle size can be calculated from a set of linear equations and the samples can be classified into categories using linear discriminant analysis or k-means clustering.
The proposed method is shown to be very effective in differentiating spectral differences which depend on scatterer diameter. Such information can be very useful diagnostically since malignant tissue is known to exhibit characteristic scatterer changes. When fully developed, this technique can be used to differentiate dissimilar areas of the tissue under investigation. One of the major advantages of this method is the fact that it does not require knowledge of the variations in spectral content of the tissue under investigation. Such an unsupervised technique could make available a tool for differentiating tissues with different size scatterers with no a-priori information regarding spectral content or scatterer size. Once verified with histology and pre-clinical studies, such a tool can prove extremely valuable for the investigation of disease tissue features which yet remain below the resolution of OCT.