https://orcid.org/0000-0003-2789-705X
Figures
Abstract
Background
The allelic variations of the apolipoprotein E (APOE) gene play a critical role in regulating lipid metabolism and significantly impact cardiovascular disease risk (CVD). This study aimed to evaluate the impact of exercise on cardiac structure and function in mouse models expressing different APOE genotypes using photon-counting computed tomography (PCCT) and deep learning-based segmentation.
Methods
A total of 140 mice were grouped based on APOE genotype (APOE2, APOE3, APOE4), sex, and exercise regimen. All mice were maintained on a controlled diet to isolate the effects of exercise. Low dose cardiac photon counting micro-CT imaging with intrinsic gating was performed using a custom-built micro-PCCT system and data was reconstructed with an iterative algorithm incorporating both temporal and spectral dimensions. A liposomal-iodine nanoparticle contrast agent was intravenously administered to uniformly opacify cardiovascular structures. Cardiac structures were segmented using a 3D U-Net deep learning model that was trained and validated on manually labeled data. Statistical analyses, including ANOVA, post-hoc analysis, and stratified group comparisons, were used to assess the effects of genotype, sex, and exercise on key cardiac metrics, including ejection fraction and cardiac index.
Results
The PCCT imaging pipeline provided high-resolution images with enhanced contrast between blood compartment and myocardium allowing for precise segmentation of cardiac features. Deep learning-based segmentation achieved high accuracy with an average Dice coefficient of 0.85. Exercise significantly improved cardiac performance, with ejection fraction increasing by up to 18% and cardiac index by 46% in exercised males, who generally benefited more from exercise. Females, particularly those with the APOE4 genotype, also showed improvements, with a 31% higher ejection fraction in exercised versus non-exercised mice. Stratified analyses confirmed that both sexes benefited from exercise, with males showing larger effect sizes. APOE3 and APOE4 genotypes derived the greatest benefit, while APOE2 mice showed no significant improvement.
Conclusions
This study demonstrates the utility of PCCT combined with deep learning segmentation in assessing the cardioprotective effects of exercise in APOE mouse models. These findings highlight the importance of genotype-specific approaches in understanding and potentially mitigating the impact of CVD through lifestyle interventions such as exercise.
Citation: Allphin AJ, Nadkarni R, Han ZY, Clark DP, Ghaghada KB, Badea A, et al. (2025) Assessing the cardioprotective effects of exercise in APOE mouse models using deep learning and photon-counting micro-CT. PLoS ONE 20(4): e0320892. https://doi.org/10.1371/journal.pone.0320892
Editor: Paul-Adrian Calburean, George Emil Palade University of Medicine Pharmacy Science and Technology of Targu Mures: Universitatea de Medicina Farmacie Stiinte si Tehnologie George Emil Palade din Targu Mures, ROMANIA
Received: October 4, 2024; Accepted: February 25, 2025; Published: April 10, 2025
Copyright: © 2025 Allphin et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.
Data Availability: Statistical data and sample code needed to recreate all segmentation results and are available at the public Gitlab repository: https://gitlab.oit.duke.edu/aja54/pcct-cardiac-pipeline.
Funding: All work was performed at the Quantitative Imaging and Analysis Lab and was supported by R01AG070149, R01 AG066184, RF1 AG057895. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.
Competing interests: The authors have declared that no competing interests exist.
Introduction
Cardiovascular disease (CVD) and Alzheimer’s disease (AD) risk are associated with aging, and modifiable lifestyle factors such as sedentary behavior and diets high in fats and sugars. These diseases are further influenced by genetic factors, notably the apolipoprotein E (APOE) gene, which plays a crucial role in lipid metabolism [1–3]. Understanding the interplay between genetic predisposition, environmental factors, and lifestyle choices such as exercise is critical for developing effective therapeutic and preventive interventions. Among the various imaging modalities, x-ray computed tomography (CT)-based quantitative imaging stands out for its ability to produce high-resolution, anatomical and functional multi-dimensional images, which are crucial for studying disease mechanisms and evaluating therapeutic interventions. In particular, photon-counting CT (PCCT) represents a significant advancement in imaging technology, offering superior image quality and enhanced quantitative capabilities compared to traditional CT methods. PCCT’s benefits include higher spatial resolution, improved tissue contrast, and reduced radiation dose, making it a promising tool for both preclinical and clinical applications [4]. Photon-counting CT has demonstrated promising results in cardiovascular imaging, particularly for detecting coronary calcifications [5], assessing stents [6], and evaluating myocardial perfusion [7]. The technology’s ability to reduce electronic noise and artifacts, combined with its multi-energy capabilities, allows for superior tissue characterization, making it highly suitable for complex cardiac imaging tasks.
Preclinical research, especially in mouse models, is an essential precursor and companion to clinical research because it allows precise control over genetic and environmental variables [8]. Our group has previously demonstrated the value of preclinical PCCT imaging in cancer and cardiac studies in mice, emphasizing its potential for detailed phenotypic characterization [9,10]. Despite these advances, the application of PCCT in preclinical cardiac imaging, particularly for evaluating the impact of exercise as a therapeutic intervention, remains underexplored and warrants further investigation.
