Cardiac Cine MRI Radiomics: Analysis of Feature Reproducibility and Evaluation of Variations in the Context of Heart Failure

Cardiac Cine MRI Radiomics: Analysis of Feature Reproducibility and Evaluation of Variations in the Context of Heart Failure | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Research Article Cardiac Cine MRI Radiomics: Analysis of Feature Reproducibility and Evaluation of Variations in the Context of Heart Failure Lea Ter-Minassian, Matthias ivantsis, Markus Hüllebrand, Hannu Zhang, and 8 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-6912341/v1 This work is licensed under a CC BY 4.0 License Status: Posted Version 1 posted You are reading this latest preprint version Abstract Purpose This work investigates the reproducibility of radiomics features of cardiac cine MRI and their applicability for heart failure (HF) characterization. Material and methods Short-axis ECG-triggered cine CMR (cardiac MRI) sequences were acquired with different spatiotemporal resolutions. Segmentation was performed by two experts delineating left ventricle blood pool (LVBP), left ventricle myocardium (LVM), and right ventricle blood pool (RVBP). Intra- and interobserver segmentation agreement was evaluated with Dice Score, Hausdorff and Average Contour Distance. Feature reproducibility was assessed via intraclass correlation to evaluate the influence of image resolutions and observer variability. Feature curves of robust features were compared between HF subtypes and age-matched controls. Results We included twelve healthy volunteers (median age M: 28, interquartile range IQR: 9) for reproducibility analysis, and 19 controls without HF (M: 62, IQR: 11), 22 HF patients with preserved ejection fraction (HFpEF) (M: 77, IQR: 8),17 HF patients with mildly reduced EF (HFmrEF) (M: 73, IQR: 10), and 15 HF patients with reduced EF (HFrEF) (M: 62, IQR: 14). On high resolution respectively 61 and 63 of 64 features for observer 1 and 2 showed good to excellent intraobserver reproducibility. Higher spatiotemporal resolution had a positive impact on reproducibility. Average LVBP and LVM feature values differ between controls, HFpEF, HFmrEF, and HFrEF but value ranges overlap. Conclusions Most radiomics features showed good to excellent reproducibility, the spatiotemporal resolution being the main cause of variability followed by observer experience. Combined assessment of LVM and LVBP features could enable CMR-based HF subtype classification. Cardiac MRI Reproducibility Variability Radiomics Cine Heart failure Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 1. Introduction Radiomics can help image-based assessment of atherosclerosis and infarcts ( 1 – 4 ), by quantifying shape and intensity patterns of segmented structures.To ensure reproducibility, the IBSI promotes standardization across tools like PyRadiomics ( 5 ), butfeature values also depend on voxel size, object positioning ( 6 ), intensity sampling ( 7 ), and preprocessing—areas often lacking in cardiac MRI studies ( 8 ). MRI intensities also vary with imaging settings and segmentation accuracy ( 9 ). Temporal sampling is crucial for the reproducibility of conventional diagnostic parameters, such as ventricular volumes ( 10 ) and strain ( 11 , 12 ). A few studies examined the effects of imaging procedure variations on radiomics feature reproducibility in cardiac cine MRI. Schofield et al. analyzed parameter variations as a proxy for multi-site or multi-vendor imaging ( 13 ). Their study found good reproducibility of features like intensity mean, standard deviation, skewness, and kurtosis in the myocardium. A test-retest study by Jang et al. identified 102 reproducible radiomics texture features, ( 14 ). They investigated the robustness of texture features under different acquisition parameters, including spatial resolution, and parallel imaging technique ( 15 ). For cine imaging, spatial resolution had the most significant effect on feature reproducibility. Most studies ( 6 , 7 ) assess reproducibility using intra-class correlation coefficients (ICCs) with varying thresholds (see Supplement Criteria for reproducibility). This paper investigates four primary aspects of radiomics feature reproducibility: The influence of spatiotemporal sampling on segmentation. The reproducibility of features with respect to segmentation. The reproducibility of features concerning spatial and temporal sampling. Value ranges of reproducible features in different heart failure subtypes compared to age-matched controls. We evaluate cardiac shape parameters, quadrature-filter-based motion vectors and texture features from previous studies to capture structural and functional information from 4D image sequences ( 3 , 14 , 16 , 17 ). 2. Materials and Methods Ethic statements This retrospective study was conducted in accordance with the Helsinki declaration. We declare no conflict of interest. The heart failure patient data originates from the EMPATHY-HF registry (German Clinical Trials Register ID: DRKS00015615). The image analysis of the healthy volunteers had been approved by the ethics committee of Charité Universitätsmedizin Berlin (application number EA2/073/16). All participants provided written consent. Dataset All imaging protocols included short-axis ECG-triggered cine CMR images. For the 12 healthy volunteers (6 females) we analyzed two parameterizations for spatiotemporal sampling to assess the influence of imaging parameters on the reproducibility of the proposed radiomics features (see Table 1 ). These datasets have previously been analysed for the optimization of protocols for strain analysis ( 11 ). They had a median age of 28 years (IQR 9) and normal left ventricular ejection fraction LVEF (median 60%; IQR 3 [59, 62]). The subset used for the inter-observer analysis consisted of eight volunteers (50% female), with median age 31 (IQR 7) and median LVEF of 62% (IQR 4 [59,63]). For the analysis of the radiomics feature suitability for the differentiation of HF subtypes, we used datasets from the EMPATHY-HF cohort ( https://drks.de/search/de/trial/DRKS00015615 ) [19]. The subgroups are: 15 patients with heart failure with reduced ejection fraction (HFrEF) (3 female) with a median age of 63 (IQR 14 [55, 69]) and a median LVEF of 34% (IQR 7 [30, 37]) 17 patients with heart failure with mild reduced ejection fraction (HFmrEF) (4 female) with a median age of 73 (IQR 10 [64, 74]) and a median LVEF of 44% (IQR 3 [42, 45]) 22 patients with heart failure with preserved ejection fraction (HFpEF) (11 female) with median age of 78 (IQR 8 [73, 81]) and a median LVEF of 59% (IQR 12 [54, 66]) 19 controls (9 female) with a median age of 62 (IQR 11 [56, 67]) and a median LVEF of 64% (IQR 8 [62, 68]) The automatically segmented EMPATHY datasets were processed to evaluate if the features deemed robust regarding imaging and expert segmentation, are sensitive to pathological changes. Table 1 Overview of imaging parameters. Parameter Healthy Volunteers HF Patients and age-matched controls Scanner Philips Achieva 1.5T Imaging sequence ECG-gated bSSFP cine Repetition time [ms] 4.13 ± 0.02 ** 2.57 ± 0.06 * 3.42 Echo time [ms] 2.07 ± 0.1 ** 1.29 ± 0.03 * 1.624–1.761 Flip angle [°] 60 In-plane resolution (isotropic) [mm] 1.39 ± 0.03 ** 1.43 ± 0.12 * 1.45–1.63 Slice thickness [mm] 5** / 10* 8 Slice spacing [mm] 15.49 ± 1.66 ** 17.66 ± 1.66 * ࿼ Temporal resolution [phases/cycle] 20* / 50** 50 K-space sampling cartesian Heart rate [bpm] 67.0 ± 8.07 ** 64.5 ± 8.43 * 47–85 Table 1 : Overview of imaging parameters considered for feature robustness analysis. *: “low-resolution” parameters **: “high-resolution” parameters Image Preprocessing We rescaled the intensities to a range of [0, 4095] and resampled the images with a cubic B-spline to a voxel-size of 1.5x1.5 mm 2 (highest resolution in the dataset). For image intensity discretization, we used the PyRadiomics default value with a bin size of 25. Image Segmentation Two experts (experienced radiologist O1, and medical trainee O2) independently segmented the volunteer sequences using a semi-automatic tool ( 18 ). A neural network automatically labeled the left ventricular blood pool (LVBP), including trabeculation and papillary muscles, the left ventricular myocardium (LVM), and the right ventricular blood pool (RVBP). The experts inspected and refined the resulting contours. For the calculation of voxel-based radiomics features, contours were rasterized to image masks (ROIs). O1 and O2 segmented the high- and low-resolution images of 8 volunteers twice to assess intra- and inter-observer variation of segmentations and derived features. For these and 4 supplementary healthy volunteers, O2 also segmented supplementary spatiotemporal resolution image sequences. HF patient and control datasets were not post-processed after automatic segmentation (see Fig. 1). Feature Extraction The first feature group contains radiomics first-order and texture features which were considered reproducible in at least three previous publications (see Table 2 ). Values were calculated using the PyRadiomics package for all rasterized ROIs and time points. The second group consists of the shape features, area, elongation, and sphericity, acting as surrogates for clinical parameters of the left ventricle ( 19 ). The third group contains ten cardio-specific features that describe anatomy, motion and the interaction between left and right ventricles. Strain features calculated with quadrature filters for motion analysis characterize local deformation ( 20 ). The Minkowski-Bouligand Dimension describes the complexity of the endocardial border ( 21 ), which can provide insights about trabeculation. Figure 1: Reproducibility analysis pipeline. Influence of the segmentation (intra- and interobserver variability) and resolution is based on the scans of healthy volunteers. Segmentation contours are converted to segmentation masks in the resampled images. Feature curves of age-matched controls are compared with those of heart failure patients to analyse the influence of pathological alterations vs. the sampling methods and observers. Table 2 Overview of extracted features, including previous publication in which features have been found to be robust. Type Feature Description Found robust in these studies First-order Energy Measures the degree of intensity inhomogeneity Alis 2021( 16 ), Jang 2020 ( 14 ), Jang 2021( 15 ), Raisi-Estrabragh 2020( 17 ) Mean The mean of the intensity histogram Alis 2021( 16 ), Raisi-Estrabragh 2020 ( 17 ), Schofield 2019 ( 13 ) Total Energy Measures the degree of intensity inhomogeneity scaled by the voxel volume Alis 2021( 16 ), Jang 2020 ( 14 ), Jang 2021( 15 ), Raisi-Estrabragh 2020( 17 ) 90th Percentile The 90th percentile of the intensity histogram Alis 2021( 16 ), Raisi-Estrabragh 2020 ( 17 ), Mancio 2021 ( 3 ) Texture (GLCM) Joint Entropy Measures the randomness of neighboring intensities Alis 2021 ( 16 ), Jang 2020 ( 14 ), Jang 2021( 15 ), Raisi-Estrabragh 2020 ( 17 ) Joint Energy Energy measures the degree of intensity homogeneity within a ROI Alis 2021 ( 16 ), Jang 2020( 14 ), Jang 2021( 15 ), Raisi-Estrabragh 2020 ( 17 ) Sum Entropy Describes the sum of differences of neighboring intensities Alis 2021( 16 ), Baessler 2018 ( 25 ), Jang 2020 ( 14 ), Jang 2021( 15 ), Raisi-Estrabragh 2020 ( 17 ) Texture (GLDM) Dependence Non-Uniformity Similarity of dependence within a ROI. Alis 2021( 16 ), Jang 2020 ( 14 ), Raisi-Estrabragh 2020 ( 17 ) Dependence Entropy Measures the degree of dependence inhomogeneity Alis 2021( 16 ), Jang 2020( 14 ), Mancio 2021( 3 ) Gray Level Non-Uniformity Measures the similarity of intensity values in the image Alis 2021( 16 ), Jang 2020 ( 14 ), Raisi-Estrabragh 2020 ( 17 ) Texture (GLRLM) Short Run Emphasis Measures the distribution of short run length. High value indicates more fine textures Alis 2021( 16 ), Baessler 2018 ( 25 ), Raisi-Estrabragh 2020 ( 17 ), Mancio 2021 ( 3 ) Gray Level Non-Uniformity Measures the similarity of intensity values in the image Alis 2021( 16 ), Jang 2020 ( 14 ), Raisi-Estrabragh 2020 ( 17 ) Run Length Non-Uniformity Measures the similarity of run lengths in the image. Lower value indicates more homogeneity in run lengths Alis 2021( 16 ), Jang 2020 ( 14 ), Raisi-Estrabragh 2020 ( 17 ) Texture (GLSZM) Gray Level Non-Uniformity Measures the similarity of intensity values in the image Alis 2021( 16 ), Jang 2020 ( 14 ), Raisi-Estrabragh 2020 ( 17 ) Zone Entropy Measures randomness in distribution of zone sizes and gray levels. High value indicates more heterogeneity in texture patterns Alis 2021( 16 ), Jang 2020 ( 14 ), Raisi-Estrabragh 2020 ( 17 ) Shape (2D) Area Basis for diagnostic parameters (ESV, EDV, EF, Myocardial Mass) Sphericity Degree to which a segmented region approximates a circle Elongation Relation between lengths of smallest and longest main axes of a segmented region Cardio-specific Relative Septum Thickness Ratio of septum thickness to heart diameter, calculated on spline contours Tautz 2020 ( 27 ) Area Ratio LV/RV Ratio of LV area to RV area, calculated on spline contours Relative RV Diameter Ratio of RV diameter to heart diameter, calculated on spline contours Radial Strain (min, max, mean) Shape deformation due to myocardial dynamics, calculated on voxelized LVM Circumferential Strain (min, max, mean) Left Blood Pool Fractal Dimension Complexity of boundary (e.g. due to trabeculation), calculated on voxelized LVBP Captur 2013 ( 21 ) Feature Calculation All features were calculated by a prototypical software implemented in MeVisLab (version 3.5a, MeVis Medical Solutions AG, Bremen, Germany). To evaluate the myocardium independently of patient size, LVM Area Ratio is calculated as LVM Area / (LVM Area + LVBP Area). Relative septum thickness represents the distance between the intersections of the LVM contours with the line connecting the LVBP and RVBP centers. The relative RV diameter is calculated analogously. Circumferential and radial strain statistics are based on analyzing voxel-wise deformation vectors relative to end-systole (first frame) with strain tensors, representing Lagrangianstrain ( 20 ). The fractal Minkowski-Bouligand (box counting) dimension is derived from the voxelized LVBP foreground segmentation and quantifies the complexity of trabecularization and blood pool boundaries ( 21 ). (Further details in the online supplement). Feature Aggregation and Time Series Features were calculated for each slice and time point, then aggregated (sum, mean, median, min, max) to create feature time curves. To assess reproducibility with respect to temporal resolution, we linearly interpolated all-time series to 50 phases. Statistical Analysis of Feature Robustness Assessment of Segmentation Results The inter- and intra-observer variation of the segmentation was assessed with the Dice similarity coefficient (DSC), the Hausdorff distance (HD), and the average contour distance (ACD) in MeVisLab. DSC quantifies the voxel overlap of the rasterized segmentation; HD is defined as the maximum and ACD as the mean of the minimal distances between points of the compared contours. Assessment of Radiomics Feature Reproducibility To assess the reproducibility of radiomics features, we performed an inter- and intra-observer variability analysis. We analyzed the difference of the feature maxima between the high and low temporal resolution to quantify the potential information loss due to lower sampling rate. For reproducibility analysis we utilized the intra-class correlation coefficient (ICC) calculated with Pingouin 0.4.0 ( 22 ) as a two-way random model, with single measures and absolute agreement (equivalent to ICC ( 2 , 1 ) in the R psych package). We adopted the levels of reliability from Koo and Li ( 23 ), with threshold values of 0.5 (poor reliability), 0.75 (moderate), 0.9 (good) and 1.0 (excellent). 