The spatial statistics of structural magnetic resonance images: application to post-acquisition quality assessment of brain MRI images

Abstract

Popular algorithms for quality evaluation of medical images are generic, global and distortion-specific. Performance is limited by complexity introduced by presence of disease signatures such as lesions. Application of classical statistics ignores spatial dependency and the unique image attributes in different imaging modalities. We propose a new no-reference method that overcomes some draw-backs of current algorithms and correlate with human visual system in terms of 'fidelity', 'usefulness' and 'naturalness'. We define the sample space as local entropy at pixel locations in a MRI volume data. Furthermore we segment each slice into eight equal angular segments and partition the sample space into low and high entropy regions. These regions, in a MRI volume data, are regarded as regionalized random variables which exhibit distinct and organized geographic patterns. Our proposal has 'usefulness' because it exploits similarity in geometry of human anatomy to build quality models from 250 brain MRI images of different subjects sourced from the Alzheimer's disease neuroimaging initiative (ADNI) database. 'Fidelity' is impacted by comparing quality models built from ADNI images with quality features extracted from a test image. 'Naturalness' comes from application of spatial statistics to describe distinct geographic patterns of different anatomic structures of brain. Quality scores computed using law of total probability takes into account presence of lesions in MRI data. Partition of feature space encourages focus on region of interest. Experimental results show that our proposal correlates with different levels of degradation and can improve discernment of a physician or a trained reader towards reliable diagnosis.

A systematic review of (semi-)automatic quality control of T1-weighted MRI scans

“…Cross-reference searching of the included articles resulted in the identification of eight more articles, of which six underwent full text review of which four were excluded from further quality assessment due to 1) assessing quality of other MR sequences (n = 2), 2) assessing quality of a file structure (n = 1) and 3) assessing quality of phantom images (n = 1). Ultimately, a total of 18 articles were included (Alfaro-Almagro et al, 2018; Bottani et al, 2022; Esteban et al, 2017; Fantini et al, 2021; Gedamu et al, 2008; Ikushima et al, 2022; Jang et al, 2018; Keshavan et al, 2019; Kim et al, 2019; Küstner et al, 2018; Mortamet et al, 2009; Osadebey et al, 2017a; Osadebey et al, 2017b; Osadebey et al, 2018; Pizarro et al, 2016; Sujit et al, 2019; White et al, 2018; Woodard and Carley-Spencer, 2006). Relevant information regarding dataset, benchmark, and performance measures is summarized in Table 1.…”

No-Reference Image Quality Assessment Of Large FLAIR MRI Datasets

“…A series of studies in the field of MR IQA includes three applications produced by Osadebey et al [46] [47] [48]. The first work considered is an IQA approach that utilizes spatial statistics to describe quality [47]. The focus of this study is to correlate artificial distortion with image quality features.…”

Image Quality Evaluation in Clinical Research: A Case Study on Brain and Cardiac MRI Images in Multi-Center Clinical Trials

“…The report in [42] predict image quality by casting the relationship between entropy and classical image quality attributes on Bayesian framework. Another report [43] computes image quality by using three separate geo-spatial feature vectors extracted from a test image to standardize corresponding Gaussian distributed quality models. Other recent reports assess image quality based on how subject motion during acquisition bias structural information and metrics derived from the image [44] – [48] .…”

“…Even for brain MRI images, the performance of these algorithms can be significantly limited by underestimation of noise level when the number of background voxels are limited or corrupted by artifacts [54] , [55] . The need for a large population to extract relevant features for the construction of quality model can be regarded as a drawback for the reports in [42] and [43] . This drawbacks makes it difficult to achieve the much desired consistent quality evaluation required in good clinical practice.…”