Rubens Fazan

An Explainable Artificial Intelligence algorithm is used to generate explanations of the decisions of Machine Learning models trained to classify the severity of Obstructive Sleep Apnea based on Heart Rate Variability. First, a Random Forest classifier is used for Obstructive Sleep Apnea severity prediction and then the Local Rule-based Explanations method is applied to provide decisions interpretability. The method employs Genetic Algorithms to generate artificial examples, which are then used to train decision trees that approximate the local behavior of the black-box model. This approach enables the extraction of human-readable rules that explain individual predictions, offering valuable insights into the decision-making process of black-box models such as Random Forests. By identifying the most influential features and decision rules, our approach aims to aid medical professionals in understanding how Heart Rate Variability correlates with Obstructive Sleep Apnea severity and enhances trust in Machine Learning-driven diagnostic tools.

Introduction: Obstructive sleep apnea (OSA) is a prevalent sleep disorder with a high rate of undiagnosed patients, primarily due to the complexity of its diagnosis made by polysomnography (PSG). Considering the severe comorbidities associated with OSA, especially in the cardiovascular system, the development of early screening tools for this disease is imperative. Heart rate variability (HRV) is a simple and non-invasive approach used as a probe to evaluate cardiac autonomic modulation, with a variety of newly developed indices lacking studies with OSA patients.Objectives: We aimed to evaluate numerous HRV indices, derived from linear but mainly nonlinear indices, combined or not with oxygen saturation indices, for detecting the presence and severity of OSA using machine learning models.Methods: ECG waveforms were collected from 291 PSG recordings to calculate 34 HRV indices. Minimum oxygen saturation value during sleep (SatMin), the percentage of total sleep time the patient spent with oxygen saturation below 90% (T90), and patient anthropometric data were also considered as inputs to the models. The Apnea-Hypopnea Index (AHI) was used to categorize into severity classes of OSA (normal, mild, moderate, severe) to train multiclass or binary (normal-to-mild and moderate-to-severe) classification models, using the Random Forest (RF) algorithm. Since the OSA severity groups were unbalanced, we used the Synthetic Minority Over-sampling Technique (SMOTE) to oversample the minority classes.Results: Multiclass models achieved a mean area under the ROC curve (AUROC) of 0.92 and 0.86 in classifying normal individuals and severe OSA patients, respectively, when using all attributes. When the groups were dichotomized into normal-to-mild OSA vs. moderate-to-severe OSA, an AUROC of 0.83 was obtained. As revealed by RF, the importance of features indicates that all feature modalities (HRV, SpO₂, and anthropometric variables) contribute to the top 10 ranks.Conclusion: The present study demonstrates the feasibility of using classification models to detect the presence and severity of OSA using these indices. Our findings have the potential to contribute to the development of rapid screening tools aimed at assisting individuals affected by this condition, to expedite diagnosis and initiate timely treatment.

Obstructive sleep apnea (OSA) is one of the most common sleep disorders and affects nearly a billion people worldwide. Furthermore, it is estimated that many patients with OSA are underdiagnosed, which contributes to the development of comorbidities, such as cardiac autonomic imbalance, leading to high cardiac risk. Heart rate variability (HRV) is a non-invasive, widely used approach to evaluating neural control of the heart. This study evaluates the relationship between HRV indices and the presence and severity of OSA. We hypothesize that HRV, especially the nonlinear methods, can serve as an easy-to-collect marker for OSA early risk stratification. Polysomnography (PSG) exams of 157 patients were classified into four groups: OSA-free (N = 26), OSA-mild (N = 39), OSA-moderate (N = 37), and OSA-severe (N = 55). The electrocardiogram was extracted from the PSG recordings, and a 15-min beat-by-beat series of RR intervals were generated every hour during the first 6 h of sleep. Linear and nonlinear HRV approaches were employed to calculate 32 indices of HRV. Specifically, time- and frequency-domain, symbolic analysis, entropy measures, heart rate fragmentation, acceleration and deceleration capacities, asymmetry measures, and fractal analysis. Results with indices of sympathovagal balance provided support to reinforce previous knowledge that patients with OSA have sympathetic overactivity. Nonlinear indices showed that HRV dynamics of patients with OSA display a loss of physiologic complexity that could contribute to their higher risk of development of cardiovascular disease. Moreover, many HRV indices were found to be linked with clinical scores of PSG. Therefore, a complete set of HRV indices, especially the ones obtained by the nonlinear approaches, can bring valuable information about the presence and severity of OSA, suggesting that HRV can be helpful for in a quick diagnosis of OSA, and supporting early interventions that could potentially reduce the development of comorbidities.

Elastase-2 (ELA-2) is an angiotensin II-generating enzyme that participates in the cardiovascular system. ELA-2 is involved in hemodynamic and autonomic control and is upregulated in myocardial infarction and hypertension. The inhibition of angiotensin-converting enzyme (ACE) increased ELA-2 expression in the carotid arteries and heart of spontaneously hypertensive rats. In this study, we sought to investigate the role of ACE inhibition in hemodynamic and autonomic balance in elastase-2 knockout (ELA-2 KO) mice. Male ELA-2 KO and C57BL/6 mice were treated with the ACE inhibitor enalapril or saline for 10 days. After treatment, mice underwent surgery for cannulation of the femoral artery and arterial pressure recordings were made five days later in awake animals. The variability of systolic blood pressure (SBP) and pulse interval (PI) was evaluated in the time and frequency domain. Spontaneous baroreflex was assessed by the sequencing method. ACE inhibition caused a significant decrease in mean arterial pressure (117±2.2 vs 100±2.8 mmHg) and an increase in heart rate (570±32 vs 655±15 bpm) in ELA-2 KO mice. Despite a tendency towards reduction in the overall heart rate variability (standard deviation of successive values: 7.6±1.1 vs 4.7±0.6 ms, P=0.08), no changes were found in the root of the mean sum of squares or in the power of the high-frequency band. ACE inhibition did not change the spontaneous baroreflex indices (gain and baroreflex effectiveness index) in ELA-2 KO mice. Altogether, this data suggested that ACE played a role in the maintenance of hemodynamic function in ELA-2 KO mice.

