多模态医学图像融合图像质量评估

目前大多数的质量评估算法都应用于自然图像的融合场景，缺乏专用评价的医学图像数据集及多模态医学融合图像的质量评估算法。针对此问题，利用17种经典的医学图像融合方法构建医学图像主观数据集，解决无专用评价数据集的问题;提出一种基于色彩相似度(color similarity，CS)和信息相似度(information similarity，IS)的客观医学图像质量评价方法。将CS模块用于测量局部颜色失真，在传统的池化层上添加背景分离模块使其适用于医学图像多背景干扰特性;将IS模块用于衡量信息失真，改进图像熵的计算方法，添加过滤模块以剔除图像噪声。实验结果表明，所提方法的预测值和主观数据集客观评分具有更好的一致性，更符合人类视觉系统。

Quality assessment of multimodal medical image fusion images

Currently, the majority of quality assessment algorithms are primarily applied to natural image fusion scenarios, resulting in a lack of dedicated evaluation datasets and quality assessment algorithms for multimodal medical fused images. To address these issues, this paper constructs a subjective dataset of medical images using 17 classical medical image fusion methods and proposes an objective medical image quality assessment method based on color similarity (CS) and information similarity (IS). The CS module is utilized to measure local color distortion, and a background separation module is incorporated into the traditional pooling layer to accommodate the multi-background interference characteristics specific to medical images. Additionally, the IS module is employed to evaluate information distortion by improving the calculation method of image entropy and introducing a filtering module for noise removal. Experimental results demonstrate that the proposed evaluation method yields better consistency with objective scores from the subjective dataset and aligns more closely with human visual perception.