Empagliflozin Protects Against Doxorubicin Cardiotoxicity: Integrative Assessment of Cardiac Kinetics and Electrophysiology Using Machine Learning in a Rat Model

Figures

• 

  Figure 1. System set-up for the anesthetic-surgical protocol and video recording of the rat hearts. (a) The procedure begins with endotracheal intubation of the rat through a tracheostomy. (b) A left lateral thoracotomy is performed to expose the heart. (c) A small thoracic retractor is used to improve visualization of the heart. The heart is carefully separated from the lung tissue to minimize movement during video recording. Subcutaneous electrodes are attached to monitor the ECG throughout the procedure. (d) A screenshot of the monitor displaying continuous ECG and oxygen saturation readings during a session. (e) Frame split view of real-time heart recording using OpenCV(v4.12.0) in Python (v3.12.0).

• 

  Figure 2. The monitoring system is composed of the following elements: (a) desktop module (laptop); (b) stereo vision camera; (c) session folders; (d) database; (e) ECG machine (CONTEC monitor).

• 

  Figure 3. Depth visualization. For visualization, the depth axis was inverted, causing the heart to be displayed as a smooth dome emerging from a flat plane. Lighter green tones indicate structures located farther from the camera, whereas darker blue tones indicate structures closer to the camera.

• 

  Figure 4. Isolated heart point cloud top-down view. (a) Control (group I); (b) EMPA (group II); (c) EMPA + DOX (group III); (d) DOX (group IV); (e) EMPA-pre + DOX (group V). Lighter green tones indicate structures located farther from the camera, whereas blue tones indicate structures closer to the camera.

• 

  Figure 5. Cannulation of the adult rat heart. (a) The rat heart was cannulated via the aorta and connected to a syringe attached to the cannula. (b) Schematic illustration depicting the cannula’s position within the aorta in relation to the aortic valve and the suture tie. The cannula should be inserted into the ascending aorta, taking care to avoid crossing the aortic valve. Created in BioRender (https://BioRender.com/ml08lou (accessed on 18 June 2026)).

• 

  Figure 6. Electrocardiographic comparisons across the experimental groups. (A) (a) Control group, with QRS, PR, QT, and QTc intervals and R- and T-wave amplitudes within normal limits. (b) The EMPA group showed an ECG pattern comparable to that of the control group. (c) EMPA + DOX group, in which EMPA attenuated DOX-induced prolongation of QRS, PR, and QTc intervals and largely preserved R- and T-wave amplitudes. (d) DOX group, displaying reduced HR, widening of QRS, PR, QT, and QTc intervals, and characteristic changes in R- and T-wave amplitudes typical of DOX-induced cardiotoxicity. (e) EMPA-pre + DOX group in which EMPA pretreatment similarly limited QRS, PR, QT, and QTc prolongation and preserved R- and T-wave amplitudes compared with the DOX group. Paper speed: 60 mm/s; amplitude calibration: 10 mm/mV. (B) Schematic representation of ECG waves, intervals, and segments. (C) ECG measurements for all groups.

• 

  Figure 7. Multi-parametric internal correlation between ECG indices and the point cloud cardiotoxicity score within the same experimental cohort. Each panel depicts the linear regression (Pearson’s r) between the computer-vision-derived cardiotoxicity score (X-axis, scale 1.0–10.0) and specific electrophysiological markers (Y-axis). Panel 1 displays QRS duration (ms), Panel 2 shows PR interval (ms), Panel 3 presents QT interval (ms), Panel 4 illustrates R-wave amplitude (R-Amp, mV), Panel 5 shows T-wave amplitude (T-Amp, mV), and Panel 6 displays Heart Rate (BPM). Each data point represents an individual subject across the five experimental groups. Statistical significance is indicated as p < 0.001 for all correlations (*** = p < 0.001).

• 

  Figure 8. Internal non-parametric Spearman rank-order correlations between the cohort-relative point cloud cardiotoxicity score and ECG parameters in the same experimental cohort. The multi-panel scatter plots show a monotonic relationship between the unsupervised optical severity score on the X-axis (graded 1.0–10.0) and electrophysiological metrics on the Y-axis. Panel 1 displays QRS duration (ms), Panel 2 shows PR interval (ms), Panel 3 presents QT interval (ms), Panel 4 illustrates R-wave amplitude (R-Amp, mV), Panel 5 shows T-wave amplitude (T-Amp, mV), and Panel 6 displays Heart Rate (BPM). All variable pairs analyzed are statistically significant at p < 0.001 (** = p < 0.01, *** = p < 0.001).

• 

  Figure 9. Internal validation of the cohort-relative point cloud cardiotoxicity score using the normalized composite ECG index from the same experimental cohort. To establish a direct 1:1 comparative baseline, specific ECG parameters (QRS, PR, QT, R-Amp, T-Amp, and BPM) were mathematically integrated and normalized into a 1.0-to-10.0 severity scale, matching the optical point cloud scoring system. Linear regression analysis (Pearson’s r = 0.954, p < 0.001) demonstrates a highly robust positive relationship between mechanical fragmentation and global electrophysiological decline; (*** = p < 0.001).

• 

  Figure 10. Internal non-parametric validation of the cohort-relative point cloud cardiotoxicity score using the normalized composite ECG index from the same experimental cohort. The scatter plot illustrates the Spearman rank-order correlation (p = 0.924, p < 0.0001) between the global ECG severity rank (X-axis) and the unsupervised spatial-kinematic score (Y-axis), both harmonized to a 1.0–10.0 scale. The high monotonic alignment supports the algorithm’s ability to stage the functional ranking across groups: from the healthy baseline (Control/EMPA), through the gradient of therapeutic efficacy (EMPA + DOX and EMPA-pre + DOX), to the severe AIC phenotype (DOX group); (*** = p < 0.001).

• 

  Figure 11. ECG-derived health scores (solid bars) vs. point cloud-derived scores (hatched bars) for all 18 subjects in the same experimental cohort. The unsupervised point cloud scoring assigned scores from 1.0 to 10.0. All four control hearts scored below 4.0. All five doxorubicin hearts (group IV) scored above 7.3. Groups III (EMPA + DOX) and V (EMPA-pre + DOX) occupied intermediate ranges, reflecting partial cardioprotection.

• 

  Figure 12. Representative phase-contrast images of rat cardiomyocytes from each experimental group. Cell morphology and overall structural integrity under the different treatment conditions.