Differentially Private Multi-task Learning

Abstract

Recently it has been shown that multiwavelength photoacoustic imaging has the potential to discriminate between normal and atheromatous areas of arterial tissue when operating in the 740-1300nm wavelength range. At this wavelength range the absorption spectrum of lipids and normal arterial tissue are significantly different allowing discrimination between one another. Also, this wavelength range has the advantage of being relatively weakly absorbed by blood. This obviates the need for a saline flush if implemented using an intravascular imaging probe. In this study we investigate the possibility of identifying regions of high lipid concentration from 2D multiwavelength photoacoustic images of vascular tissue by exploiting the unique spectral features of lipids. Recognising regions of high lipid concentration would be useful to identify plaques which are likely to rupture (vulnerable plaques). To investigate this, samples of post mortem human aortas were imaged at a range of near-infrared (NIR) wavelengths and compared to histology. Photoacoustic images were also obtained when illuminating the sample through blood. This study demonstrated that lipid rich atheromatous plaques can clearly be identified using multiwavelength photoacoustic imaging.