Previous preclinical research has focused on quantitatively comparing the cardiac anatomy and function of different APOE mouse models using only 3D left ventricle segmentations [11,12]. This left ventricle segmentation has enabled calculation of key cardiac metrics such as stroke volume, ejection fraction, and cardiac output, providing a quantitative representation of cardiac health. The different APOE genotypes used in these studies served as key models for varying degrees of risk for Alzheimer’s disease and cardiovascular disease, offering valuable insights into the genetic predisposition and progression of these conditions [1–3].
In our previous study, we developed a cardiac photon-counting CT (PCCT) pipeline to phenotype APOE mouse models and investigated the effects of high fat diet on cardiac performance across different APOE genotypes [12]. In the current study, we expanded upon our previous work by incorporating additional cardiac metrics derived from a larger set of cardiac features and by introducing exercise as an experimental variable. While exercise is known to provide significant cardiovascular benefits, its effects across different APOE genotypes remain poorly understood. To address this, we employed a deep learning-based approach for full heart segmentation, which offers greater speed, reliability, and effectiveness compared to previously used atlas-based methods limited by their reliance on predefined anatomical templates and manual corrections, leading to slower processing times and potential inaccuracies [13]. By utilizing these advanced techniques, we aim to deliver a more detailed and comprehensive assessment of the benefits of exercise in APOE mouse models genetically predisposed to cardiovascular disease.
Methods
Cardiac imaging pipeline
Fig 1 illustrates the cardiac photon-counting CT (PCCT) pipeline developed in our previous work [12] and highlights the modifications introduced in the current study [12].
We have bold-faced the items that relate to new aspects unique to this work (compared to our previous study [12]).
The pipeline integrates advanced imaging and deep learning techniques to quantitatively assess cardiac function. The core imaging methods, including contrast-enhanced cine-PCCT, intrinsic cardiac gating, and multi-energy iterative reconstruction, were adapted from our previous work [12] to additionally evaluate exercise as a group factor.
Imaging was conducted using our custom-built micro-PCCT system [14] at the Duke Quantitative Imaging and Analysis Lab. Each of the imaging datasets was reconstructed using a 5D joint iterative reconstruction algorithm, incorporating regularization along both time and energy dimensions as described previously [10,12,15]. The resulting PCCT data were then used for deep learning-based cardiac segmentation and statistical analysis. Detailed explanations of these techniques are provided in the following sections.
Mouse models
All animal procedures were approved by the Duke Institutional Animal Care and Use Committee (IACUC, protocol registry number: A173-20–08). The study utilized a cohort of 140 mice bred and maintained at Duke University Medical Center’s facilities in the Bryan Research Building for Neurobiology. At the end of the study, the mice were euthanized using an intraperitoneal injection of 250 mg/Kg pentobarbital, as approved by our institution’s animal care and use committee. We ensured that all actions were carried out humanely, and with the utmost regard for the welfare of our animals. The experimental design incorporated variation in APOE genotype (APOE2, APOE3, APOE4), sex (male, female), and exercise regimen. We also used some APOE mouse models that possess a component of humanized innate immune system through the inclusion of the human (h)NOS2 gene as described in previous work [12]. The presence or absence of this humanized component is abbreviated as “HN factor” throughout this work. The average age of all animals was 14.5 months with a standard deviation of 3.3 months. All animals were maintained on the same normal diet to eliminate dietary effects, allowing us to isolate the impact of exercise. The exercised mice were provided with running wheels for 1 hour per day, 5 days a week, over a period of 3 months. Each mouse was housed individually during the designated time, ensuring unrestricted access to the wheel. Exercise activity was monitored using the Wheel Manager software (Med Associates Inc., St. Albans, VT, USA), and the distance run was recorded at the end of each session. Table 1 details the distribution of mice across these experimental variables, and Table 2 provides percentages of exercised mice, and HN factor within each genotype. While the distribution of animals across genotypes and sexes appears relatively balanced in terms of total numbers, there is some variability in exercise participation within each group, particularly when broken down by sex.
In Vivo PCCT imaging
Mice were imaged using our custom PCCT system, which features a Varian G297 x-ray tube and a Santis 1604 PCD (Dectris, Inc.), configured with 4 energy bins and a CdTe sensor [14]. The system provides a field of view of 12.9 cm x 4.3 cm and captures high-resolution images with a pixel size of 150 µm. During imaging, mice were anesthetized with 2–3% isoflurane delivered via a nosecone, and vital signs were monitored throughout the procedure using a pneumatic pillow for respiration and electrodes for ECG. To enhance blood-pool contrast, long-circulation liposomal iodine (Lip-I) nanoparticle contrast agent was administered via retro-orbital injection. A volume of 0.012 mL (equaling 1.2 mg of I [9,16]) was injected per gram of body weight for each mouse. Scanning parameters included an 80 kVp tube voltage, 4 mA tube current, and 10 ms/exposure, resulting in a total scan time of 70 seconds and an absorbed radiation dose of approximately 118 mGy. This dose is 55 to 76 times less than the lethal dose of 6.5–9 Gy, known as LD50/30 [17]. The PCD energy thresholds were set at 25, 34, 50, and 60 keV, with the second energy threshold positioned near the K-edge of iodine (33.2 keV) to optimize iodine contrast enhancement. As in our previous studies, a semi-automated intrinsic gating approach [12] was employed to sort the projections into 10 cardiac phases during the R-R interval of the cardiac cycle, enabling high-fidelity assessment of cardiac function.
