Impact of image quality on radiomics applications

Abstract

Radiomics features extracted from medical images have been widely reported to be useful in the patient specific outcome modeling for variety of assessment and prediction purposes. Successful application of radiomics features as imaging biomarkers, however, is dependent on the robustness of the approach to the variation in each step of the modeling workflow. Variation in the input image quality is one of the main sources that impacts the reproducibility of radiomics analysis when a model is applied to broader range of medical imaging data. The quality of medical image is generally affected by both the scanner related factors such as image acquisition/reconstruction settings and the patient related factors such as patient motion. This article aimed to review the published literatures in this field that reported the impact of various imaging factors on the radiomics features through the change in image quality. The literatures were categorized by different imaging modalities and also tabulated based on the imaging parameters and the class of radiomics features included in the study. Strategies for image quality standardization were discussed based on the relevant literatures and recommendations for reducing the impact of image quality variation on the radiomics in multi-institutional clinical trial were summarized at the end of this article.

Robustness of radiomic features in 123I-ioflupane-dopamine transporter single-photon emission computer tomography scan

“…However, it is known that different image quality is a main source of irreproducibility and concern in radiomics analysis [ 35 ]. In general, there are two sources of variation in the quality of medical images, machine-dependent (e.g., acquisition and image reconstruction/post-processing) and patient-dependent factors (e.g., movement artifacts).…”

Identification of texture MRI brain abnormalities on first-episode psychosis and clinical high-risk subjects using explainable artificial intelligence

“…We selected the features that express opposite properties at microscale level (Entropy vs Energy and Homogeneity vs Contrast). Second order statistics of Entropy measure the arrangement of voxel gray-level intensities, depend on spatial relationship between gray-level intensities of the voxels [ 31 ], and have shown significant results in previous study [ 29 ]. The definitions of these features were as follows: GLCM-contrast reflects local variations in the GLCM; GLCM-energy reflects uniformity of gray-level voxel pairs; GLCM-entropy reflects randomness of gray-level voxel pairs; and GLCM-homogeneity reflects homogeneity of gray-level voxel pairs [ 61 , 62 ]; finally GLCM-sum of entropy and difference of entropy reflect second order statistics of differentiation of gray-level distribution GLCM.…”

“…Radiomics texture features are able to quantify the hidden patterns between voxel intensities and the spatial distribution of these patterns across brain regions. Meaningful comparison of texture feature results between different subjects is possible, when sMR images of the brain with similar resolution and noise levels are used, a common quantization method and the same number of gray levels in all quantized images [ 30 , 31 ]. Radiomics texture features with their potential as image-based biomarkers have been widely used across several studies, like for cancer identification [ 32 ], Alzheimer’s [ 33 ] and Parkinson’s disease [ 34 ] as neurodegenerative diseases, major depression [ 35 ], and schizophrenia [ 36 , 37 ].…”

“…Instead of applying conventional methods which show greater performance than deep neural networks [ 42 ], we employed an innovative approach that addresses a frequent concern about artificial intelligence methods, i.e., the explainability of results. Our goal is to gain insights into the examined disorders using radiomics texture features and explainable AI which achieve better performance in outcome modeling instead of statistical analysis of the radiomics texture features or the deep neural networks solely [ 31 ]. Our proposed algorithm integrates the complexity of the deep neural networks with the explainability of these networks as introduced by Bach et al [ 43 ].…”

Deep learning in MRI‐guided radiation therapy: A systematic review

“…Variability in MR imaging characteristics such as field strength, scanner manufacturer, pulse sequence, ROI or contour quality, and the feature extraction method can result in different features being significant. This variability can largely be mitigated by normalizing the data to a reference MRI and including data from multiple sources 130 …”