An efficient energy-density-dependent undersampling approach for compressive sensing in spectral domain optical coherence tomography (Conference Presentation)

Abstract

Many prior studies performed in the area of compressive optical coherence tomography (OCT) have mostly dealt with the problem of compressive sensing and sparse recovery of processed OCT images. Unlike these studies, in this paper, we study the application of compressive sensing in terms of efficient data storage and generating OCT images from undersampled raw unprocessed spectral domain OCT data. High resolution spectral domain OCT requires acquisition of enormous amount of data at very high sampling rate but such a large amount of the raw data impedes fast and efficient data storage and communication. To solve the problem of storing a large amount of data, we propose a specific undersampling method guided by the energy density of the spectral domain data in order to facilitate sparse representation of the raw data in terms of its salient frequency domain samples. This method takes into account not just the higher amplitude spectral data, as suggested in some previous studies but samples data based on nearly uniform distribution of energy over all the sampling intervals in the entire spectrum. Finally, we apply some state of the art sparse recovery methods involving L1 minimization to recover our desired high resolution images from the undersampled spectral domain data. We demonstrate the performance of our proposed scheme by comparing it with the recovery accuracy of some recent energy-guided undersampling methods and the conventional compressive sensing with random undersampling. We also compare the performance of our method with the other methods in terms of data compression ratio with respect to the reconstruction error.