Comprehensive Approach to Mri Image Preprocessing Based on Filtering and Quality Assessment

Magnetic resonance imaging (MRI) occupies an important place in modern diagnostics due to its high informativeness, but technical features of image acquisition, artefacts and noise often reduce the visual accuracy of the results. This article addresses the problem of improving the quality of MRI images by applying different filtering techniques. The study focuses on a comparative analysis of three filtering methods: Laplace filter for edge extraction, Non-Local Means (NLM) algorithm for texture preserving noise reduction and Unsharp Mask for sharpening. The key challenge is to find a balance between removing noise, preserving diagnostically relevant details and improving visual sharpness. Quantitative metrics (MSE, PSNR, SSIM) and subjective assessment of visual sharpness of images were used to evaluate the effectiveness of the methods. Experiments were performed on heterogeneous data comprising MR images of PD, T1, T2 imaging modes with FS, SE, FSE and TSE pulse sequences obtained from different scanners. As a result, a combined preprocessing approach was developed to adapt filtering to the specificity of imaging modes, maintaining a balance between noise suppression and preservation of diagnostically relevant information. Special attention is paid to the adaptation of methods to heterogeneous data, which is relevant for clinical practice, where images can be acquired on equipment of different manufacturers. The proposed approach allows standardising preprocessing of MRI data for further analysis in intelligent diagnostic systems, including machine learning algorithms. The study highlights the importance of combining numerical metrics and expert judgement to develop clinically relevant solutions. And the presented results demonstrate the effectiveness of the proposed method and its potential for use in automated MRI data analysis systems