AI-enhanced non-invasive monitoring of critical biochemical analytes in perioperative care: a review of signal processing and clinical applications

Abstract

Current perioperative monitoring of critical biochemical analytes predominantly relies on intermittent arterial blood gas analysis, which carries inherent risks of invasiveness and discontinuous data acquisition. Emerging evidence suggests that variations in electrocardiogram and photoplethysmogram waveforms may hold significant predictive value for detecting critical biochemical analytes changes. Advances in artificial intelligence analysis technologies have further accelerated the development of non-invasive monitoring tools, including real-time non-invasive blood glucose monitoring for diabetic patients and blood potassium monitoring for renal dialysis patients. This review highlights the urgent clinical need for non-invasive, continuous monitoring of the critical biochemical analytes during the perioperative period. It provides a comprehensive summary of current monitoring technologies and signals related to critical biochemical analytes, with a focus on their potential application in non-invasive blood gas monitoring. Based on existing evidence, key analytes such as serum potassium, serum calcium, lactate, and blood glucose have demonstrated robust research foundations and are the primary focus of this review. However, further clinical validation is urgently required to confirm their reliability and applicability in clinical settings. Integrating artificial intelligence with traditional monitoring systems has the potential to significantly enhance the precision, timeliness, and effectiveness of perioperative care. These findings suggest that AI-enhanced non-invasive monitoring could reduce unnecessary blood sampling while providing earlier detection of critical biochemical analytes disorder. Successful clinical translation requires standardized validation protocols and hardware-software co-development to address current limitations in measurement consistency and clinical workflow integration.

Introduction

The real-time monitoring and regulation of changes in vital signs and the internal environment during the perioperative period are essential to ensuring patient safety. Globally, approximately 300 million surgical procedures are performed annually (Meara et al., 2015), highlighting the critical need for effective peri-aesthesia monitoring systems. Monitoring critical biochemical analytes primarily involves the assessment of electrolytes (e.g. potassium, calcium), metabolic parameters (e.g. glucose, lactate), and acid-base status (e.g. pH, PCO₂). Disorders of the critical biochemical analytes are common, with over 50% of cases experiencing imbalances in blood potassium and calcium levels (Wang et al., 2018); 40% showing abnormalities in glucose and lactate levels, and more than 60% affected by acid-base disturbances (Dony et al., 2017). These disorders are closely associated with adverse patient outcomes. For example, hyperkalemia significantly increases early postoperative mortality in elderly patients undergoing orthopedic surgery (Norring-Agerskov et al., 2017). Similarly, perioperative hyperglycemia and hyperlactatemia are linked to a two- to three-fold higher risk of postoperative infections and kidney injury (He et al., 2023). Beyond short-term outcomes, these imbalances also impair long-term prognoses. For instance, metabolic acidosis during kidney transplantation has been linked to compromised renal function six months post-surgery (Tejchman et al., 2007). Consequently, effective peri-aesthesia monitoring and regulation of the critical biochemical analytes are crucial for improving patient safety and clinical outcomes.

Currently, perioperative monitoring currently relies heavily on intermittent and invasive blood gas analysis. This approach, which requires arterial blood samples and laboratory analysis, is often delayed by 20–60 minutes (Zi et al., 2023), even with in-room blood gas analyzers reducing this time to 3–15 minutes. Such delays can impair clinical decision-making and compromise patient safety. A study involving 860,000 surgical patients found that 10% of anesthesia-related deaths were related to inadequate monitoring, with 60% attributed to delays in detection and intervention (Arbous et al., 2001). Furthermore, invasive blood gas analysis carries inherent risks, particularly for pediatric patients, in whom arterial punctures are technically challenging and associated with high complication rates (Gleich et al., 2021).

Artificial intelligence (AI), driven by vast amounts of medical data, offers transformative potential in clinical analysis, diagnosis, and treatment. AI-powered medical devices, particularly those utilizing deep learning algorithms, have achieved significant advancements across various specialties, including cardiology, ophthalmology, respiratory medicine, and gastroenterology (Muehlematter et al., 2023). Despite these advancements, real-time, continuous, non-invasive perioperative monitoring of the critical biochemical analytes remains a significant unmet need. Addressing this challenge is pivotal for the evolution of perioperative monitoring and represents a cornerstone for advancing anesthesia management.