5D CT reconstruction and material decomposition
Reconstruction of the PCCT data was performed using a multi-channel iterative algorithm with an isotropic voxel size of 125 µm. The reconstruction process involved joint regularization across both the temporal and spectral dimensions using rank-sparse kernel regression (RSKR) [15,18].
Reconstruction iteratively computed that is the solution to the following equation:
The reconstructed data (columns of X) at each energy (e) and time (t) minimizes the reprojection error (R representing the system projection matrix) relative to log-transformed projection data (Y). Projections are temporally selected based on the intrinsic gating procedure previously discussed. To reduce noise in the reconstructed results, the data regularizer () RSKR reduces bilateral total variation (BTV) jointly across the spectral dimension and patch-based rank across the time dimension.
Reconstruction resulted in 5D volumes with 4 energies, 10 cardiac phases, and 3 spatial dimensions with 125 µm isotropic voxels. Iterative reconstruction times ranged from 5–9 hours depending on the volume size.
Since our reconstructed volumes have 4 energies, we could use material decomposition to generate I, photoelectric effect (PE), and Compton scattering (CS) maps. Material decomposition was performed using a variation of the method described by Alvarez and Macovski [18,19].
In essence, material decomposition was performed via matrix inversion, solving the following linear system at each voxel:
In this equation, X is the reconstructed PCCT image vectorized as columns by energy, C represents the concentrations of our basis materials (e.g., I, PE and CS or I, Ca, and H2O) for each voxel, and M is a matrix of material sensitivities at each energy. An orthogonal subspace projection approach was used to prevent negative concentrations [18]. Post-decomposition, the material maps were assigned colors and combined in ImageJ for visualization.
These material maps were used to distinguish between blood (containing iodine contrast agent) and other structures such as soft tissue and bone, potentially facilitating a more precise analysis of cardiac features.
Deep learning-based cardiac segmentation
The goal of quantitative imaging is to extract useful, interpretable, and repeatable markers of biological performance. To quantitatively assess cardiac anatomy and function using micro-PCCT, we used segmentations of cardiac features at ventricular diastole and systole to calculate metrics such as ejection fraction and cardiac output. In this study, we expanded our previous approach which used segmentation of the left ventricle to further include segmentations of both ventricles, both atria, the myocardium, and several peripheral vessels, including the pulmonary artery and aorta. High-throughput imaging requires a robust and rapid segmentation approach, making deep learning an ideal choice. Consequently, we trained a deep learning-based segmentation network using an initial subset of user-labeled data.
These training labels were created using the seed-growing and painting tools in 3D Slicer [20]. Initially, seeds were placed in each of the cardiac structures, which were then grown to generate relatively accurate segmentations [21]. Finally, the labelers manually refined the segmentations using the 3D painting tools. A total of 46 sets of manual labels were created to train and validate the segmentation network. The creation of these labels was a collaborative effort among four researchers, each with at least two years of cardiac imaging experience. Involving multiple individuals in label creation helped increase throughput and mitigate the risk of the network learning a specific individual’s bias, given the user discretion involved in both seed placement and refinement.
For this segmentation task, we selected the 3D UNet CNN architecture [22], adapted from previous work [23]. We trained two separate networks using different input data types to assess potential differential benefits. For one network, the input was a cropped (128³ voxels) 3D view of the lowest energy threshold CT image. For the other network, the input was the same cropped region but from the decomposed iodine map. The 46 manual labels (comprising both diastole and systole of 23 mice) were used for training. These labels were randomly shuffled, independent of phase, and split into dedicated training (36 segmentations), validation (5 segmentations), and test (5 segmentations) sets. To maximize network generalizability and avoid overfitting, we employed multiple data augmentation strategies, including random cropping, random rotation, and random intensity shifts. We trained the models for 200 epochs at a learning rate of 0.001 using the Adam optimizer [24]. This training process took approximately two hours on a single RTX 5000 GPU. We used cross-entropy loss as the training cost function and evaluated network performance on the dedicated test set by calculating voxel-by-voxel accuracy, Jaccard index, Dice coefficient, precision, and recall. As shown later in the results section, due to its superior performance, the CT-input model was used for cardiac metric calculation and statistical analysis.
Derived cardiac metrics
Using the segmentations, we calculated the following key cardiac metrics: stroke volume (SV), ejection fraction (EF), cardiac output (CO), cardiac index (CI), and myocardial mass (MM). Equations 3–6 show how these metrics were calculated. The heart rate (HR), used in Equation 5, was measured during the intrinsic gating procedure and thus represents the anesthetized heart rate. The mouse body mass (m) used in Equation 6 was measured using a digital scale at the start of the day in which imaging occurred.
