Tri-Modal Medical Image Fusion and Denoising Based on BitonicX Filtering
Yuchan Jie, Xiaosong Li, Fuqiang Zhou, Tao Ye
Abstract
Medical image fusion integrates beneficial information from images of different modes and has been studied as a crucial auxiliary medical technology. However, current research mainly focused on two-modal fusion problems ( <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">e.g</i> ., CT-MRI, MRI-PET), and little attention was paid to tri-modal medical image fusion ( <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">e.g</i> ., CT-MR_T2-SPECT, MR_T2-MR_Gad-SPECT). In this paper, we present a tri-modal medical image fusion and denoising method that contains fusion models (I) and (II). A three-layer decomposition based on a bitonicX filter was designed for model (I). Particularly, the adaptive morphological gradient-coupled network (AM-PCNN) and the coupled neural P (CNP) system-based rules were proposed to fuse the detail and texture layers, respectively. The middle-fused result was obtained by synthesizing the pre-fused energy, texture, and detail layers in model (I). In model (II), for the fusion of the functional image and middle-fused result, we performed a "detail-energy" decomposition with the bitonicX filter. We also propose a gradient energy clarity operator and energy-based rules for fusing the detail and energy layers. The final fused result was obtained by adding the pre-fused detail layer and the energy layer in model (II). Numerous experiments were conducted to assess the fused images objectively and subjectively, demonstrating that the proposed method developed in this study surpasses the state-of-the-art methods in noise-free and noise fusions. Meanwhile, the average levels of the proposed method are 38.39%, 2.46%, 1.01%, 36.28%, 11.23%, 20.82%, 8.46%, and 2.52% higher than that of the compared methods evaluated by the metrics including Q <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">MI</sub> , Q <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">TE</sub> , Q <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">NCIE</sub> , Q <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">P</sub> , Q <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">CB</sub> , Q <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">SSIM</sub> , CNR, and CC, respectively. Our code is publicly available.