3. Results Observer Agreement at Different Spatio-Temporal Resolutions Higher agreement was generally found at the higher spatio-temporal resolution, except for O2’s segmentations of LVM and LVBP (see Table 3 and Fig. 2). O1 had an equal or better agreement than O2 in most cases, except for a higher HD for RVBP at the high resolution. Inter-observer agreement was also higher for high-resolution segmentations (see Table 3 and Fig. 2). Table 3 Intra- and inter-observer agreement of the segmentations measured by Dice similarity coefficient (DSC), Hausdorff distance (HD) and average contour distance (ACD). * indicates the highest agreement. Structure Spatio-temporal resolution Mean DSC Mean HD [mm] Mean ACD [mm] Intra Inter Intra Inter Intra Inter O1 O2 O1 O2 O1 O2 Left Ventricular Myocardium (LVM) High 0.88* 0.84 0.88 1.98* 2.32 1.96 0.49* 0.65 0.28 Low 0.86* 0.83 0.83 2.16* 2.29 2.34 0.52* 0.65 0.40 Left Ventricular Blood Pool (LVBP) High 0.94* 0.89 0.92 1.84* 2.31 1.95 0.78* 1.13 0.62 Low 0.92* 0.89 0.89 2.09* 2.26 2.34 0.95* 1.05 0.90 Right Ventricular Blood Pool (RVBP) High 0.98* 0.97 0.99 1.94 1.41* 1.36 0.18* 0.25 0.08 Low 0.91* 0.91* 0.91 3.72* 4.93 4.51 0.53* 0.82 0.82 Influence of Segmentation on Feature Reproducibility Intra-Observer Reproducibility The ICC values for intra-observer reproducibility of high-resolution features are shown in Fig. 3 and Table 4 . O1 demonstrates a higher agreement than O2. LVM features show the most observer dependence, with O1 achieving good/excellent reproducibility for all except Elongation (moderate). O2’s LVM reproducibility ranges from moderate to excellent. LVBP features are mostly good/excellent, except for Sphericity, which has poor agreement. RVBP features show excellent reproducibility, though RVBP Sphericity has a lower ICC. Cardio-specific features are mostly good/excellent, with O2 outperforming O1 in six out of ten features. Of the 64 features we found (57, 4, 2, 1) O1’s features and (56, 7, 0, 1) O2’s features to be of (excellent, good, moderate, poor) intra-observer reproducibility, respectively. Inter-Observer Reproducibility The ICC values for the inter-observer agreement of features are displayed in Fig. 3. and Table 4 . We observe a higher ICC for the high-resolution features with excellent reproducibility for 54 of 64 features and 39 in low-resolution scans. Of the three anatomical structures, the LVM’s features show the lowest ICC values for the two resolutions, as well as the largest feature-wise variances. All features extracted from high-resolution images exhibit good or excellent reproducibility. Of the features originating from low-resolution sequences, only Elongation, GLDM Dependence Non-Uniformity, GLDM Dependence Entropy, and GLSZM Zone Entropy exhibit moderate inter-observer reproducibility, all other features’ ICC scores are good or excellent. As for the intra-observer comparison, LVBP Sphericity exhibits poor agreement at both resolutions in the inter-observer comparison. All other LVBP features are of good or excellent reproducibility. Table 4 Number of features with excellent, good, moderate, and poor inter- and intra-observer, and inter-resolution Intra-Class Correlation (ICC) values. Inter-resolution ICC are given separately by slice. Table 4. Number of features with excellent, good, moderate, and poor inter- and intra-observer, and inter-resolution Intra-Class Correlation (ICC) values. Inter-resolution ICC are given separately by slice. ICC Class Low Resolution High Resolution Inter-Resolution Inter-observer Inter-observer Intra-observer Apical Mid Ventricle Basal Observer 1 Observer 2 Excellent 39 54 57 56 4 3 1 Good 19 9 4 7 9 5 12 Moderate 5 0 2 0 23 23 20 Poor 1 1 1 1 28 33 31 All features of the RVBP are found to have good or excellent ICC scores, with the features from high-resolution images scoring consistently in the excellent range, except for Sphericity. The cardio-specific features show good or excellent agreement at both resolutions, apart from the Area Ratio LV/RV, which has only moderate reproducibility at low resolution. Influence of Spatio-Temporal Resolution on Feature Reproducibility The ICC values for the different spatio-temporal resolutions for each of the three slices are displayed in Fig. 2. Most 2D shape features show good or excellent reproducibility between low and high spatiotemporal resolution and do not vary significantly regarding the slice position. The features showing greater variability between low and high resolution are also influenced by the slice position; features of LVBP vary strongest in mid ventricular slices. Influence of Spatio-Temporal Resolution on Maximum Feature Values Figure 4 shows how maximum feature values vary with resolution. A zero deviation means no change, negative values indicate lower maxima at low resolution, and positive values indicate higher maxima at low resolution. Comparison of Robust Feature Curves in Controls and HF Patients Figure 5 shows the temporal features curves for aggregated high-resolution case features of EMPATHY volunteer datasets compared with features derived from the patients with the three HF subtypes. The average values of robust radiomics features showed a greater difference between healthy controls, HFpEF, HFmrEF, and HFrEF patients for LVM and LVBP than for RVBP. The LVBP area and run-length-non-uniformity curves have non-overlapping value ranges between the groups with normal and reduced EF. The LVM area ratio and LVM sphericity features curves showed non-overlapping value ranges for the age-matched control group and HFpEF patients in the diastolic phase (p-value < 0.05). The relative septum thickness and the circumferential strain curves of the four groups differ in curve shape but have overlapping value ranges. Robust 2D LVM shape features, like Area Ratio, Sphericity, and Strain, showed the largest differences across healthy, HFpEF, HFmrEF, and HFrEF groups. Shape features and septal thickness also differed significantly between healthy and HFpEF (p < 0.05; see Supplementary). Dependance and Run Length Uniformity of LVBP differed significantly (p < 0.05; see Supplementary) between individuals with preserved and non-preserved EF, with no overlap in values. 4. Discussion Segmentation Reproducibility A higher spatio-temporal resolution was correlated with a better intra- and inter-observer segmentation reproducibility for LVM, LVBP and RVBP (Table 3 ). We could also observe that the experience of the radiologist was correlated with a better intra-observer reproducibility for LVM and LVBP. In contrast to previous studies of LVM und LVBP ( 24 ) the intra- and inter-observer reproducibility were comparable. The differences in previous studies could be explained through the observers' strong experience (high intra-observer reproducibility) but different training (higher inter-observer variability than in our study). In line with our observation on segmentation agreement we found inter-observer feature reproducibility to be as good as intra-observer reproducibility (Fig. 3). These results differ from previous studies ( 25 ) ( 14 ). Intra-Observer and Inter-Observer Reproducibility Most Radiomics features showed good to excellent intra-observer reproducibility, with LVM features of experienced O1 being more reproducible than those of O2. The features that were found reproducible in the study of Jang 2020 showed similar results in our study. O2 achieved higher agreement than O1 for cardio-specific features. The strain features showed a good to excellent reproducibility. In the study from Schmidt et al ( 26 ) the strain features showed an excellent intra-observer reproducibility except for radial SR. Segmentation agreement is not explicitly reported but is assumingly causing the inter-observer differences. We observe a better inter-observer agreement on high spatiotemporal resolution images, with LVM’s features showing the largest differences in ICC for the two resolutions (Fig. 3). 62 of 64 features for high resolution (96%) and 51 of 64 features for low resolution (79%) showed an ICC ≥ 0.8. Jang et al. 2020 ( 14 ) found only 32.1% features showing an ICC in this range. The difference is probably caused by differences in segmentation agreement or in spatial LV coverage between the two studies. Figures 3 and 4 illustrate how the ICC and feature variation depend on slice location. Influence of Slice Position and Spatio-Temporal Sampling Our results are consistent with the previous findings in that spatial resolution is a major cause of variability of feature values ( 15 , 25 ). The ICC score for LVBP and RVBP texture features was worse in the midventricular slice which may be explained by the greater mobility of the heart at this level. This agrees with the study of Alis et al ( 16 ) where 55% of the features showed great variability through the cardiac cycle. For LVM, the agreement was lower at basal level where the segmentation agreement was worse. Few studies analyze cardiac phases other than end-diastolic and end-systolic time points, but there is limited evidence that features incorporating temporal information are more reproducible ( 14 , 16 , 27 , 28 ). Previous studies found associations between radiomics feature curve max/min values and traditional cardiac function and motions indices ( 16 , 29 ). Feature variability was higher in diastole than systole (Fig. 5), with less pronounced extremes at low resolution. Similar findings were reported by Backhaus et al. ( 11 ) for temporal resolution effects on strain values. However, we observed a higher dependence of feature reproducibility on spatial than temporal resolution (see. Supplements). Comparison between Feature Curves of the Controls and HF patients The analysis of robust radiomics feature curves between controls and patients with different HF-subtypes showed expected results for LVM 2D Shape features and cardiospecific features (Fig. 5). In line with previous publications, the circumferential strain curves of the four subgroups show different courses. However, the value ranges of the different groups overlap. In contrast to the findings in related work ( 30 ), the difference in the strain values of healthy individuals and HFpEF patients was not significant in our analysis (p = 0.24). Robust 2D Shape LVM features show non-overlapping value ranges for the controls and HFpEF subgroups with corresponding p-values < 0.05 which might enable the classification of the four subgroups based on a combination of LVBP and LVM features. Limitations The number of participants included limited our analysis. In the age-matched controls the proportion of women was higher than among the HF-patients. A multicentric study using different scans would provide a better accuracy of evaluation of the applicability of those technic in clinical settings, this could be the subject of a future study. 5. Conclusions The study demonstrates the need to align the selection of radiomics parameters and acquisition protocols. Our results imply the necessity to select high-resolution standardized protocols for the clinical use of some specific radiomics texture parameters such as Joint Entropy. Standarization is especially relevant in cardiac MRI, where long scan times challenge time management and patient comfort. Using reliable radiomics features characterizing the LVM such as area ratio, sphericity and GLRLM non-uniformity might enable the characterization of different types of heart failure. Abbreviations CMR: cardiac MRI LVBP: Left ventricle blood pool LVM: left ventricle myocardium RVBP: right ventricle blood pool O1: observer 1 O2: observer 2 HF: Heart Failure HFrEF: Heart Failure with reduced ejection fraction HFpEF: Heart Failure with preserved ejection fraction HFmrEF: Heart Failure with mild reduced ejection fraction DSC:Dice similarity coefficient (DSC) HD: Hausdorff distance ACD: average contour distance (ACD) ICC: Intraclass correlation coefficient Declarations Conflicts of Interests The authors declare no conflict of interest. Acknowledgments / Funding This work has been funded by the German Ministry for Education and Research (BIFOLD, 01IS18037E), the German Research Foundation (SFB 1470, INST 335/815-1, 515294457, HE 7312/7 − 1), and the DZHK (German Centre for Cardiovascular Research). Author Contribution Anja Hennemuth, Sebastian Kelle, Djawid Hashemi, Frank Edelmann and Jeanette Schulz-Menger conceived the presented idea. Lea Ter-Minassian contributed to the data management and carried out the experiments with the help of Hannu Zhang. Markus Hüllebrand developed the software program. Ann Laube,Lennart Tautz and Matthias Ivantsits performed the analytic calculations. Jennifer Erley verified the analytical method. Lea Ter-Minassian wrote the manuscript. Anja Hennemuth supervised the work with the help of Markus Hüllebrand. All authors discussed the results and contributed to the final manuscript. All authors reviewed the manuscript. References Kolossváry M, Karády J, Szilveszter B, Kitslaar P, Hoffmann U, Merkely B, et al. Radiomic Features Are Superior to Conventional Quantitative Computed Tomographic Metrics to Identify Coronary Plaques with Napkin-Ring Sign. Circ Cardiovasc Imaging. 