SV was calculated as the difference between end-diastolic volume (EDV) and end-systolic volume (ESV):
EF, representing the percentage of blood ejected from the left ventricle, was calculated as:
CO, representing the volume of blood pumped per minute, was calculated as:
CI which represents the weight-normalized version of CO, was calculated as:
MM was approximated using the average volume of the segmented myocardium across both the diastolic and systolic phases. This myocardial volume was converted to MM using an assumed tissue density of 1.053 g/mL [25].
By default, the EDV and ESV were derived from the segmentation of the left ventricle. We have also included some metrics that were calculated using the right ventricle segmentation. In these cases, we have added the prefix “RV” to indicate this distinction. For example, EF is the ejection fraction calculated from the left ventricle; RVEF is the ejection fraction calculated from the right ventricle.
Multi-factor statistical analysis
We first assessed the normality of each cardiac metric using the Shapiro-Wilk test [26]. Homogeneity of variance was checked using Levene’s test to evaluate whether variances were equal across groups [27]. Three metrics (mass, ejection fraction and myocardial mass) did not satisfy the normality test. Mass did not satisfy either test.
For metrics that passed the normality and homogeneity tests, we applied multi-factor ANOVA to evaluate the main effects and interactions of exercise, sex, genotype, and HN factor. More explicitly, we used linear models for each cardiac measure (e.g., Cardiac output ~ Genotype*Exercise*Sex*HN), where Genotype represents APOE2, APOE3, and APOE4, Exercise indicates exercised or non-exercised groups, and HN is 0 for mouse lines with mNos2 background and 1 for those with mNos2-/- hNOS. These results are presented both numerically and visually, providing a clear understanding of the genotype, exercise, sex, and HN factor interaction effects on cardiac performance. We applied the Benjamini-Hochberg procedure for False Discovery Rate (FDR) correction to adjust p-values for multiple comparisons [28]. The effect size was measured using eta-squared [29] to estimate the proportion of variance explained by each factor.
Metrics that failed the normality and/or homogeneity of variance tests were analyzed using other applicable methods. For mass, which failed both assumptions, we used the Kruskal-Wallis test [30] with Dunn’s post hoc test [31] for basic multifactor comparisons. For ejection fraction and myocardial mass, which failed only the normality assumption, we used generalized linear models (GLMs) for full analysis of factors and their interactions. Specifically for the metrics in this work, we employed GLMs with the Gamma distribution family and log link function, as recommended for skewed continuous data [32].
Stratified subgroup exercise comparison
We also performed a series of stratified group comparisons to examine the influence of exercise within specific subgroups, such as sex and genotype, using the Mann-Whitney U test for non-parametric comparisons. This approach was chosen to account for potential violations of normality in smaller subgroups. For example, we compared exercised and non-exercised male mice separately. We extended these stratified comparisons across combinations of sex and genotype to explore how exercise manifests differently in these subgroups. We once again applied FDR corrections to all p-values in these stratified group comparisons to reduce the likelihood of Type I errors. All analyses were performed using Python including the statsmodels and scipy packages. A significance threshold of 0.05 was used for all statistical tests.
Results
Segmentation results
Table 3 contains the segmentation performance metrics calculated on the test set. We have included calculations of accuracy, Jaccard index, Dice coefficient, precision, and recall for each anatomical segment as well as an average across all segments. For example, the average Dice coefficient across all segmented structures when trained using CT images was 0.85 which is considered acceptable for most medical images. We note that the segmentation performance for CT images is better than for I maps (average Dice: 0.72). This is most likely due to the higher noise in I images which leads to increased uncertainty along feature boundaries. We also note that the left ventricle has the lowest overlap scores (Dice and Jaccard) likely due to the ambiguity present in separating the aorta from the left ventricle. That region required the most discretion on the part of the label creator. Theoretically the network prediction represents an unbiased result, but further work would be required to verify that claim. Figure 2 shows a representative example indicating qualitative segmentation performance.
Within the images labelled “Label” and “Prediction,” the outline of each segmentation is shown as an overlay on either a CT image (upper half of figure) or a decomposed I image (bottom half of figure). The CT images are shown as attenuation maps with units of 1/voxel size while the I images represent concentrations with units of mg/mL. Each segmentation color represents a different anatomical region. We reiterate that the segmentations are voxelated 3D semantic segmentations; we show only the 2D outline of the segmentations in these images for visual simplicity.
Overview of cardiac metrics
Table 4 provides the average and standard deviation of the cardiac metrics, categorized by sex, genotype, and exercise factor. Values are presented as means with standard deviations in parentheses. The raw volumetric measurements for each heart chamber can be found in the supplemental S1 Table 1.
Table 4 offers a general overview of the sample measurements of key cardiac metrics within our cohort grouped by sex, genotype, and exercise. Most relevant to this work, we measured differences between the mean values of exercised mice and nonexercised mice within some sex and genotype groups. For example, exercised mean EF values were 31% higher in APOE3 males, 18% higher in APOE4 males, 16% higher in APOE3 females, and 31% higher in APOE4 females. Similarly, exercised mean SV values were 27% higher in APOE2 males, 44% in APOE3 males, and 65% higher in APOE4 males, but 14% in APOE3 females and 33% in APOE4 female. Mean CO values also improved with exercised CO values being 51% higher in APOE3 males and 75% higher in APOE4 males, compared to 24% in APOE3 females and 41% in APOE4 females. This suggests a possible sex- and genotype-specific improvement in the heart’s ability to pump blood.
