A novel algorithm to optimize generalized gamma distributed multiplicative noise with implications on speckle removal from OCT images

Optical coherence tomography (OCT) images are corrupted by multiplicative generalized gamma distributed speckle noise that significantly degrades the contrast to noise ratio (CNR) of microstructural compartments in biological applications. This work proposes a novel algorithm to optimize the negative log likelihood of the spatial distribution of speckle. Specifically, the proposed method formulates a penalized negative log likelihood (P-NLL) cost function and proposes a majorize-minimize-based optimization method that removes speckle from OCT images. The optimization reduces to solving an iterative gradient descent problem. We demonstrate the usefulness of the proposed method by removing speckle in OCT images of uniform phantoms with varying scattering coefficients and human brain tissue.