2017 Dec 1;10(12). Li XN, Yin WH, Sun Y, Kang H, Luo J, Chen K, et al. Identification of pathology-confirmed vulnerable atherosclerotic lesions by coronary computed tomography angiography using radiomics analysis. Eur Radiol. 2022 Jun;32(6):4003–13. Mancio J, Pashakhanloo F, El-Rewaidy H, Jang J, Joshi G, Csecs I, et al. Machine learning phenotyping of scarred myocardium from cine in hypertrophic cardiomyopathy. Available from: https://academic.oup.com/ehjcimaging/article/23/4/532/6199608 Fahmy AS, Rowin EJ, Arafati A, Al-Otaibi T, Maron MS, Nezafat R. Radiomics and deep learning for myocardial scar screening in hypertrophic cardiomyopathy. Journal of Cardiovascular Magnetic Resonance. 2022 Dec 1;24(1). Zwanenburg A, Vallières M, Abdalah MA, Aerts HJWL, Andrearczyk V, Apte A, et al. The Image Biomarker Standardization Initiative: Standardized Quantitative Radiomics for High-Throughput Image-based Phenotyping. Radiology. 2020 May;295(2):328–38. Bernatz S, Zhdanovich Y, Ackermann J, Koch I, Wild PJ, Pinto dos Santos D, et al. Impact of rescanning and repositioning on radiomic features employing a multi-object phantom in magnetic resonance imaging. Scientific Reports |. 123AD;11:14248. Parekh V, Jacobs MA. Radiomics: a new application from established techniques. Expert Rev Precis Med Drug Dev. 2016 Mar 3;1(2):207–26. Chang S, Han K, Suh YJ, Choi BW. Quality of science and reporting for radiomics in cardiac magnetic resonance imaging studies: a systematic review. Eur Radiol. 2022 Jul 1;32(7):4361–73. Hassani C, Saremi F, Varghese BA, Duddalwar V. Myocardial Radiomics in Cardiac MRI. American Journal of Roentgenology. 2020 Mar;214(3):536–45. Inoue Y, Nomura Y, Nakaoka T, Watanabe M, Kiryu S, Okubo T, et al. Effect of temporal resolution on the estimation of left ventricular function by cardiac MR imaging. Magn Reson Imaging. 2005 Jun;23(5):641–5. Backhaus SJ, Metschies G, Billing M, Schmidt-Rimpler J, Kowallick JT, Gertz RJ, et al. Defining the optimal temporal and spatial resolution for cardiovascular magnetic resonance imaging feature tracking. Journal of Cardiovascular Magnetic Resonance. 2021 Dec 1;23(1). Yang W, Li H, He J, Yin G, An J, Forman C, et al. Left Ventricular Strain Measurements Derived from MR Feature Tracking: A Head‐to‐Head Comparison of a Higher Temporal Resolution Method With a Conventional Method. Journal of Magnetic Resonance Imaging. 2022 Sep 10;56(3):801–11. Schofield R, Ganeshan B, Fontana M, Nasis A, Castelletti S, Rosmini S, et al. Texture analysis of cardiovascular magnetic resonance cine images differentiates aetiologies of left ventricular hypertrophy. Clin Radiol. 2019 Feb;74(2):140–9. Jang J, Ngo LH, Mancio J, Kucukseymen S, Rodriguez J, Pierce P, et al. Reproducibility of Segmentation-based Myocardial Radiomic Features with Cardiac MRI. Radiol Cardiothorac Imaging. 2020 Jun 25;2(3):e190216. Jang J, El-Rewaidy H, Ngo LH, Mancio J, Csecs I, Rodriguez J, et al. Sensitivity of Myocardial Radiomic Features to Imaging Parameters in Cardiac MR Imaging. J Magn Reson Imaging. 2021 Sep;54(3):787–94. Alis D, Yergin M, Asmakutlu O, Topel C, Karaarslan E. The influence of cardiac motion on radiomics features: radiomics features of non-enhanced CMR cine images greatly vary through the cardiac cycle. Eur Radiol. 2021 May 13;31(5):2706–15. Raisi-Estabragh Z, Gkontra P, Jaggi A, Cooper J, Augusto J, Bhuva AN, et al. Repeatability of Cardiac Magnetic Resonance Radiomics: A Multi-Centre Multi-Vendor Test-Retest Study. Front Cardiovasc Med. 2020 Dec 2;7. Huellebrand M, Ivantsits M, Tautz L, Kelle S, Hennemuth A. A Collaborative Approach for the Development and Application of Machine Learning Solutions for CMR-Based Cardiac Disease Classification. Front Cardiovasc Med. 2022 Mar 10;9. Salgo IS, Tsang W, Ackerman W, Ahmad H, Chandra S, Cardinale M, et al. Geometric Assessment of Regional Left Ventricular Remodeling by Three-Dimensional Echocardiographic Shape Analysis Correlates with Left Ventricular Function. Journal of the American Society of Echocardiography. 2012 Jan;25(1):80–8. Tobon-Gomez C, De Craene M, McLeod K, Tautz L, Shi W, Hennemuth A, et al. Benchmarking framework for myocardial tracking and deformation algorithms: An open access database. Med Image Anal. 2013 Aug;17(6):632–48. Captur G, Muthurangu V, Cook C, Flett AS, Wilson R, Barison A, et al. Quantification of left ventricular trabeculae using fractal analysis. Journal of Cardiovascular Magnetic Resonance. 2013 Jan;15(1):36. Vallat R. Pingouin: statistics in Python. J Open Source Softw. 2018 Nov 19;3(31):1026. Koo TK, Li MY. A Guideline of Selecting and Reporting Intraclass Correlation Coefficients for Reliability Research. J Chiropr Med. 2016 Jun;15(2):155–63. Suinesiaputra A, Bluemke DA, Cowan BR, Friedrich MG, Kramer CM, Kwong R, et al. Quantification of LV function and mass by cardiovascular magnetic resonance: Multi-center variability and consensus contours. Journal of Cardiovascular Magnetic Resonance. 2015 Jul 28;17(1). Baeßler B, Weiss K, Pinto dos Santos D. Robustness and Reproducibility of Radiomics in Magnetic Resonance Imaging. Invest Radiol. 2019 Apr;54(4):221–8. Schmidt B, Dick A, Treutlein M, Schiller P, Bunck AC, Maintz D, et al. Intra- and inter-observer reproducibility of global and regional magnetic resonance feature tracking derived strain parameters of the left and right ventricle. Eur J Radiol. 2017 Apr;89:97–105. Tautz L, Zhang H, Hüllebrand M, Ivantsits M, Kelle S, Kuehne T, et al. Cardiac radiomics: an interactive approach for 4D data exploration. Current Directions in Biomedical Engineering. 2020 Sep 17;6(1). Larroza A, López‐Lereu MP, Monmeneu J V., Gavara J, Chorro FJ, Bodí V, et al. Texture analysis of cardiac cine magnetic resonance imaging to detect nonviable segments in patients with chronic myocardial infarction. Med Phys. 2018 Apr 22;45(4):1471–80. Lin K, Sarnari R, Carr JC, Markl M. Cine MRI ‐Derived Radiomics Features of the Cardiac Blood Pool: Periodicity, Specificity, and Reproducibility. Journal of Magnetic Resonance Imaging. 2023 Sep 19;58(3):807–14. Witt UE, Müller ML, Beyer RE, Wieditz J, Salem S, Hashemi D, et al. A simplified approach to discriminate between healthy subjects and patients with heart failure using cardiac magnetic resonance myocardial deformation imaging Keywords global longitudinal strain • cut-off • early identification of heart failure • cardiac magnetic resonance imaging • deformation imaging. European Heart Journal - Imaging Methods and Practice [Internet]. 2024;2:93. Available from: https://doi.org/10.1093/ehjimp/qyae093 Table 1 - embedded object The embedded object within Table 1 in the last column for the row labeled "Slice spacing [mm]" is not available with this version. Additional Declarations No competing interests reported. Supplementary Files SupplementaryMaterial2.docx GraphicalAbstract1.pptx Cite Share Download PDF Status: Posted Version 1 posted You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-6912341","acceptedTermsAndConditions":true,"allowDirectSubmit":true,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":476890855,"identity":"dbaaa971-3788-43d5-b4d2-8db2faf9caa4","order_by":0,"name":"Lea Ter-Minassian","email":"","orcid":"","institution":"Deutsches Herzzentrum der Charité","correspondingAuthor":false,"prefix":"","firstName":"Lea","middleName":"","lastName":"Ter-Minassian","suffix":""},{"id":476890856,"identity":"a2f2e097-6e8c-4837-9baf-b74898b1cf7f","order_by":1,"name":"Matthias ivantsis","email":"","orcid":"","institution":"Deutsches Herzzentrum der Charité","correspondingAuthor":false,"prefix":"","firstName":"Matthias","middleName":"","lastName":"ivantsis","suffix":""},{"id":476890857,"identity":"ffa29b26-d17b-4374-a372-56970d6a31b5","order_by":2,"name":"Markus Hüllebrand","email":"","orcid":"","institution":"Deutsches Herzzentrum der Charité","correspondingAuthor":false,"prefix":"","firstName":"Markus","middleName":"","lastName":"Hüllebrand","suffix":""},{"id":476890858,"identity":"fc7406f0-e8a0-4f09-8f49-e7e0741749c7","order_by":3,"name":"Hannu Zhang","email":"","orcid":"","institution":"Charité – Universitätsmedizin Berlin, corporate member of Freie Universität Berlin, Humboldt Universität zu Berlin","correspondingAuthor":false,"prefix":"","firstName":"Hannu","middleName":"","lastName":"Zhang","suffix":""},{"id":476890859,"identity":"d09a5c5f-0dbb-4d9b-aa45-009a2a589d81","order_by":4,"name":"Ann Laube","email":"","orcid":"","institution":"Deutsches Herzzentrum der Charité","correspondingAuthor":false,"prefix":"","firstName":"Ann","middleName":"","lastName":"Laube","suffix":""},{"id":476890860,"identity":"22f2d2f3-4873-4d58-abce-18955353f4c0","order_by":5,"name":"Lennart Tautz","email":"","orcid":"","institution":"Fraunhofer Institute for Digital Medicine MEVIS","correspondingAuthor":false,"prefix":"","firstName":"Lennart","middleName":"","lastName":"Tautz","suffix":""},{"id":476890861,"identity":"c9e377c4-569d-4a8f-ac06-d393cf7a5c8b","order_by":6,"name":"Jeanette Schulz-Menger","email":"","orcid":"","institution":"Charité – Universitätsmedizin Berlin, corporate member of Freie Universität Berlin, Humboldt Universität zu Berlin","correspondingAuthor":false,"prefix":"","firstName":"Jeanette","middleName":"","lastName":"Schulz-Menger","suffix":""},{"id":476890862,"identity":"5c1124ad-fef4-44e2-b4a2-a167f534a76c","order_by":7,"name":"Djawid Hashemi","email":"","orcid":"","institution":"Deutsches Herzzentrum der Charité","correspondingAuthor":false,"prefix":"","firstName":"Djawid","middleName":"","lastName":"Hashemi","suffix":""},{"id":476890863,"identity":"3ae34938-2cb3-44f6-a63d-294317fcd12f","order_by":8,"name":"Jennifer Erley","email":"","orcid":"","institution":"University Medical Center Hamburg-Eppendorf","correspondingAuthor":false,"prefix":"","firstName":"Jennifer","middleName":"","lastName":"Erley","suffix":""},{"id":476890864,"identity":"c061cb31-8144-475c-b421-4338b2a1a051","order_by":9,"name":"Sebastian Kelle","email":"","orcid":"","institution":"Deutsches Herzzentrum der Charité","correspondingAuthor":false,"prefix":"","firstName":"Sebastian","middleName":"","lastName":"Kelle","suffix":""},{"id":476890865,"identity":"f04296ff-e349-4324-be58-03e028a9ec87","order_by":10,"name":"Frank Edelmann","email":"","orcid":"","institution":"Deutsches Herzzentrum der Charité","correspondingAuthor":false,"prefix":"","firstName":"Frank","middleName":"","lastName":"Edelmann","suffix":""},{"id":476890866,"identity":"4b6d1805-785e-49c5-9dbf-fb6e9fca9895","order_by":11,"name":"Anja Hennemuth","email":"data:image/png;base64,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","orcid":"","institution":"Deutsches Herzzentrum der Charité","correspondingAuthor":true,"prefix":"","firstName":"Anja","middleName":"","lastName":"Hennemuth","suffix":""}],"badges":[],"createdAt":"2025-06-17 08:53:28","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-6912341/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-6912341/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":85846603,"identity":"e7612d8d-fbb8-4595-ab44-4a5a950dba8c","added_by":"auto","created_at":"2025-07-02 09:43:15","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":276261,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cem\u003eReproducibility analysis pipeline. Influence of the segmentation (intra- and interobserver variability) and resolution is based on the scans of healthy volunteers.\u003c/em\u003e\u003c/p\u003e\n\u003cp\u003e\u003cem\u003eSegmentation contours are converted to segmentation masks in the resampled images. \u0026nbsp;Feature curves of age-matched controls are compared with those of heart failure patients to analyse the influence of pathological alterations vs. the sampling methods and observers.\u003c/em\u003e\u003c/p\u003e","description":"","filename":"1.png","url":"https://assets-eu.researchsquare.com/files/rs-6912341/v1/ac8f57c14aaebab8526db843.png"},{"id":85846605,"identity":"6d736326-345c-48a7-9f26-d681f5a323cd","added_by":"auto","created_at":"2025-07-02 09:43:15","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":79471,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cem\u003eAgreement of the expert-curated segmentations assessed by Dice Score (top row), Hausdorff Distance (middle row) and Average Contour Distance (bottom row).\u003c/em\u003e\u003c/p\u003e\n\u003cp\u003e\u003cem\u003eIntra-observer agreement values have been pooled for both observers (left column). Note the agreement in LVM and LVBP segmentations, and the variation in RVBP segmentations, and the differences in agreement between high and low spatio-temporal resolutions.\u003c/em\u003e\u003c/p\u003e","description":"","filename":"2.png","url":"https://assets-eu.researchsquare.com/files/rs-6912341/v1/4da5f7efd2f73528a54b33c8.png"},{"id":85847118,"identity":"c9196ee7-8141-4145-ba66-fdee7f0c3e1e","added_by":"auto","created_at":"2025-07-02 09:51:15","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":368420,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cem\u003eAgreement of radiomic and cardiospecific features depending on observers, resolutions and slice positions for LVM, LVBP and RBP.