Genotype-specific responses showed that APOE4 mice had the most pronounced improvements in cardiac function. Exercised mean CO values were 75% higher in APOE4 males and 30% higher in females from the same genotype highlighting a strong cardioprotective effect of exercise in this genotype. APOE2 mice exhibit more modest improvements with 27% higher mean SV values in males and 12% higher mean SV values females suggesting a weaker but still positive response to exercise.
Sex differences were apparent, with males generally having higher baseline CO and MM compared to females. Despite this, exercise improved cardiac efficiency, measured as mean CI, in both sexes. APOE4 males showed a 46% improvement in CI, while APOE4 females displayed a similar 44% improvement, demonstrating that the benefits of exercise are not exclusive to one sex.
Exercise had to have little to no impact on body mass or MM. These findings suggest that exercise predominantly enhanced cardiac function rather than inducing changes in size or structure.
To summarize, within our cohort, the clearest measured differences were in the cardiac performance metric between exercised and nonexercised mice. These differences were most pronounced in APOE4 females, APOE4 males, and APOE3 males. However, statistical analysis presented in the following section is required to understand the significance of these differences.
Multi-factor statistical results
Mass failed both the normality and homogeneity of variance assumptions and was therefore analyzed using the Kruskal-Wallis and Dunn’s tests. As shown in Fig 3, significant mass differences were observed in our cohort between genotype levels and between sexes but notably not between exercise groups.
The lines with asterisks indicate significant differences (p<0.05) identified by the Kruskal-Wallis test and (when needed) Dunn’s post hoc test. These plots coupled with other findings indicate that mass differences in our cohort can be attributed to sex and genotype but not to exercise.
EF and MM each failed the assumption of normality and were analyzed using GLMs. For EF, no main effects were found to be significant but there were significant interactions between genotype and exercise as well as sex and exercise. Specifically, APOE4 mice without exercise showed a significant (p=0.014) reduction in EF compared to baseline. Additionally, males without exercise also showed a significant (p=0.017) reduction in EF. For MM, sex was found to be a highly significant (p<0.0001) main effect with males showing an overall larger MM than females. Interestingly, the interaction term between sex and HN factor was significant (p=0.009), demonstrating a potential difference in how the human Nos2 gene effects male and female mice. Specifically, males with the HN factor showed a decrease in MM.
All other metrics outside of mass, EF, and MM passed the relevant assumptions and were analyzed as part of a multi-factor ANOVA. As shown in Table 5, exercise was a significant predictor with very large effect sizes for nearly all key cardiac metrics including SV, CO, CI, RVSV, and RVEF. Sex was also a significant predictor for almost all cardiac metrics; however, CI was an exception. This indicates that some sex differences, such as those for CO, may be attributed to sex differences in body mass rather than sex differences in exercise response. Genotype was shown to be a significant predictor with a small to medium effect size for SV and CI. The exercise-sex interaction was also significant with small to medium effect sizes for SV and CO.
Stratified subgroup exercise comparison results
Within our stratified analysis of exercise group differences, we focused only on EF, CI, and RVEF. EF is a normalized measure that accounts for baseline differences in heart size. CI is a normalized measure that accounts for differences in body mass. These normalized metrics were chosen to reduce the impact of the mass differences (Fig 3) which are not exercise related. Figs 4 and 5 contain key violin plots demonstrating the stratified group differences our study revealed between exercised and nonexercised mice. Table 6 gives a summary of all significant stratified group differences. Across the board, exercise has a marked positive impact on EF, CI, and RVEF. This improvement is especially pronounced in males, as indicated in Fig 4, where both EF and RVEF are significantly higher in exercised males compared to their non-exercised counterparts (p < 0.0007 for EF and RVEF, Cohen’s D of 1.16 and 1.08, respectively). The effect is also observed in females, but to a lesser degree, with significant improvement in CI (p = 0.0142, Cohen’s D of 0.89). Fig 5 highlights that the response to exercise is genotype dependent. APOE3 and APOE4 mice show significant improvements in EF, CI, and RVEF with exercise. In particular, the APOE4 group shows the most robust response, with EF increasing significantly (p = 0.0028, Cohen’s D of 1.27) and CI showing the largest effect size (Cohen’s D of 1.64). Conversely, the APOE2 group exhibits no significant improvements with exercise, indicating a less responsive phenotype to exercise interventions in this genotype. The large effect sizes (Cohen’s D ranging from 0.89 to 1.64) reported for different metrics and subgroups highlight the importance of exercise in enhancing cardiac function, particularly in the APOE3 and APOE4 genotypes and in males.
The bars with asterisks indicate statistically significant differences revealed by a Mann Whitney U test (P<0.05).
genotype. The bars with asterisks indicate statistically significant differences revealed by a Mann Whitney U test.
