Congenital diaphragmatic hernia: automatic lung and liver MRI segmentation with nnU-Net, reproducibility of pyradiomics features, and a machine learning application for the classification of liver herniation.

Prenatal assessment of lung size and liver position is essential to stratify congenital diaphragmatic hernia (CDH) fetuses in risk categories, guiding counseling, and patient management. Manual segmentation on fetal MRI provides a quantitative estimation of total lung volume and liver herniation. However, it is time-consuming and operator-dependent. In this study, we utilized a publicly available deep learning (DL) segmentation system (nnU-Net) to automatically contour CDH-affected fetal lungs and liver on MRI sections. Concordance between automatic and manual segmentation was assessed by calculating the Jaccard coefficient. Pyradiomics standard features were then extracted from both manually and automatically segmented regions. The reproducibility of features between the two groups was evaluated through the Wilcoxon rank-sum test and intraclass correlation coefficients (ICCs). We finally tested the reliability of the automatic-segmentation approach by building a ML classifier system for the prediction of liver herniation based on support vector machines (SVM) and trained on shape features computed both in the manual and nnU-Net-segmented organs. We compared the area under the classifier receiver operating characteristic curve (AUC) in the two cases. Pyradiomics features calculated in the manual ROIs were partly reproducible by the same features calculated in nnU-Net segmented ROIs and, when used in the ML procedure, to predict liver herniation (both AUC around 0.85).          Conclusion: Our results suggest that automatic MRI segmentation is feasible, with good reproducibility of pyradiomics features, and that a ML system for liver herniation prediction offers good reliability.          Trial registration: https://clinicaltrials.gov/ct2/show/NCT04609163?term=NCT04609163&draw=2&rank=1 ; Clinical Trial Identification no. NCT04609163. What is Known: • Magnetic resonance imaging (MRI) is crucial for prenatal congenital diaphragmatic hernia (CDH) assessment. It enables the quantification of the total lung volume and the extent of liver herniation, which are essential for stratifying the severity of CDH, guiding counseling, and patient management. • The manual segmentation of MRI scans is a time-consuming process that is heavily reliant upon the skill set of the operator. What is New: • MRI lung and liver automatic segmentation using the deep learning nnU-Net system is feasible, with good Jaccard coefficient values and satisfactory reproducibility of pyradiomics features compared to manual results. • A feasible ML system for predicting liver herniation could improve prenatal assessments and CDH patient management.

Robustness of Radiomic Features: Two-Dimensional versus Three-Dimensional MRI-Based Feature Reproducibility in Lipomatous Soft-Tissue Tumors.

This retrospective study aimed to compare the intra- and inter-observer manual-segmentation variability in the feature reproducibility between two-dimensional (2D) and three-dimensional (3D) magnetic-resonance imaging (MRI)-based radiomic features. The study included patients with lipomatous soft-tissue tumors that were diagnosed with histopathology and underwent MRI scans. Tumor segmentation based on the 2D and 3D MRI images was performed by two observers to assess the intra- and inter-observer variability. In both the 2D and the 3D segmentations, the radiomic features were extracted from the normalized images. Regarding the stability of the features, the intraclass correlation coefficient (ICC) was used to evaluate the intra- and inter-observer segmentation variability. Features with ICC > 0.75 were considered reproducible. The degree of feature robustness was classified as low, moderate, or high. Additionally, we compared the efficacy of 2D and 3D contour-focused segmentation in terms of the effects of the stable feature rate, sensitivity, specificity, and diagnostic accuracy of machine learning on the reproducible features. In total, 93 and 107 features were extracted from the 2D and 3D images, respectively. Only 35 features from the 2D images and 63 features from the 3D images were reproducible. The stable feature rate for the 3D segmentation was more significant than for the 2D segmentation (58.9% vs. 37.6%, p = 0.002). The majority of the features for the 3D segmentation had moderate-to-high robustness, while 40.9% of the features for the 2D segmentation had low robustness. The diagnostic accuracy of the machine-learning model for the 2D segmentation was close to that for the 3D segmentation (88% vs. 90%). In both the 2D and the 3D segmentation, the specificity values were equal to 100%. However, the sensitivity for the 2D segmentation was lower than for the 3D segmentation (75% vs. 83%). For the 2D + 3D radiomic features, the model achieved a diagnostic accuracy of 87% (sensitivity, 100%, and specificity, 80%). Both 2D and 3D MRI-based radiomic features of lipomatous soft-tissue tumors are reproducible. With a higher stable feature rate, 3D contour-focused segmentation should be selected for the feature-extraction process.

