This study was primarily designed to examine some of the fundamental aspects of 2D TA applied to non-contrast cine MR images of the LV myocardium. The work was performed at a single center and was standardized as much as possible by minimizing variations associated with the imaging factors, such as the same scanner, consistent MR protocol, image resolution, etc. By using this approach, our aim was to focus on searching for the highest power of each texture parameter to highlight differences associated with the myocardium, as opposed to being influenced by differences associated with the experimental conditions.

The original hypothesis of the work was to examine three different conditions, which we thought would influence the LV texture parameters by different amounts. Our expectation was that the LV TA variables would differ considerably with cardiac phase but be more stable with gender and age. This was largely borne out by our results (Tables 3–5), where mainly significant differences between the TA parameters were identified between those measured at ED versus ES, and those derived from males versus females. However, the TA parameters were more stable when it came to identifying differences between age groups, where the TA changes were considerably more subtle.

While TA has been shown in prior studies to differentiate cardiomyopathies and myocarditis from healthy controls—particularly when combined with mapping/late gadolinium enhancement (LGE)—we focused on cine images in this study for several reasons. First, this work was part of a larger population-based protocol assessing routine cardiac MR (CMR) variables alongside whole-body MR angiography (assessment of atherosclerotic plaque sites). Given that cine imaging is universally acquired in CMR, we aimed to rigorously evaluate TA on this widely available sequence, ensuring broader applicability compared to more specialized techniques (e.g., mapping or LGE). Our goal was to establish a foundational understanding of TA performance on standard cine data before progressing to advanced sequences. The study design (e.g., ED vs ES, male vs female, young vs old) remains equally valid for future TA applications in T1/T2 mapping or LGE—a logical next step.

It is known from previous studies that the ROI size can influence texture outcomes to some extent [4, 17, 24]. To investigate this further, we performed a small sub-study. By keeping the imaging parameters as consistent as possible, we were able to identify that most of the TA features tested were consistent with our ROI sizes, and those that were inconsistent were discarded from the rest of our work.

Of the n = 3 texture variables that were strongly correlated with ROI size, two of them were from the run-length matrix family (‘Average_RLNonUni’ and ‘Average_GLevNonU’), which calculates runs of pixels at a specific gray-level value in a definite orientation. The other variable was from the co-occurrence matrix-long distance family (‘S5-Average_Entropy’), which indicates the degree of disarray of the gray-level signal in the image. Similar results were also presented in a brain MRI TA study [25], where the authors hypothesized that the reason for ‘run-length matrix family’ variations is simply due to the fact that a larger ROI will result in more runs, or in the case of co-occurrence, more degrees of disarray. Of some concern, in many clinical studies of the utility of radiomics, gray-level non-uniformity has been one of the strongest texture features in assisting the separation of different myocardial disease [9, 26, 27]. Our work suggests these prior works should be interpreted with caution, as the difference between the groups may simply be picking up the difference in the degree of hypertrophy between the cohorts, rather than a unique texture fingerprint.

Texture variations between ED and ES images were hypothesized to be the largest; myocardial tissue pattern was quite different from each other between these two states [28]. The significant differences in myocardial texture between ED and ES observed during texture analysis may be attributed to several factors: (1) Myocardial thickening, strain, wall motion, and shear forces during the cardiac cycle alter the orientation and arrangement of myocardial fibers. These changes cause directional variations in tissue movement, leading to detectable texture differences [29]. (2) Intramyocardial blood flow: Perfusion differences between ED and ES affect local tissue contrast and texture patterns. (3) Stiffness and elasticity: The myocardium becomes stiffer during systole due to active contraction, whereas diastolic relaxation increases elasticity. These mechanical property changes influence texture characteristics [30]. (4) Imaging artifacts: Technique-dependent artifacts (e.g., in echocardiography) may also contribute to observed variations between ED and ES phases [31].

Previous studies that have reported cardiac MR texture analysis have preferred to focus on just one cardiac phase [6,7,8, 26, 32], while others have not mentioned the phase used, or have used the whole cardiac cycle of images [33]. Our findings confirm those of Alis et al [34], who examined radiomics in 59 healthy adults, and found that only 22 out of 352 showed strong stability over the cardiac cycle. Our study confirms these prior results, extending these by also showing that these differences hold true even after removing features associated with ROI size (which would be larger in systole). We also showed them to be present in a sex and age stratified analysis, as well as in considering the whole myocardium, or just the septal or free wall myocardium. The study highlights the importance of ensuring a consistent phase is used for radiomic analysis and warns against extrapolating findings from studies using systole to those using diastole.