Discussion
Our results confirm that automated segmentation allows for fast and accurate quantitative measurements. Qualitative assessment of the segmentations provides added confidence in the accuracy of this work. The segmentation performance (as shown in Table 3) using CT images as input generally outperformed the results when using I maps. This confirms previous findings that segmentation with CT input offers higher accuracy and overlap metrics, such as the Dice and Jaccard coefficients, reflecting clearer tissue boundaries [12]. In contrast, segmentation with I maps as input shows a notable decline in these metrics, reflecting the challenge of segmenting images with enhanced noise or less distinct tissue boundaries. Precision remains high for both inputs, indicating that the model avoids false positives, but recall is somewhat lower, suggesting it misses some boundaries, especially with I maps.
The multi-factor statistical analyses highlight clear differences in cardiac performance in our cohort. For nearly all cardiac functional metrics, exercise was revealed to be a significant predictor with a large effect size (Table 5). This confirms the beneficial influence of exercise on cardiac performance across the cohort. In addition to exercise, sex and genotype were also significant factors, although their effects were smaller. Notably, the interaction between exercise and sex suggests that males benefit more from exercise, as supported by the stratified subgroup analysis. This interaction aligns with the stronger exercise-induced improvements in males observed for EF, CI, and RVEF, as shown in Figs 4 and 5.
Interestingly, the results reveal no significant impact of exercise on overall mass, as demonstrated by the Kruskal-Wallis and Dunn’s post hoc analyses in Fig 3. Mass differences in our cohort are driven primarily by sex and genotype, rather than by exercise. This reinforces the idea that exercise predominantly affects cardiac function rather than structural metrics like heart mass. The analysis also revealed an interaction between sex and the HN factor influencing right ventricular stroke volume, suggesting that the humanized innate immune component might modulate sex-specific responses to cardiovascular stimuli like exercise.
The stratified group comparisons showed that exercise significantly enhances cardiac metrics, including EF, CI, and RVEF. Male mice generally exhibit greater improvements than females (as seen in Fig 4), reflecting sex-specific responses likely driven by physiological factors such as hormonal differences, as males typically exhibit a greater capacity for exercise-induced cardiac hypertrophy and functional improvement. Furthermore, cardiac function of APOE2 genotypes was unaffected by exercise while both APOE3 and APOE4 mice showed a significant improvement with exercise (Fig 5). These findings suggest that the APOE4 genotype is particularly responsive to exercise, consistent with previous studies [33,34] showing that the increased disease risk in this genotype can be mitigated by exercise.
These findings underscore the importance of personalized approaches when considering exercise as a therapeutic intervention, especially for cardiovascular conditions. The variability in responses, as indicated by the standard deviations, suggests underlying biological mechanisms that may differ across individuals, warranting further exploration. Exercise stands out as a strong determinant of cardiac performance, particularly for male mice with APOE3 and APOE4 genotypes, reinforcing the potential of targeted exercise regimens for those with genetic predispositions. Differences between genotypes and sexes also provide practical insights for lifestyle interventions tailored to specific genetic backgrounds, particularly in males, where exercise consistently improves multiple cardiac metrics, while female benefits are genotype restricted. Still, within sex differences indicate that APOE4 females benefit more than APOE3 and APOE2 females.
The observed differences in exercise benefits between male and female mice can likely be explained by the distinct biological effects of sex hormones on cardiovascular function. Estrogen, which is more prevalent in females, offers cardioprotective effects such as improved endothelial function, anti-inflammatory properties, and enhanced lipid metabolism. This protective baseline could reduce the additional benefits of exercise, as female cardiovascular health is already supported by hormonal protection, particularly premenopausal females with higher estrogen levels [35,36]. In males, testosterone promotes cardiac hypertrophy and enhanced cardiovascular performance, which could amplify the effects of exercise. Additionally, APOE4 males typically exhibit a more pro-inflammatory and pro-atherogenic profile, which exercise helps to improve, aligning with previous findings on the cardiovascular risks associated with the APOE4 genotype [2]. These results are consistent with the broader understanding that sex-specific responses to exercise are complex and heavily influenced by genetic background.
In our previous study [12], we found that high-fat diet negatively affects cardiac performance, especially in APOE4 and APOE2 genotypes, leading to increased left ventricular volumes and reduced ejection fractions. In contrast, the current study demonstrates that exercise significantly enhances cardiac function, particularly in male APOE3 and APOE4 mice, suggesting that exercise serves as a robust cardioprotective intervention. The findings highlight the contrasting impacts of high-fat diet and exercise, underscoring the importance of lifestyle modifications in managing genetic predispositions to cardiovascular disease. Specifically, exercise not only mitigates the cardiovascular risks associated with APOE genotypes, but also offers a targeted intervention to improve cardiac function in susceptible populations. While exercise was evaluated independent of diet in this work, future work may include joint evaluation of diet and exercise.
Limitations
While this study demonstrates significant findings regarding the cardioprotective effects of exercise across APOE genotypes using photon-counting CT and deep learning, several limitations should be acknowledged:
- 1. Sample Size and Subgroup Variability: Although the study includes a relatively large cohort of 140 mice, the sample sizes for some subgroups (e.g., genotype and sex combinations) are limited, particularly for non-exercised mice. This may have reduced the statistical power of some analyses, especially for less responsive genotypes such as APOE2.