\u003c/em\u003e\u003c/p\u003e\n\u003cp\u003e\u003cem\u003eColumns 1 and 2: intra- and inter-observer agreement grouped by anatomical structure and observer for the high-resolution image sequences. Color-coding indicates repeated assessments for observer 1 (yellow), observer 2 (green), and the inter-observer agreement (blue). In column 2 low is represented by circle glyphs and high resolution by cross glyphs. Column 3: inter-resolution agreement for both observers for the apical (purple), midventricular (pink), and basal (orange) slice positions. Differences in background shading indicate the ICC reproducibility thresholds from Koo and Li (23)\u003c/em\u003e\u003c/p\u003e","description":"","filename":"3.png","url":"https://assets-eu.researchsquare.com/files/rs-6912341/v1/fc10688a65111be571a879fe.png"},{"id":85846608,"identity":"9f857b13-ce52-48d2-8f23-ddc8746fbd2f","added_by":"auto","created_at":"2025-07-02 09:43:15","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":200306,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cem\u003eInter-resolution maximum feature value differences normalized by the largest absolute maximum value of the corresponding feature value pair for LVM (A) LVBP (B) RVBP (C) and Cardiospecific features (D)\u003c/em\u003e\u003c/p\u003e","description":"","filename":"4.png","url":"https://assets-eu.researchsquare.com/files/rs-6912341/v1/3d4ee52c75423476168ceb22.png"},{"id":85847120,"identity":"38aae98b-98fa-43fc-baad-769c6aaee1a6","added_by":"auto","created_at":"2025-07-02 09:51:15","extension":"png","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":1255463,"visible":true,"origin":"","legend":"\u003cp\u003eComparison of high-resolution feature ranges between age-matched controls and HFpEF, HFmrEF, HFrEF patients for LVM (A), LVBP (B), RVBP (C), and Cardiospecific features (D).\u003c/p\u003e\n\u003cp\u003e\u003cem\u003eFor each group the area shows the value range with the corresponding curve showing the average values.\u003c/em\u003e\u003c/p\u003e","description":"","filename":"5.png","url":"https://assets-eu.researchsquare.com/files/rs-6912341/v1/22ce22f7796e4ed41d885ec0.png"},{"id":90689198,"identity":"76961bc1-4242-4363-b4ac-6b4f807a7dd9","added_by":"auto","created_at":"2025-09-05 17:46:42","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":3183854,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-6912341/v1/4f196496-dd8d-43a7-bcca-0a389d968387.pdf"},{"id":85847117,"identity":"23107321-1374-4a06-95e7-4456e502365c","added_by":"auto","created_at":"2025-07-02 09:51:15","extension":"docx","order_by":0,"title":"","display":"","copyAsset":false,"role":"supplement","size":274441,"visible":true,"origin":"","legend":"","description":"","filename":"SupplementaryMaterial2.docx","url":"https://assets-eu.researchsquare.com/files/rs-6912341/v1/e976db77fdbc91cfa877a52d.docx"},{"id":85846626,"identity":"7b7ec862-b9f6-47c5-8ebb-f805d71db974","added_by":"auto","created_at":"2025-07-02 09:43:15","extension":"pptx","order_by":1,"title":"","display":"","copyAsset":false,"role":"supplement","size":2725163,"visible":true,"origin":"","legend":"","description":"","filename":"GraphicalAbstract1.pptx","url":"https://assets-eu.researchsquare.com/files/rs-6912341/v1/78ae194660c6aee767e69d83.pptx"}],"financialInterests":"No competing interests reported.","formattedTitle":"Cardiac Cine MRI Radiomics: Analysis of Feature Reproducibility and Evaluation of Variations in the Context of Heart Failure","fulltext":[{"header":"1. Introduction","content":"\u003cp\u003eRadiomics can help image-based assessment of atherosclerosis and infarcts (\u003cspan additionalcitationids=\"CR2 CR3\" citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e), by quantifying shape and intensity patterns of segmented structures.To ensure reproducibility, the IBSI promotes standardization across tools like PyRadiomics (\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e), butfeature values also depend on voxel size, object positioning (\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e), intensity sampling (\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e), and preprocessing\u0026mdash;areas often lacking in cardiac MRI studies (\u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e). MRI intensities also vary with imaging settings and segmentation accuracy (\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eTemporal sampling is crucial for the reproducibility of conventional diagnostic parameters, such as ventricular volumes (\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e) and strain (\u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e, \u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e). A few studies examined the effects of imaging procedure variations on radiomics feature reproducibility in cardiac cine MRI. Schofield et al. analyzed parameter variations as a proxy for multi-site or multi-vendor imaging (\u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e). Their study found good reproducibility of features like intensity mean, standard deviation, skewness, and kurtosis in the myocardium. A test-retest study by Jang et al. identified 102 reproducible radiomics texture features, (\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e). They investigated the robustness of texture features under different acquisition parameters, including spatial resolution, and parallel imaging technique (\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e). For cine imaging, spatial resolution had the most significant effect on feature reproducibility. Most studies (\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e, \u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e) assess reproducibility using intra-class correlation coefficients (ICCs) with varying thresholds (see Supplement Criteria for reproducibility).\u003c/p\u003e \u003cp\u003eThis paper investigates four primary aspects of radiomics feature reproducibility:\u003c/p\u003e \u003cp\u003e \u003col\u003e \u003cspan\u003e \u003cli\u003e \u003cp\u003eThe influence of spatiotemporal sampling on segmentation.\u003c/p\u003e \u003c/li\u003e \u003c/span\u003e \u003cspan\u003e \u003cli\u003e \u003cp\u003eThe reproducibility of features with respect to segmentation.\u003c/p\u003e \u003c/li\u003e \u003c/span\u003e \u003cspan\u003e \u003cli\u003e \u003cp\u003eThe reproducibility of features concerning spatial and temporal sampling.\u003c/p\u003e \u003c/li\u003e \u003c/span\u003e \u003cspan\u003e \u003cli\u003e \u003cp\u003eValue ranges of reproducible features in different heart failure subtypes compared to age-matched controls.\u003c/p\u003e \u003c/li\u003e \u003c/span\u003e \u003c/ol\u003e \u003c/p\u003e \u003cp\u003eWe evaluate cardiac shape parameters, quadrature-filter-based motion vectors and texture features from previous studies to capture structural and functional information from 4D image sequences (\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e, \u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e, \u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e, \u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e).\u003c/p\u003e"},{"header":"2. Materials and Methods","content":"\u003cp\u003e \u003cb\u003eEthic statements\u003c/b\u003e \u003c/p\u003e \u003cp\u003e This retrospective study was conducted in accordance with the Helsinki declaration.\u003c/p\u003e \u003cp\u003eWe declare no conflict of interest.\u003c/p\u003e \u003cp\u003eThe heart failure patient data originates from the EMPATHY-HF registry (German Clinical Trials Register ID: DRKS00015615).\u003c/p\u003e \u003cp\u003eThe image analysis of the healthy volunteers had been approved by the ethics committee of Charit\u0026eacute; Universit\u0026auml;tsmedizin Berlin (application number EA2/073/16).\u003c/p\u003e \u003cp\u003e All participants provided written consent.\u003c/p\u003e \u003cp\u003e \u003cb\u003eDataset\u003c/b\u003e \u003c/p\u003e \u003cp\u003eAll imaging protocols included short-axis ECG-triggered cine CMR images. For the 12 healthy volunteers (6 females) we analyzed two parameterizations for spatiotemporal sampling to assess the influence of imaging parameters on the reproducibility of the proposed radiomics features (see Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e). These datasets have previously been analysed for the optimization of protocols for strain analysis (\u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e). They had a median age of 28 years (IQR 9) and normal left ventricular ejection fraction LVEF (median 60%; IQR 3 [59, 62]). The subset used for the inter-observer analysis consisted of eight volunteers (50% female), with median age 31 (IQR 7) and median LVEF of 62% (IQR 4 [59,63]).\u003c/p\u003e \u003cp\u003eFor the analysis of the radiomics feature suitability for the differentiation of HF subtypes, we used datasets from the EMPATHY-HF cohort (\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://drks.de/search/de/trial/DRKS00015615\u003c/span\u003e\u003cspan address=\"https://drks.de/search/de/trial/DRKS00015615\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e) [19]. The subgroups are:\u003c/p\u003e \u003cp\u003e \u003cul\u003e \u003cli\u003e \u003cp\u003e15 patients with heart failure with reduced ejection fraction (HFrEF) (3 female) with a median age of 63 (IQR 14 [55, 69]) and a median LVEF of 34% (IQR 7 [30, 37])\u003c/p\u003e \u003c/li\u003e \u003cli\u003e \u003cp\u003e17 patients with heart failure with mild reduced ejection fraction (HFmrEF) (4 female) with a median age of 73 (IQR 10 [64, 74]) and a median LVEF of 44% (IQR 3 [42, 45])\u003c/p\u003e \u003c/li\u003e \u003cli\u003e \u003cp\u003e22 patients with heart failure with preserved ejection fraction (HFpEF) (11 female) with median age of 78 (IQR 8 [73, 81]) and a median LVEF of 59% (IQR 12 [54, 66])\u003c/p\u003e \u003c/li\u003e \u003cli\u003e \u003cp\u003e19 controls (9 female) with a median age of 62 (IQR 11 [56, 67]) and a median LVEF of 64% (IQR 8 [62, 68])\u003c/p\u003e \u003c/li\u003e \u003c/ul\u003e \u003c/p\u003e \u003cp\u003eThe automatically segmented EMPATHY datasets were processed to evaluate if the features deemed robust regarding imaging and expert segmentation, are sensitive to pathological changes.\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab1\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 1\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eOverview of imaging parameters.\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"3\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eParameter\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eHealthy Volunteers\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eHF Patients and age-matched controls\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eScanner\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c3\" namest=\"c2\"\u003e \u003cp\u003ePhilips Achieva 1.5T\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eImaging sequence\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c3\" namest=\"c2\"\u003e \u003cp\u003eECG-gated bSSFP cine\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eRepetition time [ms]\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e4.13\u0026thinsp;\u0026plusmn;\u0026thinsp;0.02 **\u003c/p\u003e \u003cp\u003e2.57\u0026thinsp;\u0026plusmn;\u0026thinsp;0.06 *\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e3.42\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eEcho time [ms]\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e2.07\u0026thinsp;\u0026plusmn;\u0026thinsp;0.1 **\u003c/p\u003e \u003cp\u003e1.29\u0026thinsp;\u0026plusmn;\u0026thinsp;0.03 *\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e1.624\u0026ndash;1.761\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eFlip angle [\u0026deg;]\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c3\" namest=\"c2\"\u003e \u003cp\u003e60\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eIn-plane resolution (isotropic) [mm]\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e1.39\u0026thinsp;\u0026plusmn;\u0026thinsp;0.03 **\u003c/p\u003e \u003cp\u003e1.43\u0026thinsp;\u0026plusmn;\u0026thinsp;0.12 *\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e1.45\u0026ndash;1.63\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eSlice thickness [mm]\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e5** / 10*\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e8\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eSlice spacing [mm]\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e15.49\u0026thinsp;\u0026plusmn;\u0026thinsp;1.66 **\u003c/p\u003e \u003cp\u003e17.66\u0026thinsp;\u0026plusmn;\u0026thinsp;1.66 *\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e࿼\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eTemporal resolution [phases/cycle]\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e20* / 50**\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e50\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eK-space sampling\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c3\" namest=\"c2\"\u003e \u003cp\u003ecartesian\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eHeart rate [bpm]\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e67.0\u0026thinsp;\u0026plusmn;\u0026thinsp;8.07 **\u003c/p\u003e \u003cp\u003e64.5\u0026thinsp;\u0026plusmn;\u0026thinsp;8.43 *\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e47\u0026ndash;85\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003eTable\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e: \u003cspan type=\"ItalicSmallCaps\" class=\"ItalicSmallCaps\" name=\"Emphasis\"\u003eOverview of imaging parameters considered for feature robustness analysis.