CMMIQA: a prompt-driven cross-modality multi-organ medical image quality assessment model

Background: Three-dimensional (3D) medical imaging technology plays an important role in clinical diagnosis, whose image quality is the cornerstone of effective diagnosis and optimization of imaging equipment. Therefore, the assessment of medical images quality has received a wide range of attention. In this paper, we are committed to developing a way that can comprehensively assess the quality of medical images of different modality and organs. Methods: We first used real rather than synthetic data to create the first 3D Cross-Modality Multi-Organ Medical Image Quality Assessment Database (CMMIQA-DB) for assessment of computed tomography (CT) and magnetic resonance imaging (MRI) image quality. Based on the database, a Cross-Modality Multi-Organ Medical Image Quality Assessment model with hybrid-convolution (Hybrid-conv) and aware prompts was proposed. This model uses a Hybrid-conv network to extract features from a variety of modality images. Simultaneously, quality regression training is performed with the help of aware prompts, and the quality prediction scores of different modality images are finally produced. Results: The experimental results of Spearman rank correlation coefficient (SRCC), Pearson linear correlation coefficient (PLCC), and root mean square error (RMSE) on reached 0.7476, 0.7153 and 0.3345 respectively, and the experimental results of 0.8618, 0.8705 and 0.1589 on manually labeled two-dimensional (2D) dataset respectively. Conclusions: Experimental results show that this model is superior to existing no-reference image quality assessment (IQA) methods and medical IQA models, while being able to transfer to other 2D and 3D medical image datasets as a foundation.