- 2. Cross-Sectional Design: This study examines the effects of exercise at a single time point. A longitudinal design would allow us to evaluate how cardiac function evolves with exercise over time, providing deeper insights into the dynamics of exercise-induced adaptations.
- 3. Potential Bias in Deep Learning Segmentation: Although the deep learning model achieved high accuracy and Dice coefficients, the training and validation datasets were based on semi-automatically labeled data from a subset of mice. This may introduce labeling biases that could influence the segmentation performance, especially in complex cardiac structures.
- 4. Absence of Functional Behavior Assessment: While cardiac metrics were rigorously quantified, the study did not include functional assessments such as behavioral or endurance testing to link cardiac improvements to broader physiological or performance outcomes.
- 5. No Exploration of Molecular Mechanisms: The study provides a phenotypic analysis of cardiac function but does not explore the molecular or cellular mechanisms underlying the observed genotype- and sex-specific responses to exercise. Investigating these mechanisms could offer valuable insights into the biological pathways affected by exercise.
- 6. Clinical Translation: While this study demonstrates the utility of photon-counting CT in preclinical research, direct translation of the findings to human studies may be challenging due to species-specific differences in cardiac physiology and APOE-related effects.
These limitations highlight areas for improvement in future studies, such as incorporating longitudinal designs, expanding subgroup sizes, and exploring molecular mechanisms. Despite these constraints, the study provides valuable insights into genotype- and sex-specific exercise responses and underscores the utility of advanced imaging technologies in preclinical cardiovascular research.
Conclusion
These findings underscore the complex interplay between APOE genotype, sex, and exercise in modulating cardiac function. The significant improvements in cardiac metrics due to exercise, particularly in male and female APOE3 and APOE4 mice, suggest that targeted lifestyle interventions could have differential benefits depending on sex and genetic background, potentially mitigating cardiovascular risks associated with specific APOE genotypes.
Supporting information
S1 Table 1. Supplemental table showing all cardiac chamber volumes. Averages and standard deviations for measured cardiac volumes grouped by sex, genotype, and exercise state. Standard deviations are shown in parentheses below each average value. From left to right, the measurements are of the diastolic left ventricle (DLV), systolic left ventricle (SLV), diastolic right ventricle (DRV), systolic right ventricle (SRV), diastolic left atrium (DLA), systolic left atrium (SLA), diastolic right atrium (DRA), and systolic left atrium (SRA).
https://doi.org/10.1371/journal.pone.0320892.s001
(DOCX)
References
- 1. Mahley RW. Apolipoprotein E: From cardiovascular disease to neurodegenerative disorders. J Mol Med (Berl). 2016;94(7):739–46. pmid:27277824
- 2. Mahley RW, Rall SC Jr. Apolipoprotein E: far more than a lipid transport protein. Annu Rev Genomics Hum Genet. 2000;1:507–37. pmid:11701639
- 3. Liu C-C, Liu C-C, Kanekiyo T, Xu H, Bu G. Apolipoprotein E and Alzheimer disease: risk, mechanisms and therapy. Nat Rev Neurol. 2013;9(2):106–18. pmid:23296339
- 4. Taguchi K, Iwanczyk JS. Vision 20/20: Single photon counting x-ray detectors in medical imaging. Med Phys. 2013;40(10):100901. pmid:24089889
- 5. Feldle P, Scheuber M, Grunz J-P, Heidenreich JF, Pannenbecker P, Nora C, et al. Virtual non-iodine photon-counting CT-angiography for aortic valve calcification scoring. Sci Rep. 2024;14(1):4724. pmid:38413684
- 6. Stein T, Taron J, Verloh N, Doppler M, Rau A, Hagar MT, et al. Photon-counting computed tomography of coronary and peripheral artery stents: a phantom study. Sci Rep. 2023;13(1):14806. pmid:37684412
- 7. Meloni A. Cardiovascular applications of photon-counting CT technology: A revolutionary new diagnostic step. J Cardiovasc Dev Dis. n.d.;10(9).