\u003c/span\u003e\u003c/p\u003e \u003cp\u003e \u003cspan type=\"SmallCaps\" class=\"SmallCaps\" name=\"Emphasis\"\u003e*: \u0026ldquo;low-resolution\u0026rdquo; parameters\u003c/span\u003e \u003c/p\u003e \u003cp\u003e \u003cspan type=\"SmallCaps\" class=\"SmallCaps\" name=\"Emphasis\"\u003e**: \u0026ldquo;high-resolution\u0026rdquo; parameters\u003c/span\u003e \u003c/p\u003e \u003cp\u003e \u003cb\u003eImage Preprocessing\u003c/b\u003e \u003c/p\u003e \u003cp\u003eWe rescaled the intensities to a range of [0, 4095] and resampled the images with a cubic B-spline to a voxel-size of 1.5x1.5 mm\u003csup\u003e2\u003c/sup\u003e (highest resolution in the dataset). For image intensity discretization, we used the PyRadiomics default value with a bin size of 25.\u003c/p\u003e \u003cp\u003e \u003cb\u003eImage Segmentation\u003c/b\u003e \u003c/p\u003e \u003cp\u003eTwo experts (experienced radiologist O1, and medical trainee O2) independently segmented the volunteer sequences using a semi-automatic tool (\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e). A neural network automatically labeled the left ventricular blood pool (LVBP), including trabeculation and papillary muscles, the left ventricular myocardium (LVM), and the right ventricular blood pool (RVBP). The experts inspected and refined the resulting contours. For the calculation of voxel-based radiomics features, contours were rasterized to image masks (ROIs).\u003c/p\u003e \u003cp\u003eO1 and O2 segmented the high- and low-resolution images of 8 volunteers twice to assess intra- and inter-observer variation of segmentations and derived features. For these and 4 supplementary healthy volunteers, O2 also segmented supplementary spatiotemporal resolution image sequences. HF patient and control datasets were not post-processed after automatic segmentation (see Fig.\u0026nbsp;1).\u003c/p\u003e \u003cp\u003e \u003cb\u003eFeature Extraction\u003c/b\u003e \u003c/p\u003e \u003cp\u003eThe first feature group contains radiomics first-order and texture features which were considered reproducible in at least three previous publications (see Table\u0026nbsp;\u003cspan refid=\"Tab2\" class=\"InternalRef\"\u003e2\u003c/span\u003e). Values were calculated using the PyRadiomics package for all rasterized ROIs and time points.\u003c/p\u003e \u003cp\u003eThe second group consists of the shape features, area, elongation, and sphericity, acting as surrogates for clinical parameters of the left ventricle (\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eThe third group contains ten cardio-specific features that describe anatomy, motion and the interaction between left and right ventricles.\u003c/p\u003e \u003cp\u003eStrain features calculated with quadrature filters for motion analysis characterize local deformation (\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e). The Minkowski-Bouligand Dimension describes the complexity of the endocardial border (\u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e), which can provide insights about trabeculation.\u003c/p\u003e \u003cp\u003e \u003cem\u003eFigure 1: Reproducibility analysis pipeline. Influence of the segmentation (intra- and interobserver variability) and resolution is based on the scans of healthy volunteers.\u003c/em\u003e \u003c/p\u003e \u003cp\u003e \u003cem\u003eSegmentation contours are converted to segmentation masks in the resampled images. Feature curves of age-matched controls are compared with those of heart failure patients to analyse the influence of pathological alterations vs. the sampling methods and observers.\u003c/em\u003e \u003c/p\u003e\u003cp\u003eTable 2 Overview of extracted features, including previous publication in which features have been found to be robust.\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"No\" id=\"Taba\" border=\"1\"\u003e \u003ccolgroup cols=\"4\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eType\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eFeature\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eDescription\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eFound robust in these studies\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\" morerows=\"3\" rowspan=\"4\"\u003e \u003cp\u003eFirst-order\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eEnergy\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eMeasures the degree of intensity inhomogeneity\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eAlis 2021(\u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e), Jang 2020 (\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e), Jang 2021(\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e), Raisi-Estrabragh 2020(\u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eMean\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eThe mean of the intensity histogram\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eAlis 2021(\u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e), Raisi-Estrabragh 2020 (\u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e), Schofield 2019 (\u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eTotal Energy\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eMeasures the degree of intensity inhomogeneity scaled by the voxel volume\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eAlis 2021(\u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e), Jang 2020 (\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e), Jang 2021(\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e), Raisi-Estrabragh 2020(\u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e90th Percentile\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eThe 90th percentile of the intensity histogram\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eAlis 2021(\u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e), Raisi-Estrabragh 2020 (\u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e), Mancio 2021 (\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\" morerows=\"2\" rowspan=\"3\"\u003e \u003cp\u003eTexture (GLCM)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eJoint Entropy\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eMeasures the randomness of neighboring intensities\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eAlis 2021 (\u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e), Jang 2020 (\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e), Jang 2021(\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e), Raisi-Estrabragh 2020 (\u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eJoint Energy\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eEnergy measures the degree of intensity homogeneity within a ROI\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eAlis 2021 (\u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e), Jang 2020(\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e), Jang 2021(\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e), Raisi-Estrabragh 2020 (\u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eSum Entropy\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eDescribes the sum of differences of neighboring intensities\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eAlis 2021(\u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e), Baessler 2018 (\u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e), Jang 2020 (\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e), Jang 2021(\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e), Raisi-Estrabragh 2020 (\u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\" morerows=\"2\" rowspan=\"3\"\u003e \u003cp\u003eTexture (GLDM)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eDependence Non-Uniformity\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eSimilarity of dependence within a ROI.\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eAlis 2021(\u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e), Jang 2020 (\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e), Raisi-Estrabragh 2020 (\u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eDependence Entropy\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eMeasures the degree of dependence inhomogeneity\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eAlis 2021(\u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e), Jang 2020(\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e), Mancio 2021(\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eGray Level Non-Uniformity\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eMeasures the similarity of intensity values in the image\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eAlis 2021(\u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e), Jang 2020 (\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e), Raisi-Estrabragh 2020 (\u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\" morerows=\"2\" rowspan=\"3\"\u003e \u003cp\u003eTexture (GLRLM)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eShort Run Emphasis\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eMeasures the distribution of short run length. High value indicates more fine textures\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eAlis 2021(\u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e), Baessler 2018 (\u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e), Raisi-Estrabragh 2020 (\u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e), Mancio 2021 (\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eGray Level Non-Uniformity\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eMeasures the similarity of intensity values in the image\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eAlis 2021(\u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e), Jang 2020 (\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e), Raisi-Estrabragh 2020 (\u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eRun Length Non-Uniformity\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eMeasures the similarity of run lengths in the image. Lower value indicates more homogeneity in run lengths\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eAlis 2021(\u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e), Jang 2020 (\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e), Raisi-Estrabragh 2020 (\u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\" morerows=\"1\" rowspan=\"2\"\u003e \u003cp\u003eTexture (GLSZM)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eGray Level Non-Uniformity\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eMeasures the similarity of intensity values in the image\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eAlis 2021(\u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e), Jang 2020 (\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e), Raisi-Estrabragh 2020 (\u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eZone Entropy\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eMeasures randomness in distribution of zone sizes and gray levels. High value indicates more heterogeneity in texture patterns\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eAlis 2021(\u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e), Jang 2020 (\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e), Raisi-Estrabragh 2020 (\u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\" morerows=\"2\" rowspan=\"3\"\u003e \u003cp\u003eShape (2D)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eArea\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eBasis for diagnostic parameters (ESV, EDV, EF, Myocardial Mass)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eSphericity\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eDegree to which a segmented region approximates a circle\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eElongation\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eRelation between lengths of smallest and longest main axes of a segmented region\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\" morerows=\"5\" rowspan=\"6\"\u003e \u003cp\u003eCardio-specific\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eRelative Septum Thickness\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eRatio of septum thickness to heart diameter, calculated on spline contours\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eTautz 2020 (\u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e27\u003c/span\u003e)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eArea Ratio LV/RV\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eRatio of LV area to RV area, calculated on spline contours\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eRelative RV Diameter\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eRatio of RV diameter to heart diameter, calculated on spline contours\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eRadial Strain (min, max, mean)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\" morerows=\"1\" rowspan=\"2\"\u003e \u003cp\u003eShape deformation due to myocardial dynamics, calculated on voxelized LVM\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eCircumferential Strain (min, max, mean)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eLeft Blood Pool Fractal Dimension\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eComplexity of boundary (e.g. due to trabeculation), calculated on voxelized LVBP\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eCaptur 2013 (\u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003e \u003cem\u003eFeature Calculation\u003c/em\u003e \u003c/p\u003e \u003cp\u003eAll features were calculated by a prototypical software implemented in MeVisLab (version 3.5a, MeVis Medical Solutions AG, Bremen, Germany). To evaluate the myocardium independently of patient size, LVM Area Ratio is calculated as LVM Area / (LVM Area\u0026thinsp;+\u0026thinsp;LVBP Area). Relative septum thickness represents the distance between the intersections of the LVM contours with the line connecting the LVBP and RVBP centers. The relative RV diameter is calculated analogously. Circumferential and radial strain statistics are based on analyzing voxel-wise deformation vectors relative to end-systole (first frame) with strain tensors, representing Lagrangianstrain (\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e). The fractal Minkowski-Bouligand (box counting) dimension is derived from the voxelized LVBP foreground segmentation and quantifies the complexity of trabecularization and blood pool boundaries (\u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e). (Further details in the online supplement).