Repeatability of Multiparametric Prostate MRI Radiomics Features (1807.06089v2)

Abstract: In this study we assessed the repeatability of the values of radiomics features for small prostate tumors using test-retest Multiparametric Magnetic Resonance Imaging (mpMRI) images. The premise of radiomics is that quantitative image features can serve as biomarkers characterizing disease. For such biomarkers to be useful, repeatability is a basic requirement, meaning its value must remain stable between two scans, if the conditions remain stable. We investigated repeatability of radiomics features under various preprocessing and extraction configurations including various image normalization schemes, different image pre-filtering, 2D vs 3D texture computation, and different bin widths for image discretization. Image registration as means to re-identify regions of interest across time points was evaluated against human-expert segmented regions in both time points. Even though we found many radiomics features and preprocessing combinations with a high repeatability (Intraclass Correlation Coefficient (ICC) > 0.85), our results indicate that overall the repeatability is highly sensitive to the processing parameters (under certain configurations, it can be below 0.0). Image normalization, using a variety of approaches considered, did not result in consistent improvements in repeatability. There was also no consistent improvement of repeatability through the use of pre-filtering options, or by using image registration between timepoints to improve consistency of the region of interest localization. Based on these results we urge caution when interpreting radiomics features and advise paying close attention to the processing configuration details of reported results. Furthermore, we advocate reporting all processing details in radiomics studies and strongly recommend making the implementation available.

• The paper evaluated radiomics feature repeatability in small prostate tumors using mpMRI, highlighting the high sensitivity of feature stability to preprocessing configurations like normalization and filtering.
• Key findings include that image normalization did not consistently improve repeatability for all image types and that pre-filtering options significantly varied repeatability.
• The study underscores the need for detailed reporting of processing parameters in radiomics research to ensure reproducibility and identifies features with higher repeatability than volume measure as promising candidates.

Repeatability of Multiparametric Prostate MRI Radiomics Features

The paper, "Repeatability of Multiparametric Prostate MRI Radiomics Features," by Schwier et al. presents a comprehensive evaluation of radiomics feature repeatability for small prostate tumors using multiparametric MRI (mpMRI). Radiomics, which involves extracting quantitative features from medical images, holds promise for enhancing disease characterization and improving diagnostic and prognostic capabilities. However, the reliability of radiomics features, especially in clinical applications, demands repeatability across scans under stable conditions. This paper meticulously investigates the influence of multiple preprocessing parameters on the repeatability of these features.

Methodological Overview

The authors utilized a test-retest mpMRI dataset comprising 15 men with confirmed or suspected prostate cancer. A crucial aspect of their methodology involved assessing the repeatability of radiomics features under different preprocessing configurations, such as image normalization schemes, pre-filtering options, and texture computation dimensions. Feature extraction was performed using the pyradiomics package, adhering to definitions from the Imaging Biomarkers Standardization Initiative (IBSI). The researchers measured repeatability using the Intraclass Correlation Coefficient (ICC) and used tumor volume as a reference measure for ICC comparison.

Key Findings

Several critical insights emerged from the paper:

1. Sensitivity to Processing Parameters: Overall, feature repeatability exhibited high sensitivity to the preprocessing configuration. While many features achieved ICCs above 0.85, some configurations resulted in ICCs dropping below zero.
2. Normalization Impact: Contrary to expectations, image normalization did not consistently enhance repeatability across features. Normalization improved ICCs for ADC images in most cases, but not for T2w images.
3. Bin Width Influence: Differences in bin widths had a measurable impact on repeatability, though the effect was relatively small across most features.
4. Pre-filtering Variability: The application of pre-filtering options induced substantial variation in ICC values, with no clear consensus on which filters provided improvements across all features.
5. ROI Re-identification and Segmentation Consistency: Attempting to re-identify ROIs by image registration did not consistently enhance repeatability, illustrating the challenges posed by segmentation variability between scans.

Implications and Considerations

The findings underscore the complexities inherent in deriving stable radiomics features from mpMRI data. Given the variability introduced by preprocessing choices, caution is warranted when interpreting radiomics studies and selecting features for predictive models in prostate cancer imaging. The lack of universally stable preprocessing recommendations in the paper highlights the need for detailed reporting of processing parameters in radiomics research to facilitate reproducibility and comparison.

Furthermore, the results emphasize the importance of feature-specific preprocessing configurations to achieve reliable repeatability. Future studies may benefit from incorporating repeatability analysis into their feature selection processes to enhance the validity of radiomics-based biomarkers.

Conclusion and Future Directions

While this paper did not identify universally stable preprocessing configurations, it identified several features with notably higher repeatability than the reference volume measure. This highlights promising candidates for future research into prognostic feature development. The thorough analysis presented in this paper forms a basis for further investigation into feature stability across various clinical applications and underscores the necessity for detailed methodological transparency in radiomics studies.