Making Radiomics More Reproducible across Scanner and Imaging Protocol Variations: A Review of Harmonization Methods.

Radiomics converts medical images into mineable data via a high-throughput extraction of quantitative features used for clinical decision support. However, these radiomic features are susceptible to variation across scanners, acquisition protocols, and reconstruction settings. Various investigations have assessed the reproducibility and validation of radiomic features across these discrepancies. In this narrative review, we combine systematic keyword searches with prior domain knowledge to discuss various harmonization solutions to make the radiomic features more reproducible across various scanners and protocol settings. Different harmonization solutions are discussed and divided into two main categories: image domain and feature domain. The image domain category comprises methods such as the standardization of image acquisition, post-processing of raw sensor-level image data, data augmentation techniques, and style transfer. The feature domain category consists of methods such as the identification of reproducible features and normalization techniques such as statistical normalization, intensity harmonization, ComBat and its derivatives, and normalization using deep learning. We also reflect upon the importance of deep learning solutions for addressing variability across multi-centric radiomic studies especially using generative adversarial networks (GANs), neural style transfer (NST) techniques, or a combination of both. We cover a broader range of methods especially GANs and NST methods in more detail than previous reviews.

Deep neural network-based approach to improving radiomics analysis reproducibility in liver cancer: effect on image resampling.

Objectives.To test the effect of traditional up-sampling slice thickness (ST) methods on the reproducibility of CT radiomics features of liver tumors and investigate the improvement using a deep neural network (DNN) scheme.Methods.CT images with ≤ 1 mm ST in the public dataset were converted to low-resolution (3 mm, 5 mm) CT images. A DNN model was trained for the conversion from 3 mm ST and 5 mm ST to 1 mm ST and compared with conventional interpolation-based methods (cubic, linear, nearest) using structural similarity (SSIM) and peak-signal-to-noise-ratio (PSNR). Radiomics features were extracted from the tumor and tumor ring regions. The reproducibility of features from images converted using DNN and interpolation schemes were assessed using the concordance correlation coefficients (CCC) with the cutoff of 0.85. The paired t-test and Mann-Whitney U test were used to compare the evaluation metrics, where appropriate.Results.CT images of 108 patients were used for training (n = 63), validation (n = 11) and testing (n = 34). The DNN method showed significantly higher PSNR and SSIM values (p < 0.05) than interpolation-based methods. The DNN method also showed a significantly higher CCC value than interpolation-based methods. For features in the tumor region, compared with the cubic interpolation approach, the reproducible features increased from 393 (82%) to 422(88%) for the conversion of 3-1 mm, and from 305(64%) to 353(74%) for the conversion of 5-1 mm. For features in the tumor ring region, the improvement was from 395 (82%) to 431 (90%) and from 290 (60%) to 335 (70%), respectively.Conclusions.The DNN based ST up-sampling approach can improve the reproducibility of CT radiomics features in liver tumors, promoting the standardization of CT radiomics studies in liver cancer.

Assessing the reproducibility of CBCT-derived radiomics features using a novel three-dimensional printed phantom.