\u003c/p\u003e \u003cp\u003e \u003cem\u003eFeature Aggregation and Time Series\u003c/em\u003e \u003c/p\u003e \u003cp\u003eFeatures were calculated for each slice and time point, then aggregated (sum, mean, median, min, max) to create feature time curves. To assess reproducibility with respect to temporal resolution, we linearly interpolated all-time series to 50 phases.\u003c/p\u003e \u003cp\u003e \u003cb\u003eStatistical Analysis of Feature Robustness\u003c/b\u003e \u003c/p\u003e \u003cp\u003e \u003cem\u003eAssessment of Segmentation Results\u003c/em\u003e \u003c/p\u003e \u003cp\u003eThe inter- and intra-observer variation of the segmentation was assessed with the Dice similarity coefficient (DSC), the Hausdorff distance (HD), and the average contour distance (ACD) in MeVisLab. DSC quantifies the voxel overlap of the rasterized segmentation; HD is defined as the maximum and ACD as the mean of the minimal distances between points of the compared contours.\u003c/p\u003e \u003cp\u003e \u003cem\u003eAssessment of Radiomics Feature Reproducibility\u003c/em\u003e \u003c/p\u003e \u003cp\u003eTo assess the reproducibility of radiomics features, we performed an inter- and intra-observer variability analysis. We analyzed the difference of the feature maxima between the high and low temporal resolution to quantify the potential information loss due to lower sampling rate.\u003c/p\u003e \u003cp\u003eFor reproducibility analysis we utilized the intra-class correlation coefficient (ICC) calculated with Pingouin 0.4.0 (\u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e) as a two-way random model, with single measures and absolute agreement (equivalent to ICC (\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e, \u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e) in the R psych package). We adopted the levels of reliability from Koo and Li (\u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e), with threshold values of 0.5 (poor reliability), 0.75 (moderate), 0.9 (good) and 1.0 (excellent).\u003c/p\u003e"},{"header":"3. Results","content":"\u003cp\u003e \u003cb\u003eObserver Agreement at Different Spatio-Temporal Resolutions\u003c/b\u003e \u003c/p\u003e \u003cp\u003eHigher agreement was generally found at the higher spatio-temporal resolution, except for O2\u0026rsquo;s segmentations of LVM and LVBP (see Table\u0026nbsp;\u003cspan refid=\"Tab3\" class=\"InternalRef\"\u003e3\u003c/span\u003e and Fig.\u0026nbsp;2). O1 had an equal or better agreement than O2 in most cases, except for a higher HD for RVBP at the high resolution. Inter-observer agreement was also higher for high-resolution segmentations (see Table\u0026nbsp;\u003cspan refid=\"Tab3\" class=\"InternalRef\"\u003e3\u003c/span\u003e and Fig.\u0026nbsp;2).\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab3\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 3\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eIntra- and inter-observer agreement of the segmentations measured by Dice similarity coefficient (DSC), Hausdorff distance (HD) and average contour distance (ACD). * indicates the highest agreement.\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"11\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c6\" colnum=\"6\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c7\" colnum=\"7\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c8\" colnum=\"8\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c9\" colnum=\"9\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c10\" colnum=\"10\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c11\" colnum=\"11\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eStructure\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eSpatio-temporal resolution\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colspan=\"3\" nameend=\"c5\" namest=\"c3\"\u003e \u003cp\u003eMean DSC\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colspan=\"3\" nameend=\"c8\" namest=\"c6\"\u003e \u003cp\u003eMean HD [mm]\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colspan=\"3\" nameend=\"c11\" namest=\"c9\"\u003e \u003cp\u003eMean ACD [mm]\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c4\" namest=\"c3\"\u003e \u003cp\u003eIntra\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eInter\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c7\" namest=\"c6\"\u003e \u003cp\u003eIntra\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003eInter\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c10\" namest=\"c9\"\u003e \u003cp\u003eIntra\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c11\"\u003e \u003cp\u003eInter\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e\u003cb\u003eO1\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e\u003cb\u003eO2\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e\u003cb\u003eO1\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e\u003cb\u003eO2\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003e\u003cb\u003eO1\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c10\"\u003e \u003cp\u003e\u003cb\u003eO2\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c11\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\" morerows=\"1\" rowspan=\"2\"\u003e \u003cp\u003eLeft Ventricular Myocardium (LVM)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eHigh\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e\u003cb\u003e0.88*\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e\u003cb\u003e0.84\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e\u003cb\u003e0.88\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e\u003cb\u003e1.98*\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e2.32\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e\u003cb\u003e1.96\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003e\u003cb\u003e0.49*\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c10\"\u003e \u003cp\u003e\u003cb\u003e0.65\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c11\"\u003e \u003cp\u003e\u003cb\u003e0.28\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eLow\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0.86*\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0.83\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.83\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e2.16*\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e\u003cb\u003e2.29\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e2.34\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003e0.52*\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c10\"\u003e \u003cp\u003e\u003cb\u003e0.65\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c11\"\u003e \u003cp\u003e0.40\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\" morerows=\"1\" rowspan=\"2\"\u003e \u003cp\u003eLeft Ventricular Blood Pool (LVBP)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eHigh\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e\u003cb\u003e0.94*\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e\u003cb\u003e0.89\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e\u003cb\u003e0.92\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e\u003cb\u003e1.84*\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e2.31\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e\u003cb\u003e1.95\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003e\u003cb\u003e0.78*\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c10\"\u003e \u003cp\u003e1.13\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c11\"\u003e \u003cp\u003e\u003cb\u003e0.62\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eLow\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0.92*\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e\u003cb\u003e0.89\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.89\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e2.09*\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e\u003cb\u003e2.26\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e2.34\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003e0.95*\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c10\"\u003e \u003cp\u003e\u003cb\u003e1.05\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c11\"\u003e \u003cp\u003e0.90\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\" morerows=\"1\" rowspan=\"2\"\u003e \u003cp\u003eRight Ventricular Blood Pool (RVBP)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eHigh\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e\u003cb\u003e0.98*\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e\u003cb\u003e0.97\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e\u003cb\u003e0.99\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e\u003cb\u003e1.94\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e\u003cb\u003e1.41*\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e\u003cb\u003e1.36\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003e\u003cb\u003e0.18*\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c10\"\u003e \u003cp\u003e\u003cb\u003e0.25\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c11\"\u003e \u003cp\u003e\u003cb\u003e0.08\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eLow\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0.91*\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0.91*\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.91\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e3.72*\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e4.93\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e4.51\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003e0.53*\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c10\"\u003e \u003cp\u003e0.82\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c11\"\u003e \u003cp\u003e0.82\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003e \u003cb\u003eInfluence of Segmentation on Feature Reproducibility\u003c/b\u003e \u003c/p\u003e \u003cp\u003e \u003cem\u003eIntra-Observer Reproducibility\u003c/em\u003e \u003c/p\u003e \u003cp\u003eThe ICC values for intra-observer reproducibility of high-resolution features are shown in Fig.\u0026nbsp;3 and Table\u0026nbsp;\u003cspan refid=\"Tab4\" class=\"InternalRef\"\u003e4\u003c/span\u003e. O1 demonstrates a higher agreement than O2. LVM features show the most observer dependence, with O1 achieving good/excellent reproducibility for all except Elongation (moderate). O2\u0026rsquo;s LVM reproducibility ranges from moderate to excellent. LVBP features are mostly good/excellent, except for Sphericity, which has poor agreement. RVBP features show excellent reproducibility, though RVBP Sphericity has a lower ICC. Cardio-specific features are mostly good/excellent, with O2 outperforming O1 in six out of ten features.\u003c/p\u003e \u003cp\u003eOf the 64 features we found (57, 4, 2, 1) O1\u0026rsquo;s features and (56, 7, 0, 1) O2\u0026rsquo;s features to be of (excellent, good, moderate, poor) intra-observer reproducibility, respectively.\u003c/p\u003e \u003cp\u003e \u003cem\u003eInter-Observer Reproducibility\u003c/em\u003e \u003c/p\u003e \u003cp\u003eThe ICC values for the inter-observer agreement of features are displayed in Fig.\u0026nbsp;3. and Table\u0026nbsp;\u003cspan refid=\"Tab4\" class=\"InternalRef\"\u003e4\u003c/span\u003e. We observe a higher ICC for the high-resolution features with excellent reproducibility for 54 of 64 features and 39 in low-resolution scans.\u003c/p\u003e \u003cp\u003eOf the three anatomical structures, the LVM\u0026rsquo;s features show the lowest ICC values for the two resolutions, as well as the largest feature-wise variances. All features extracted from high-resolution images exhibit good or excellent reproducibility. Of the features originating from low-resolution sequences, only Elongation, GLDM Dependence Non-Uniformity, GLDM Dependence Entropy, and GLSZM Zone Entropy exhibit moderate inter-observer reproducibility, all other features\u0026rsquo; ICC scores are good or excellent.\u003c/p\u003e \u003cp\u003eAs for the intra-observer comparison, LVBP Sphericity exhibits poor agreement at both resolutions in the inter-observer comparison. All other LVBP features are of good or excellent reproducibility.\u003c/p\u003e\u003cp\u003eTable 4 Number of features with excellent, good, moderate, and poor inter- and intra-observer, and inter-resolution Intra-Class Correlation (ICC) values. Inter-resolution ICC are given separately by slice.\u003c/p\u003e \n\u003cp\u003e\u003cem\u003eTable 4. Number of features with excellent, good, moderate, and poor inter- and intra-observer, and inter-resolution Intra-Class Correlation (ICC) values. Inter-resolution ICC are given separately by slice.