We evaluated the effects of exercise training (ET) on the profile of mood states (POMS), heart rate variability, spontaneous baroreflex sensitivity (BRS), and sleep disturbance severity in patients with obstructive sleep apnea (OSA). Forty-four patients were randomized into 2 groups, 18 patients completed the untrained period and 16 patients completed the exercise training (ET). Beat-to-beat heart rate and blood pressure were simultaneously collected for 5 min at rest. Heart rate variability (RR interval) was assessed in time domain and frequency domain (FFT spectral analysis). BRS was analyzed with the sequence method, and POMS was analyzed across the 6 categories (tension, depression, hostility, vigor, fatigue, and confusion). ET consisted of 3 weekly sessions of aerobic exercise, local strengthening, and stretching exercises (72 sessions, achieved in 40±3.9 weeks). Baseline parameters were similar between groups. The comparisons between groups showed that the changes in apnea-hypopnea index, arousal index, and O2 desaturation in the exercise group were significantly greater than in the untrained group (P<0.05). The heart rate variability and BRS were significantly higher in the exercise group compared with the untrained group (P<0.05). ET increased peak oxygen uptake (P<0.05) and reduced POMS fatigue (P<0.05). A positive correlation (r=0.60, P<0.02) occurred between changes in the fatigue item and OSA severity. ET improved heart rate variability, BRS, fatigue, and sleep parameters in patients with OSA. These effects were associated with improved sleep parameters, fatigue, and cardiac autonomic modulation, with ET being a possible protective factor against the deleterious effects of hypoxia on these components in patients with OSA.

The sequence method is an important approach to assess the baroreflex function, mainly because it is based on the spontaneous fluctuations of beat-by-beat arterial pressure (for example, systolic arterial pressure or SAP) and pulse interval (PI). However, some studies revealed that the baroreflex effectiveness index (BEI), calculated through the sequence method, shows an intriguing oscillatory pattern as function of the delay between SAP and PI. It has been hypothesized that this pattern is related to the respiratory influence on SAP and/or PI variability, limiting the SAP ramps to 3 or 4 beats of length. In this study, this hypothesis was tested by assessing the sequence method using raw (original) and filtered series. Results were contrasted to the well-established transfer function, estimated between SAP and PI. Continuous arterial pressure recordings were obtained from healthy rats (N = 61) and beat-by-beat series of SAP and PI were generated. Low-pass (LP) and high-pass (HP) filtered series of SAP and PI were created by filtering the original series with a cutoff frequency of 0.8 Hz. Original series were analyzed by either the sequence method or cross-spectral analysis (transfer function) at low- (LF) and high- (HF) frequency bands, while filtered series were evaluated only by the sequence method. Baroreflex sensitivity (BRS) and BEI of original series, calculated by sequence method, was highly (85–90%) determined by HP series, with no significant association between original and LP series. A high correlation (>0.7) was found between the BRS estimated from original series (sequence method) and HF band (transfer function), as well as for LP series (sequence method) and LF band (transfer function). These findings confirmed the hypothesis that the sequence method quantifies only the high-frequency components of the baroreflex, neglecting the low-frequency influences, such as the Mayer waves. Therefore, we propose using both the original and LP filtered time series for a broader assessment of the baroreflex function using the sequence method.

Even though the coronary reperfusion process is the most important tool to preserve cardiac function, after myocardial infarction, reperfusion of acutely ischemic myocardium can induce injury. We aimed to evaluate the functional and molecular aspects 4 weeks after myocardial ischemia-reperfusion (IR) in rats. Male Wistar rats (N = 47) were subjected to myocardial IR by short-term (30 min) ligation and subsequent reperfusion of the left descending coronary artery. Control rats (N = 7) underwent the same surgical maneuver without coronary ligation. After 4 weeks, rats had their cardiac function examined by ventricular pressure recording under basal condition or pharmacological stress. Myocardial fibrosis and molecular mediators of IR injury (reactive oxygen species, tumor necrosis factor-alpha and matrix-metalloproteinase-2) were assessed as well. Most of the rats subjected to IR did not show macroscopic signs of infarct, while only 17% of these animals showed large myocardial infarction scars. Of note, all animals submitted to IR presented the functional and molecular parameters altered when compared with the control subjects. Cardiac function was attenuated in all animals submitted to IR, regardless the presence or size of macroscopic cardiac scars. Interstitial fibrosis, matrix-metalloproteinase-2 activity and the expression of tumor necrosis factor-alpha were higher in the myocardium of all IR rats as compared to the control subjects (p<0.05). Myocardium superoxide anion and hydrogen peroxide were increased in rats without or with mild cardiac scars. These results show that IR leads to myocardial injury in rats. Besides, even the animals with an apparent healthy myocardium (without infarct scar) presented cardiac dysfunction and molecular changes that may contribute to the development of heart failure over time.