PURPOSE: Radiomics modeling is an exciting avenue for enhancing clinical decision making and personalized treatment. Radiation oncology patients often undergo routine imaging for position verification, particularly using LINAC-mounted cone beam computed tomography (CBCT). The wealth of imaging data collected in modern radiation therapy presents an ideal use case for radiomics modeling. Despite this, texture feature (TF) calculation can be limited by concerns over feature stability and reproducibility; in theory, this issue is compounded by the relatively poor image quality of CBCT, as well as variation of acquisition and reconstruction parameters. METHODS: In this study, we developed and validated a novel three-dimensional (3D) printed phantom for evaluating CBCT-based TF reliability. The phantom has a cylindrical shape (22 cm diameter and 25.5 cm height) with five inner inserts designed to hold custom-printed rods (1 cm diameter and 10-20 cm height) of various materials, infill shapes, and densities. TF reproducibility was evaluated across and within three LINACs from a single vendor using sets of three consecutive CBCT taken with the head, thorax, and pelvis clinical imaging protocols. PyRadiomics was used to extract a standard set of TFs from regions of interest centered on each rod. Two-way mixed effects absolute agreement intra-class correlation coefficient (ICC) was used to evaluate TF reproducibility, with features showing ICC values above 0.9 considered robust if their Bonferroni-corrected p-value was below 0.05. RESULTS: A total of 63, 87, and 83 features exhibited test-retest reliability for the head, thorax, and pelvis imaging protocols respectively. When assessing stability between discreet imaging sessions on the same LINAC, these numbers were reduced to 5, 63, and 70 features, respectively. The thorax and pelvis protocols maintained a rich candidate feature space in inter-LINAC analysis with 61 and 65 features, respectively, exceeding the ICC criteria. Crucially, no features were deemed reproducible when compared between protocols. CONCLUSIONS: We have developed a 3D phantom for consistent evaluation of TF stability and reproducibility. For LINACs from a single vendor, our study found a substantial number of features available for robust radiomics modeling from CBCT imaging. However, some features showed variations across LINACs. Studies involving CBCT-based radiomics must preselect features prior to their use in clinical-based models.

Breast DCE-MRI radiomics: a robust computer-aided system based on reproducible BI-RADS features across the influence of datasets bias and segmentation methods.

PURPOSE: A highly accurate and robust computer-aided system based on quantitative high-throughput Breast Imaging Reporting and Data System (BI-RADS) features from dynamic contrast-enhanced magnetic resonance imaging (DCE-MRI) can drive the success of radiomic applications in breast cancer diagnosis. We aim to build a stable system with highly reproducible radiomics features, which can make diagnostic performance independent of datasets bias and segmentation methods. METHOD: We applied a dataset of 267 patients including 136 malignant and 131 benign tumors from two MRI manufacturers, where 211 cases from a Philips system and 55 cases from a GE system. First, manual annotations, 3D-Unet and 2D-Unet were applied as different segmentation methods. Second, we designed and extracted 3172 features from six modalities of DCE-MRI based on BI-RADS. Third, the feature selection was conducted. Between-class distance was utilized to eliminate the effect of dataset bias caused by two machines. Concordance correlation coefficient, intraclass correlation coefficient and deviation were employed to evaluate the influence of three segmentation methods. We further eliminated features redundancy using genetic algorithm. Finally, three classifiers including support vector machine (SVM), the bagged trees and K-Nearest Neighbor were evaluated by their performance for diagnosing malignant and benign tumors. RESULTS: A total of 246 features were preserved to have high stability and reproducibility. The final feature set showed the robust performance under these factors and achieved the area under curve of 0.88, the accuracy of 0.824, the sensitivity of 0.844, the specificity of 0.807 in differentiating benign and malignant tumors with the SVM classifier using manually segmentation results. CONCLUSION: The final selected 246 features are reproducible and show little dependence on segmentation methods and data perturbation. The high stability and effectiveness of diagnosis across these factors illustrate that the preserved features can be used for prognostic analysis and help radiologists in the diagnosis of breast cancer.

Clinical impact of variability on CT radiomics and suggestions for suitable feature selection: a focus on lung cancer.

