Bone SPECT/CT image fusion based on the discrete Hermite transform and sparse representation

Nuclear medicine imaging techniques have become fundamental modalities for metabolic evaluation of the human body. Single Photon Emission Computed Tomography (SPECT) is among the most used tools for functional image analysis. When combined with anatomical imaging such as Computed Tomography (CT), functional information can be analyzed with better performance since lesion localization can be substantially improved. In this paper, we present a transform-based fusion scheme applied to bone SPECT/CT image analysis. The method uses the discrete Hermite transform for image feature coding. It consists of a powerful mathematical tool that locally projects an input image onto the set of Hermite polynomials, providing low frequency and detail information. The image fusion is performed by using coefficients obtained from the transformation. Two different fusion strategies were designed based on coefficient content. Low frequency coefficients are fused through a sparse representation mechanism meanwhile coefficients with fine details are combined using local directional information and variance. Prior knowledge based on segmentation of bone tissues is included in the SR-based fusion rule. The final fused image is recovered by using the inverse transform from the resulting fused coefficients. The approach was evaluated with registered bone SPECT and CT images. It was also compared with other approaches of the state of art demonstrating its feasibility. We also present an evaluation of our method with brain tissues by using common images found in literature. Frequently used metrics were considered to quantitatively assess the performance of the proposed method.

Automated summary

This paper introduces a transform-based fusion scheme for bone SPECT/CT image analysis, using the Hermite transform for image feature coding. Two different fusion strategies were designed based on coefficient content, and the final fused image was recovered using the inverse transform.