\u003c/em\u003e\u003cem\u003e\u0026nbsp;\u003c/em\u003e\u003c/p\u003e\n\u003ctable border=\"1\" cellspacing=\"0\" cellpadding=\"0\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd rowspan=\"3\" valign=\"bottom\" style=\"width: 82px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eICC Class\u003c/strong\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 80px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eLow Resolution\u003c/strong\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"3\" style=\"width: 242px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eHigh Resolution\u003c/strong\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"3\" style=\"width: 194px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eInter-Resolution\u003c/strong\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd rowspan=\"2\" valign=\"top\" style=\"width: 80px;\"\u003e\n \u003cp\u003eInter-observer\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd rowspan=\"2\" valign=\"top\" style=\"width: 81px;\"\u003e\n \u003cp\u003eInter-observer\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 161px;\"\u003e\n \u003cp\u003eIntra-observer\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd rowspan=\"2\" style=\"width: 65px;\"\u003e\n \u003cp\u003eApical\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd rowspan=\"2\" style=\"width: 65px;\"\u003e\n \u003cp\u003eMid Ventricle\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd rowspan=\"2\" style=\"width: 65px;\"\u003e\n \u003cp\u003eBasal\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 81px;\"\u003e\n \u003cp\u003eObserver 1\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 81px;\"\u003e\n \u003cp\u003eObserver 2\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 82px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eExcellent\u003c/strong\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 80px;\"\u003e\n \u003cp\u003e39\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 81px;\"\u003e\n \u003cp\u003e54\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 81px;\"\u003e\n \u003cp\u003e57\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 81px;\"\u003e\n \u003cp\u003e56\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 65px;\"\u003e\n \u003cp\u003e4\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 65px;\"\u003e\n \u003cp\u003e3\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 65px;\"\u003e\n \u003cp\u003e1\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 82px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eGood\u003c/strong\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 80px;\"\u003e\n \u003cp\u003e19\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 81px;\"\u003e\n \u003cp\u003e9\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 81px;\"\u003e\n \u003cp\u003e4\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 81px;\"\u003e\n \u003cp\u003e7\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 65px;\"\u003e\n \u003cp\u003e9\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 65px;\"\u003e\n \u003cp\u003e5\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 65px;\"\u003e\n \u003cp\u003e12\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 82px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eModerate\u003c/strong\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 80px;\"\u003e\n \u003cp\u003e5\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 81px;\"\u003e\n \u003cp\u003e0\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 81px;\"\u003e\n \u003cp\u003e2\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 81px;\"\u003e\n \u003cp\u003e0\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 65px;\"\u003e\n \u003cp\u003e23\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 65px;\"\u003e\n \u003cp\u003e23\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 65px;\"\u003e\n \u003cp\u003e20\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 82px;\"\u003e\n \u003cp\u003e\u003cstrong\u003ePoor\u003c/strong\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 80px;\"\u003e\n \u003cp\u003e1\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 81px;\"\u003e\n \u003cp\u003e1\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 81px;\"\u003e\n \u003cp\u003e1\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 81px;\"\u003e\n \u003cp\u003e1\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 65px;\"\u003e\n \u003cp\u003e28\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 65px;\"\u003e\n \u003cp\u003e33\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 65px;\"\u003e\n \u003cp\u003e31\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003eAll features of the RVBP are found to have good or excellent ICC scores, with the features from high-resolution images scoring consistently in the excellent range, except for Sphericity.\u003c/p\u003e \u003cp\u003e The cardio-specific features show good or excellent agreement at both resolutions, apart from the Area Ratio LV/RV, which has only moderate reproducibility at low resolution.\u003c/p\u003e \u003cp\u003e \u003cem\u003eInfluence of Spatio-Temporal Resolution on Feature Reproducibility\u003c/em\u003e \u003c/p\u003e \u003cp\u003eThe ICC values for the different spatio-temporal resolutions for each of the three slices are displayed in Fig.\u0026nbsp;2.\u003c/p\u003e \u003cp\u003eMost 2D shape features show good or excellent reproducibility between low and high spatiotemporal resolution and do not vary significantly regarding the slice position. The features showing greater variability between low and high resolution are also influenced by the slice position; features of LVBP vary strongest in mid ventricular slices.\u003c/p\u003e \u003cp\u003e \u003cem\u003eInfluence of Spatio-Temporal Resolution on Maximum Feature Values\u003c/em\u003e \u003c/p\u003e \u003cp\u003eFigure \u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e4\u003c/span\u003e shows how maximum feature values vary with resolution. A zero deviation means no change, negative values indicate lower maxima at low resolution, and positive values indicate higher maxima at low resolution.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003e \u003cem\u003eComparison of Robust Feature Curves in Controls and HF Patients\u003c/em\u003e \u003c/p\u003e \u003cp\u003eFigure 5 shows the temporal features curves for aggregated high-resolution case features of EMPATHY volunteer datasets compared with features derived from the patients with the three HF subtypes.\u003c/p\u003e\u003cp\u003eThe average values of robust radiomics features showed a greater difference between healthy controls, HFpEF, HFmrEF, and HFrEF patients for LVM and LVBP than for RVBP.\u003c/p\u003e \u003cp\u003eThe LVBP area and run-length-non-uniformity curves have non-overlapping value ranges between the groups with normal and reduced EF.\u003c/p\u003e \u003cp\u003eThe LVM area ratio and LVM sphericity features curves showed non-overlapping value ranges for the age-matched control group and HFpEF patients in the diastolic phase (p-value\u0026thinsp;\u0026lt;\u0026thinsp;0.05).\u003c/p\u003e \u003cp\u003eThe relative septum thickness and the circumferential strain curves of the four groups differ in curve shape but have overlapping value ranges.\u003c/p\u003e \u003cp\u003eRobust 2D LVM shape features, like Area Ratio, Sphericity, and Strain, showed the largest differences across healthy, HFpEF, HFmrEF, and HFrEF groups. Shape features and septal thickness also differed significantly between healthy and HFpEF (p\u0026thinsp;\u0026lt;\u0026thinsp;0.05; see Supplementary).\u003c/p\u003e \u003cp\u003eDependance and Run Length Uniformity of LVBP differed significantly (p\u0026thinsp;\u0026lt;\u0026thinsp;0.05; see Supplementary) between individuals with preserved and non-preserved EF, with no overlap in values.\u003c/p\u003e"},{"header":"4. Discussion","content":"\u003cp\u003e \u003cem\u003eSegmentation Reproducibility\u003c/em\u003e \u003c/p\u003e \u003cp\u003eA higher spatio-temporal resolution was correlated with a better intra- and inter-observer segmentation reproducibility for LVM, LVBP and RVBP (Table\u0026nbsp;\u003cspan refid=\"Tab3\" class=\"InternalRef\"\u003e3\u003c/span\u003e). We could also observe that the experience of the radiologist was correlated with a better intra-observer reproducibility for LVM and LVBP. In contrast to previous studies of LVM und LVBP (\u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e) the intra- and inter-observer reproducibility were comparable.\u003c/p\u003e \u003cp\u003eThe differences in previous studies could be explained through the observers' strong experience (high intra-observer reproducibility) but different training (higher inter-observer variability than in our study).\u003c/p\u003e \u003cp\u003eIn line with our observation on segmentation agreement we found inter-observer feature reproducibility to be as good as intra-observer reproducibility (Fig.\u0026nbsp;3). These results differ from previous studies (\u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e) (\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e).\u003c/p\u003e \u003cp\u003e \u003cem\u003eIntra-Observer and Inter-Observer Reproducibility\u003c/em\u003e \u003c/p\u003e \u003cp\u003eMost Radiomics features showed good to excellent intra-observer reproducibility, with LVM features of experienced O1 being more reproducible than those of O2. The features that were found reproducible in the study of Jang 2020 showed similar results in our study.\u003c/p\u003e \u003cp\u003eO2 achieved higher agreement than O1 for cardio-specific features. The strain features showed a good to excellent reproducibility. In the study from Schmidt et al (\u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e) the strain features showed an excellent intra-observer reproducibility except for radial SR. Segmentation agreement is not explicitly reported but is assumingly causing the inter-observer differences.\u003c/p\u003e \u003cp\u003eWe observe a better inter-observer agreement on high spatiotemporal resolution images, with LVM\u0026rsquo;s features showing the largest differences in ICC for the two resolutions (Fig.\u0026nbsp;3). 62 of 64 features for high resolution (96%) and 51 of 64 features for low resolution (79%) showed an ICC\u0026thinsp;\u0026ge;\u0026thinsp;0.8. Jang et al. 2020 (\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e) found only 32.1% features showing an ICC in this range. The difference is probably caused by differences in segmentation agreement or in spatial LV coverage between the two studies. Figures\u0026nbsp;3 and \u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e4\u003c/span\u003e illustrate how the ICC and feature variation depend on slice location.\u003c/p\u003e \u003cp\u003e \u003cem\u003eInfluence of Slice Position and Spatio-Temporal Sampling\u003c/em\u003e \u003c/p\u003e \u003cp\u003eOur results are consistent with the previous findings in that spatial resolution is a major cause of variability of feature values (\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e, \u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eThe ICC score for LVBP and RVBP texture features was worse in the midventricular slice which may be explained by the greater mobility of the heart at this level. This agrees with the study of Alis et al (\u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e) where 55% of the features showed great variability through the cardiac cycle. For LVM, the agreement was lower at basal level where the segmentation agreement was worse.\u003c/p\u003e \u003cp\u003eFew studies analyze cardiac phases other than end-diastolic and end-systolic time points, but there is limited evidence that features incorporating temporal information are more reproducible (\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e, \u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e, \u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e27\u003c/span\u003e, \u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e28\u003c/span\u003e).\u003c/p\u003e \u003cp\u003ePrevious studies found associations between radiomics feature curve max/min values and traditional cardiac function and motions indices (\u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e, \u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e29\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eFeature variability was higher in diastole than systole (Fig.\u0026nbsp;5), with less pronounced extremes at low resolution. Similar findings were reported by Backhaus et al. (\u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e) for temporal resolution effects on strain values.\u003c/p\u003e \u003cp\u003eHowever, we observed a higher dependence of feature reproducibility on spatial than temporal resolution (see. Supplements).\u003c/p\u003e \u003cp\u003e \u003cem\u003eComparison between Feature Curves of the Controls and HF patients\u003c/em\u003e \u003c/p\u003e \u003cp\u003eThe analysis of robust radiomics feature curves between controls and patients with different HF-subtypes showed expected results for LVM 2D Shape features and cardiospecific features (Fig.\u0026nbsp;5).\u003c/p\u003e \u003cp\u003eIn line with previous publications, the circumferential strain curves of the four subgroups show different courses. However, the value ranges of the different groups overlap. In contrast to the findings in related work (\u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e30\u003c/span\u003e), the difference in the strain values of healthy individuals and HFpEF patients was not significant in our analysis (p\u0026thinsp;=\u0026thinsp;0.24). Robust 2D Shape LVM features show non-overlapping value ranges for the controls and HFpEF subgroups with corresponding p-values\u0026thinsp;\u0026lt;\u0026thinsp;0.05 which might enable the classification of the four subgroups based on a combination of LVBP and LVM features.\u003c/p\u003e \u003cp\u003e \u003cb\u003eLimitations\u003c/b\u003e \u003c/p\u003e \u003cp\u003eThe number of participants included limited our analysis. In the age-matched controls the proportion of women was higher than among the HF-patients.\u003c/p\u003e \u003cp\u003eA multicentric study using different scans would provide a better accuracy of evaluation of the applicability of those technic in clinical settings, this could be the subject of a future study.\u003c/p\u003e"},{"header":"5. Conclusions","content":"\u003cp\u003eThe study demonstrates the need to align the selection of radiomics parameters and acquisition protocols. Our results imply the necessity to select high-resolution standardized protocols for the clinical use of some specific radiomics texture parameters such as Joint Entropy. Standarization is especially relevant in cardiac MRI, where long scan times challenge time management and patient comfort. Using reliable radiomics features characterizing the LVM such as area ratio, sphericity and GLRLM non-uniformity might enable the characterization of different types of heart failure.