- 8. Rosenthal N, Brown S. The mouse ascending: perspectives for human-disease models. Nat Cell Biol. 2007;9(9):993–9. pmid:17762889
- 9. Badea CT, Clark DP, Holbrook M, Srivastava M, Mowery Y, Ghaghada KB. Functional imaging of tumor vasculature using iodine and gadolinium-based nanoparticle contrast agents: a comparison of spectral micro-CT using energy integrating and photon counting detectors. Phys Med Biol. 2019;64(6):065007. pmid:30708357
- 10. Clark DP, Holbrook M, Lee C-L, Badea CT. Photon-counting cine-cardiac CT in the mouse. PLoS One. 2019;14(9):e0218417. pmid:31536493
- 11. Liehn EA, Lupan A-M, Diaconu R, Ioana M, Streata I, Manole C, et al. Heart function assessment during aging in apolipoprotein E knock-out mice. Discoveries (Craiova). 2021;9(3):e136. pmid:34816004
- 12. Allphin AJ, Mahzarnia A, Clark DP, Qi Y, Han ZY, Bhandari P, et al. Advanced photon counting CT imaging pipeline for cardiac phenotyping of apolipoprotein E mouse models. PLoS One. 2023;18(10):e0291733. pmid:37796905
- 13. Clark D, Badea A, Liu Y, Johnson GA, Badea CT. Registration-based segmentation of murine 4D cardiac micro-CT data using symmetric normalization. Phys Med Biol. 2012;57(19):6125–45. pmid:22971564
- 14. Holbrook MD, Clark DP, Badea CT. Dual source hybrid spectral micro-CT using an energy-integrating and a photon-counting detector. Phys Med Biol. 2020;65(20):205012. pmid:32702686
- 15. Clark DP, Badea CT. MCR toolkit: A GPU-based toolkit for multi-channel reconstruction of preclinical and clinical x-ray CT data. Med Phys. 2023;50(8):4775–96. pmid:37285215
- 16. Annapragada AV, Hoffman E, Divekar A, Karathanasis E, Ghaghada KB. High-resolution CT vascular imaging using blood pool contrast agents. Methodist Debakey Cardiovasc J. 2012;8(1):18–22. pmid:22891106
- 17. Williams JP, Brown SL, Georges GE, Hauer-Jensen M, Hill RP, Huser AK, et al. Animal models for medical countermeasures to radiation exposure. Radiat Res. 2010;173(4):557–78. pmid:20334528
- 18. Clark DP, Badea CT. Hybrid spectral CT reconstruction. PLoS One. 2017;12(7):e0180324. pmid:28683124
- 19. Alvarez RE, Macovski A. Energy-selective reconstructions in X-ray computerized tomography. Phys Med Biol. 1976;21(5):733–44. pmid:967922
- 20. Fedorov A, Beichel R, Kalpathy-Cramer J, Finet J, Fillion-Robin J-C, Pujol S, et al. 3D Slicer as an image computing platform for the Quantitative Imaging Network. Magn Reson Imaging. 2012;30(9):1323–41. pmid:22770690
- 21. Pinter C, Lasso A, Fichtinger G. Polymorph segmentation representation for medical image computing. Comput Methods Programs Biomed. 2019;171:19–26. pmid:30902247
- 22.
Ronneberger O, Fischer P, Brox T. U-Net: Convolutional Networks for Biomedical Image Segmentation. Medical Image Computing and Computer-Assisted Intervention – MICCAI 2015. Cham: Springer International Publishing; 2015.
- 23. Falk T, Mai D, Bensch R, Çiçek Ö, Abdulkadir A, Marrakchi Y, et al. U-Net: deep learning for cell counting, detection, and morphometry. Nat Methods. 2019;16(1):67–70. pmid:30559429
- 24. Kingma DP. Adam: A method for stochastic optimization. arXiv preprint. 2014.
- 25. Vinnakota KC, Bassingthwaighte JB. Myocardial density and composition: a basis for calculating intracellular metabolite concentrations. Am J Physiol Heart Circ Physiol. 2004;286(5):H1742-9. pmid:14693681
- 26. Shapiro SS, Wilk MB. An analysis of variance test for normality (complete samples). Biometrika. 1965;52(3–4):591–611.
- 27. Zimmerman DW. A note on preliminary tests of equality of variances. Br J Math Stat Psychol. 2004;57(Pt 1):173–81. pmid:15171807
- 28. Benjamini Y, Hochberg Y. Controlling the False Discovery Rate: A Practical and Powerful Approach to Multiple Testing. Journal of the Royal Statistical Society Series B: Statistical Methodology. 1995;57(1):289–300.
- 29. Nakagawa S, Cuthill IC. Effect size, confidence interval and statistical significance: a practical guide for biologists. Biol Rev Camb Philos Soc. 2007;82(4):591–605. pmid:17944619
- 30. Kruskal WH, Wallis WA. Use of Ranks in One-Criterion Variance Analysis. Journal of the American Statistical Association. 1952;47(260):583–621.
- 31. Dunn OJ. Multiple Comparisons Using Rank Sums. Technometrics, 1964;6(3):241–252.
- 32. Ng V, Cribbie R. Using the gamma generalized linear model for modeling continuous, skewed and heteroscedastic outcomes in psychology. Current Psychology. 2017;36(1):1–10.
- 33. Tokgöz S, Claassen JAHR. Exercise as Potential Therapeutic Target to Modulate Alzheimer’s Disease Pathology in APOE ε4 Carriers: A Systematic Review. Cardiol Ther. 2021;10(1):67–88. pmid:33403644
- 34. Foley KE, Diemler CA, Hewes AA, Garceau DT, Sasner M, Howell GR. APOE ε4 and exercise interact in a sex-specific manner to modulate dementia risk factors. Alzheimers Dement (N Y). 2022;8(1):e12308. pmid:35783454
- 35. Ueda K, et al. Sex differences and regulatory actions of estrogen in cardiovascular system. Frontiers in Physiology. 2021;12:738218.
- 36.
Lundsgaard AM, Fritzen AM, Kiens B. Sex-Specific Effects on Exercise Metabolism. In Exercise Metabolism. McConell G, Editor. Cham: Springer International Publishing; 2022. p. 337–56.