BACKGROUND: Radiomics suffers from feature reproducibility. We studied the variability of radiomics features and the relationship of radiomics features with tumor size and shape to determine guidelines for optimal radiomics study. METHODS: We dealt with 260 lung nodules (180 for training, 80 for testing) limited to 2 cm or less. We quantified how voxel geometry (isotropic/anisotropic) and the number of histogram bins, factors commonly adjusted in multi-center studies, affect reproducibility. First, features showing high reproducibility between the original and isotropic transformed voxel settings were identified. Second, features showing high reproducibility in various binning settings were identified. Two hundred fifty-two features were computed and features with high intra-correlation coefficient were selected. Features that explained nodule status (benign/malignant) were retained using the least absolute shrinkage selector operator. Common features among different settings were identified, and the final features showing high reproducibility correlated with nodule status were identified. The identified features were used for the random forest classifier to validate the effectiveness of the features. The properties of the uncalculated feature were inspected to suggest a tentative guideline for radiomics studies. RESULTS: Nine features showing high reproducibility for both the original and isotropic voxel settings were selected and used to classify nodule status (AUC 0.659-0.697). Five features showing high reproducibility among different binning settings were selected and used in classification (AUC 0.729-0.748). Some texture features are likely to be successfully computed if a nodule was larger than 1000 mm3. CONCLUSIONS: Features showing high reproducibility among different settings correlated with nodule status were identified.

Stability of radiomic features of apparent diffusion coefficient (ADC) maps for locally advanced rectal cancer in response to image pre-processing.

Quantitative imaging features (radiomics) extracted from apparent diffusion coefficient (ADC) maps of rectal cancer patients can provide additional information to support treatment decision. Most available radiomic computational packages allow extraction of hundreds to thousands of features. However, two major factors can influence the reproducibility of radiomic features: interobserver variability, and imaging filtering applied prior to features extraction. In this exploratory study we seek to determine to what extent various commonly-used features are reproducible with regards to the mentioned factors using ADC maps from two different clinics (56 patients). Features derived from intensity distribution histograms are less sensitive to manual tumour delineation differences, noise in ADC images, pixel size resampling and intensity discretization. Shape features appear to be strongly affected by delineation quality. On the whole, textural features appear to be poorly or moderately reproducible with respect to the image pre-processing perturbations we reproduced.

Radiomic Feature Robustness and Reproducibility in Quantitative Bone Radiography: A Study on Radiologic Parameter Changes.

The purpose of this study was to investigate the robustness of different radiography radiomic features over different radiologic parameters including kV, mAs, filtration, tube angles, and source skin distance (SSD). A tibia bone phantom was prepared and all imaging studies was conducted on this phantom. Different radiologic parameters including kV, mAs, filtration, tube angles, and SSD were studied. A region of interest was drawn on the images and many features from different feature sets including histogram, gradient, run-length matrix, co-occurrence matrix, autoregressive model, and wavelet derived parameters were extracted. All radiomic features were categorized based on coefficient of variation (COV). Bland-Altman analysis also was used to evaluate the mean, standard deviation, and upper/lower reproducibility limits for radiomic features in response to variation in each testing parameters. Results on COV in all features showed that 22%, 34%, and 45% of features were most robust (COV ≤ 5%) against kV, mAs, and SSD respectively and there was no robust features against filtration and tube angle. Also, all features (100%) and 76% of which showed large variations (COV > 20%) against filtrations and tube angle respectively. Autoregressive model feature set has no robust features against all radiologic parameters. Features including sum-average, sum-entropy, correlation, mean, and percentile (50, 90, and 99) belong to co-occurrence matrix and histogram feature sets were found as most robust features. Bland-Altman analysis showed the high reproducibity of some feature sets against radiologic parameter changes. The results presented here indicated that radiologic parameters have great impacts on radiomic feature values and caution should be taken into account when work with these features. In quantitative bone studies, robust features with low COV can be selected for clinical or research applications. Reproducible features also can be obtained using Bland-Altman analysis.