Gender differences are reported in studies that males may have a greater ‘physiological’ hypertrophy or bigger myocytes, along with greater myocyte cell loss with ageing [35,36,37,38,39,40,41]. In our study, we found significant differences in myocardial texture between men and women. Interestingly, we found that sex differences were present in the younger groups, but disappeared in the oldest group (≥ 64 years). A similar pattern has been observed in T1 mapping, where gender differences in native T1 are well described [40, 42]. In one study, while there is a significant gender difference in younger people (< 45 years, where T1 values in female myocardium are higher than those of males), in older age groups (≥ 45 years), this difference disappears [40]. These effects may reflect differences in the myocardium brought about by sex hormones, with these differences then becoming less pronounced in the post-menopausal period. Unfortunately, we do not have the menopausal status of the women in this study, which will require further study in future studies.

Aging causes structural and functional changes of the heart, with measures such as myocardial triglyceride [43], cardiac fibrosis and cardiomyocyte hypertrophy are known to increase with age [41, 44,45,46,47,48]. In line with this, we have observed the texture features to change with age, which likely reflect several underlying physiological and microstructural changes. These features exhibit phase-dependent variability, with end-diastole providing the most reliable measurements. This temporal pattern primarily likely results from superior diastolic image quality, as the left ventricle achieves maximal volume and minimal motion during this quiescent phase of the cardiac cycle.

From the ‘helical ventricular myocardial band of Torrent-Guasp’ [49], we know that the LV three-dimensional structure is different between the septal and lateral walls, where two layers of the fibers radiate in different directions [50]. Also, in hypertrophic cardiomyopathy (HCM) patients, cardiac pathological abnormalities such as fibrosis are more commonly found at the septum [51]. It is unknown whether the MR textural basis of the septum and the lateral wall is different in a healthy population, and there is therefore a need to understand this regionally specific texture further. In our study, we found that TA at the septal wall was more sensitive to variations between genders and age groups, while the lateral wall texture was more stable with less variation related to cardiac phase, gender, and age groups. This is likely a consequence of the lateral wall being most susceptible to artifacts, including susceptibility artifacts and chemical shift artifacts due to the close proximity of the epicardial fat and adjacent lung tissue. Similar findings have been shown in T1 mapping, where the septum produces the most reliable and reproducible region for tissue characterization. Consequently, we suggest a preferential use of the septal wall when appropriate for radiomic analysis. This approach may be particularly useful in the early stages of disease. However, the whole myocardium is also recommended as an ROI for analysis due to several advantages: (1) The septum was found to be superior in detecting change relative to the lateral wall, but this superiority was not as evident when compared to the whole myocardium, especially when examining subtle changes such as those related to age. (2) The whole myocardium ROI offers more reproducible segmentation (a capability available on most MR workstations), ensuring greater measurement robustness. This approach is also more clinically practical and provides comprehensive coverage of myocardial tissue—particularly valuable for diffuse cardiomyopathies where pathological changes may not be regionally confined.

There were n = 15 texture features that were identified to be repeatable and feasible for use in this study (Descriptions of all 15 texture features are available in Supplementary Material S-5). However, the most reliable parameters may not necessarily demonstrate effectiveness for all occasions (e.g., S2-AVERAGE_SumOfSqs was most useful for the identification of different genders and age groups but could not be used for cardiac phase differentiation). The use of these TA variables in clinical studies may be determined by the specific application and the cohort being studied.

In terms of measurement repeatability, the only previously published work is reported in [6]. The most important texture values they selected in their final study were Teta1 and Perc.01, which were excluded from our study. We found that these autoregressive model texture features (and also the histogram texture features) had very poor repeatability. These differences may be influenced by several aspects, including different field strength of magnet (1.5 T vs 3.0 T), MRI protocol, RF coils, etc.

Potential physiological mechanisms underlying these observed changes across different conditions primarily include: (1) Myocardial wall thickening alters partial volume effects, while perfusion changes modify signal intensity distribution, and myofibre structural reorganization impacts spatial tissue uniformity. (2) Increased microstructural complexity/disorganization, combined with mildly anisotropic fiber arrangement, influences both texture orientation patterns and signal characteristics. Illustrative Example: The parameter S2-AVERAGE_SumAverg demonstrates consistent statistical significance when comparing different age groups at ED. As this variable quantifies the sum of average gray-level intensities within local myocardial regions, its discriminative power likely captures: early extracellular matrix reorganization preceding clinically detectable fibrosis, subtle lipid deposition patterns that alter local signal averages, and microstructural changes associated with normal myocardial aging.