\u003c/p\u003e"},{"header":"Abbreviations","content":"\u003cp\u003eCMR: cardiac MRI\u003c/p\u003e\n\u003cp\u003eLVBP: Left ventricle blood pool\u003c/p\u003e\n\u003cp\u003eLVM: left ventricle myocardium\u003c/p\u003e\n\u003cp\u003eRVBP: right ventricle blood pool\u003c/p\u003e\n\u003cp\u003eO1: observer 1\u003c/p\u003e\n\u003cp\u003eO2: observer 2\u003c/p\u003e\n\u003cp\u003eHF: Heart Failure\u003c/p\u003e\n\u003cp\u003eHFrEF: Heart Failure with reduced ejection fraction\u003c/p\u003e\n\u003cp\u003eHFpEF: Heart Failure with preserved ejection fraction\u003c/p\u003e\n\u003cp\u003eHFmrEF: Heart Failure with mild reduced ejection fraction\u003c/p\u003e\n\u003cp\u003eDSC:Dice similarity coefficient (DSC)\u003c/p\u003e\n\u003cp\u003eHD: Hausdorff distance\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eACD: average contour distance (ACD)\u003c/p\u003e\n\u003cp\u003eICC: Intraclass correlation coefficient\u003c/p\u003e"},{"header":"Declarations","content":"\u003ch2\u003eConflicts of Interests\u003c/h2\u003e \u003cp\u003eThe authors declare no conflict of interest.\u003c/p\u003e \u003ch2\u003eAcknowledgments / Funding\u003c/h2\u003e \u003cp\u003eThis work has been funded by the German Ministry for Education and Research (BIFOLD, 01IS18037E), the German Research Foundation (SFB 1470, INST 335/815-1, 515294457, HE 7312/7\u0026thinsp;\u0026minus;\u0026thinsp;1), and the DZHK (German Centre for Cardiovascular Research).\u003c/p\u003e \u003c/p\u003e\u003ch2\u003eAuthor Contribution\u003c/h2\u003e\u003cp\u003eAnja Hennemuth, Sebastian Kelle, Djawid Hashemi, Frank Edelmann and Jeanette Schulz-Menger conceived the presented idea. Lea Ter-Minassian contributed to the data management and carried out the experiments with the help of Hannu Zhang. Markus H\u0026uuml;llebrand developed the software program. Ann Laube,Lennart Tautz and Matthias Ivantsits performed the analytic calculations. Jennifer Erley verified the analytical method. Lea Ter-Minassian wrote the manuscript. Anja Hennemuth supervised the work with the help of Markus H\u0026uuml;llebrand. All authors discussed the results and contributed to the final manuscript. All authors reviewed the manuscript.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\n\u003cli\u003eKolossv\u0026aacute;ry M, Kar\u0026aacute;dy J, Szilveszter B, Kitslaar P, Hoffmann U, Merkely B, et al. Radiomic Features Are Superior to Conventional Quantitative Computed Tomographic Metrics to Identify Coronary Plaques with Napkin-Ring Sign. Circ Cardiovasc Imaging. 2017 Dec 1;10(12). \u003c/li\u003e\n\u003cli\u003eLi XN, Yin WH, Sun Y, Kang H, Luo J, Chen K, et al. Identification of pathology-confirmed vulnerable atherosclerotic lesions by coronary computed tomography angiography using radiomics analysis. Eur Radiol. 2022 Jun;32(6):4003\u0026ndash;13. \u003c/li\u003e\n\u003cli\u003eMancio J, Pashakhanloo F, El-Rewaidy H, Jang J, Joshi G, Csecs I, et al. Machine learning phenotyping of scarred myocardium from cine in hypertrophic cardiomyopathy. Available from: https://academic.oup.com/ehjcimaging/article/23/4/532/6199608\u003c/li\u003e\n\u003cli\u003eFahmy AS, Rowin EJ, Arafati A, Al-Otaibi T, Maron MS, Nezafat R. Radiomics and deep learning for myocardial scar screening in hypertrophic cardiomyopathy. Journal of Cardiovascular Magnetic Resonance. 2022 Dec 1;24(1). \u003c/li\u003e\n\u003cli\u003eZwanenburg A, Valli\u0026egrave;res M, Abdalah MA, Aerts HJWL, Andrearczyk V, Apte A, et al. The Image Biomarker Standardization Initiative: Standardized Quantitative Radiomics for High-Throughput Image-based Phenotyping. Radiology. 2020 May;295(2):328\u0026ndash;38. \u003c/li\u003e\n\u003cli\u003eBernatz S, Zhdanovich Y, Ackermann J, Koch I, Wild PJ, Pinto dos Santos D, et al. Impact of rescanning and repositioning on radiomic features employing a multi-object phantom in magnetic resonance imaging. Scientific Reports |. 123AD;11:14248. \u003c/li\u003e\n\u003cli\u003eParekh V, Jacobs MA. Radiomics: a new application from established techniques. Expert Rev Precis Med Drug Dev. 2016 Mar 3;1(2):207\u0026ndash;26. \u003c/li\u003e\n\u003cli\u003eChang S, Han K, Suh YJ, Choi BW. Quality of science and reporting for radiomics in cardiac magnetic resonance imaging studies: a systematic review. Eur Radiol. 2022 Jul 1;32(7):4361\u0026ndash;73. \u003c/li\u003e\n\u003cli\u003eHassani C, Saremi F, Varghese BA, Duddalwar V. Myocardial Radiomics in Cardiac MRI. American Journal of Roentgenology. 2020 Mar;214(3):536\u0026ndash;45. \u003c/li\u003e\n\u003cli\u003eInoue Y, Nomura Y, Nakaoka T, Watanabe M, Kiryu S, Okubo T, et al. Effect of temporal resolution on the estimation of left ventricular function by cardiac MR imaging. Magn Reson Imaging. 2005 Jun;23(5):641\u0026ndash;5. \u003c/li\u003e\n\u003cli\u003eBackhaus SJ, Metschies G, Billing M, Schmidt-Rimpler J, Kowallick JT, Gertz RJ, et al. Defining the optimal temporal and spatial resolution for cardiovascular magnetic resonance imaging feature tracking. Journal of Cardiovascular Magnetic Resonance. 2021 Dec 1;23(1). \u003c/li\u003e\n\u003cli\u003eYang W, Li H, He J, Yin G, An J, Forman C, et al. Left Ventricular Strain Measurements Derived from MR Feature Tracking: A Head‐to‐Head Comparison of a Higher Temporal Resolution Method With a Conventional Method. Journal of Magnetic Resonance Imaging. 2022 Sep 10;56(3):801\u0026ndash;11. \u003c/li\u003e\n\u003cli\u003eSchofield R, Ganeshan B, Fontana M, Nasis A, Castelletti S, Rosmini S, et al. Texture analysis of cardiovascular magnetic resonance cine images differentiates aetiologies of left ventricular hypertrophy. Clin Radiol. 2019 Feb;74(2):140\u0026ndash;9. \u003c/li\u003e\n\u003cli\u003eJang J, Ngo LH, Mancio J, Kucukseymen S, Rodriguez J, Pierce P, et al. Reproducibility of Segmentation-based Myocardial Radiomic Features with Cardiac MRI. Radiol Cardiothorac Imaging. 2020 Jun 25;2(3):e190216. \u003c/li\u003e\n\u003cli\u003eJang J, El-Rewaidy H, Ngo LH, Mancio J, Csecs I, Rodriguez J, et al. Sensitivity of Myocardial Radiomic Features to Imaging Parameters in Cardiac MR Imaging. J Magn Reson Imaging. 2021 Sep;54(3):787\u0026ndash;94. \u003c/li\u003e\n\u003cli\u003eAlis D, Yergin M, Asmakutlu O, Topel C, Karaarslan E. The influence of cardiac motion on radiomics features: radiomics features of non-enhanced CMR cine images greatly vary through the cardiac cycle. Eur Radiol. 2021 May 13;31(5):2706\u0026ndash;15. \u003c/li\u003e\n\u003cli\u003eRaisi-Estabragh Z, Gkontra P, Jaggi A, Cooper J, Augusto J, Bhuva AN, et al. Repeatability of Cardiac Magnetic Resonance Radiomics: A Multi-Centre Multi-Vendor Test-Retest Study. Front Cardiovasc Med. 2020 Dec 2;7. \u003c/li\u003e\n\u003cli\u003eHuellebrand M, Ivantsits M, Tautz L, Kelle S, Hennemuth A. A Collaborative Approach for the Development and Application of Machine Learning Solutions for CMR-Based Cardiac Disease Classification. Front Cardiovasc Med. 2022 Mar 10;9. \u003c/li\u003e\n\u003cli\u003eSalgo IS, Tsang W, Ackerman W, Ahmad H, Chandra S, Cardinale M, et al. Geometric Assessment of Regional Left Ventricular Remodeling by Three-Dimensional Echocardiographic Shape Analysis Correlates with Left Ventricular Function. Journal of the American Society of Echocardiography. 2012 Jan;25(1):80\u0026ndash;8. \u003c/li\u003e\n\u003cli\u003eTobon-Gomez C, De Craene M, McLeod K, Tautz L, Shi W, Hennemuth A, et al. Benchmarking framework for myocardial tracking and deformation algorithms: An open access database. Med Image Anal. 2013 Aug;17(6):632\u0026ndash;48. \u003c/li\u003e\n\u003cli\u003eCaptur G, Muthurangu V, Cook C, Flett AS, Wilson R, Barison A, et al. Quantification of left ventricular trabeculae using fractal analysis. Journal of Cardiovascular Magnetic Resonance. 2013 Jan;15(1):36. \u003c/li\u003e\n\u003cli\u003eVallat R. Pingouin: statistics in Python. J Open Source Softw. 2018 Nov 19;3(31):1026. \u003c/li\u003e\n\u003cli\u003eKoo TK, Li MY. A Guideline of Selecting and Reporting Intraclass Correlation Coefficients for Reliability Research. J Chiropr Med. 2016 Jun;15(2):155\u0026ndash;63. \u003c/li\u003e\n\u003cli\u003eSuinesiaputra A, Bluemke DA, Cowan BR, Friedrich MG, Kramer CM, Kwong R, et al. Quantification of LV function and mass by cardiovascular magnetic resonance: Multi-center variability and consensus contours. Journal of Cardiovascular Magnetic Resonance. 2015 Jul 28;17(1). \u003c/li\u003e\n\u003cli\u003eBae\u0026szlig;ler B, Weiss K, Pinto dos Santos D. Robustness and Reproducibility of Radiomics in Magnetic Resonance Imaging. Invest Radiol. 2019 Apr;54(4):221\u0026ndash;8. \u003c/li\u003e\n\u003cli\u003eSchmidt B, Dick A, Treutlein M, Schiller P, Bunck AC, Maintz D, et al. Intra- and inter-observer reproducibility of global and regional magnetic resonance feature tracking derived strain parameters of the left and right ventricle. Eur J Radiol. 2017 Apr;89:97\u0026ndash;105. \u003c/li\u003e\n\u003cli\u003eTautz L, Zhang H, H\u0026uuml;llebrand M, Ivantsits M, Kelle S, Kuehne T, et al. Cardiac radiomics: an interactive approach for 4D data exploration. Current Directions in Biomedical Engineering. 2020 Sep 17;6(1). \u003c/li\u003e\n\u003cli\u003eLarroza A, L\u0026oacute;pez‐Lereu MP, Monmeneu J V., Gavara J, Chorro FJ, Bod\u0026iacute; V, et al. Texture analysis of cardiac cine magnetic resonance imaging to detect nonviable segments in patients with chronic myocardial infarction. Med Phys. 2018 Apr 22;45(4):1471\u0026ndash;80. \u003c/li\u003e\n\u003cli\u003eLin K, Sarnari R, Carr JC, Markl M. Cine \u0026lt;scp\u0026gt;MRI\u0026lt;/scp\u0026gt; ‐Derived Radiomics Features of the Cardiac Blood Pool: Periodicity, Specificity, and Reproducibility. Journal of Magnetic Resonance Imaging. 2023 Sep 19;58(3):807\u0026ndash;14. \u003c/li\u003e\n\u003cli\u003eWitt UE, M\u0026uuml;ller ML, Beyer RE, Wieditz J, Salem S, Hashemi D, et al. A simplified approach to discriminate between healthy subjects and patients with heart failure using cardiac magnetic resonance myocardial deformation imaging Keywords global longitudinal strain \u0026bull; cut-off \u0026bull; early identification of heart failure \u0026bull; cardiac magnetic resonance imaging \u0026bull; deformation imaging. European Heart Journal - Imaging Methods and Practice [Internet]. 2024;2:93. Available from: https://doi.org/10.1093/ehjimp/qyae093\u003c/li\u003e\n\u003c/ol\u003e"},{"header":"Table 1 - embedded object","content":"\u003cp\u003eThe embedded object within Table 1 in the last column for the row labeled \"Slice spacing [mm]\" is not available with this version.\u003c/p\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":true,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true},"keywords":"Cardiac MRI, Reproducibility, Variability, Radiomics, Cine, Heart failure","lastPublishedDoi":"10.21203/rs.3.rs-6912341/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-6912341/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003ch2\u003ePurpose\u003c/h2\u003e \u003cp\u003e \u003cspan type=\"SmallCaps\" class=\"SmallCaps\" name=\"Emphasis\"\u003eThis work investigates the reproducibility of radiomics features of cardiac cine MRI and their applicability for heart failure (HF) characterization.\u003c/span\u003e \u003c/p\u003e\u003ch2\u003eMaterial and methods\u003c/h2\u003e \u003cp\u003e \u003cspan type=\"SmallCaps\" class=\"SmallCaps\" name=\"Emphasis\"\u003eShort-axis ECG-triggered cine CMR (cardiac MRI) sequences were acquired with different spatiotemporal resolutions. Segmentation was performed by two experts delineating left ventricle blood pool (LVBP), left ventricle myocardium (LVM), and right ventricle blood pool (RVBP). Intra- and interobserver segmentation agreement was evaluated with Dice Score, Hausdorff and Average Contour Distance. Feature reproducibility was assessed via intraclass correlation to evaluate the influence of image resolutions and observer variability. Feature curves of robust features were compared between HF subtypes and age-matched controls.\u003c/span\u003e \u003c/p\u003e\u003ch2\u003eResults\u003c/h2\u003e \u003cp\u003e \u003cspan type=\"SmallCaps\" class=\"SmallCaps\" name=\"Emphasis\"\u003eWe included twelve healthy volunteers (median age M: 28, interquartile range IQR: 9) for reproducibility analysis, and 19 controls without HF (M: 62, IQR: 11), 22 HF patients with preserved ejection fraction (HFpEF) (M: 77, IQR: 8),17 HF patients with mildly reduced EF (HFmrEF) (M: 73, IQR: 10), and 15 HF patients with reduced EF (HFrEF) (M: 62, IQR: 14). On high resolution respectively 61 and 63 of 64 features for observer 1 and 2 showed good to excellent intraobserver reproducibility. Higher spatiotemporal resolution had a positive impact on reproducibility. Average LVBP and LVM feature values differ between controls, HFpEF, HFmrEF, and HFrEF but value ranges overlap.\u003c/span\u003e \u003c/p\u003e\u003ch2\u003eConclusions\u003c/h2\u003e \u003cp\u003e \u003cspan type=\"SmallCaps\" class=\"SmallCaps\" name=\"Emphasis\"\u003eMost radiomics features showed good to excellent reproducibility, the spatiotemporal resolution being the main cause of variability followed by observer experience. Combined assessment of LVM and LVBP features could enable CMR-based HF subtype classification.\u003c/span\u003e \u003c/p\u003e","manuscriptTitle":"Cardiac Cine MRI Radiomics: Analysis of Feature Reproducibility and Evaluation of Variations in the Context of Heart Failure","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2025-07-02 09:43:10","doi":"10.21203/rs.3.rs-6912341/v1","editorialEvents":[{"type":"communityComments","content":0}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"dede2ead-7095-4ed1-9edb-aaa1f84ceb2e","owner":[],"postedDate":"July 2nd, 2025","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"posted","subjectAreas":[],"tags":[],"updatedAt":"2025-09-05T17:38:33+00:00","versionOfRecord":[],"versionCreatedAt":"2025-07-02 09:43:10","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-6912341","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-6912341","identity":"rs-6912341","version":["v1"]},"buildId":"8U1c8b4HqxoKbykW_rLl7","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}