This review aims to explore recent advancements in non-invasive monitoring technologies for critical biochemical analytes, including glucose, lactate, and electrolytes, and evaluates the feasibility of non-invasive blood gas analysis in peri-anesthetic care. By addressing existing gaps, this review aimed to promote intelligent perioperative monitoring systems and enhance patient safety in surgical settings.

Methods

This review included studies focusing on critical biochemical analytes, including but not limited to blood glucose, lactate, and electrolytes (e.g. serum potassium, serum calcium), etc. A systematic search of PubMed was performed on Feb 12, 2026 (Supplementary 1), with the search restricted to human studies. Letters, commentaries, editorials, and case reports were not included. The initial search yielded 645 titles and abstracts, which were independently reviewed by two reviewers. From these, 143 full-text articles were selected for detailed evaluation. Additionally, informal searches were conducted to address specific topics, and reference lists from previous reviews were examined to identify relevant studies. No formal criteria beyond those stated were applied for the inclusion of studies. The primary data extracted included study population, study design, type of signal source, analysis techniques, and predictive performance. Furthermore, we reviewed monitoring technologies and devices approved by the Food and Drug Administration (FDA) and the China National Medical Products Administration (NMPA) that are relevant to critical biochemical analytes monitoring. This review was conducted in adherence to the applicable SANRA (Scale for the Assessment of Narrative Review Articles) guidelines.

Blood glucose monitoring

Hyperglycemia affects 80% of cardiac surgery patients, 20–40% of non-cardiac surgery patients, and 35% of patients with vascular disease (Long et al., 2019). Studies have shown that postoperative hyperglycemia significantly increases the risk of surgical site infections, reoperations, and mortality, regardless of diabetes status (Odds ratio [OR]: infection 2.0, reoperation 1.8, mortality 2.71) (Kwon et al., 2013). In extracorporeal circulation surgeries, glucose levels may need to be monitored at 10-minute intervals to facilitate early detection and timely intervention, improving outcomes (Kotagal et al., 2015). Finney’s study further underscored the importance of accurate and timely blood glucose monitoring by linking poor glucose control to increased mortality in critically ill patients (Finney, 2003).

Blood glucose monitoring methods can be categorized into invasive, minimally invasive, and non-invasive approaches based on the degree of tissue trauma involved. Invasive methods, such as capillary blood glucose testing via fingerstick, remain the most widely used in clinical practice. These methods are highly accurate but can cause discomfort and are impractical for frequent measurements. Minimally invasive techniques, such as glucose monitoring via interstitial fluid obtained through fine microneedles, reduce tissue damage compared to traditional invasive methods. However, these methods are susceptible to inaccuracies due to factors like sweat production and local environmental conditions (Kulcu et al., 2003). Driven by the growing demand for minimally invasive and portable monitoring devices, numerous advanced and well-established diagnostic technologies have been developed. Table 1 presents FDA-approved minimally invasive glucose monitoring devices that are commercially available (Tierney et al., 2001; Barassi et al., 2015; Christiansen et al., 2017; Saha et al., 2017; Christiansen et al., 2018; Garg et al., 2022; Karinka et al., 2022). The earliest non-invasive glucose monitoring device, the GlucoWatch Biographer, was introduced in 2002. Resembling a conventional wristwatch, it paved the way for further developments in glucose monitoring technology. The Eversense Continuous Glucose Monitoring System (CGMS), for example, employs a minimally invasive procedure to implant a glucose sensor beneath the skin. The sensor predicts glucose concentration by detecting differences in fluorescence intensities, achieving an error rate of ≤20% in 93% of the data. The Guardian Connect device, approved by the FDA in 2018, achieves a 91.8% error rate of ≤20% in sensor-detected glucose data. The FreeStyle Libre series, updated to its third generation and approved by the FDA in May 2022, is marketed as the world’s thinnest and most accurate continuous glucose monitoring (CGM) device. The third-generation FreeStyle Libre is 70% smaller than its predecessor and achieves 93.2% of predictions within a 20% margin of error. The most recent advancement, the Dexcom G7 Continuous Glucose Monitoring System, received FDA approval in April 2024. The Dexcom G7 sensor, placed on the arm, demonstrates the highest accuracy among currently available devices, with 95.3% of monitoring data exhibiting an error rate of ≤20% and the lowest mean absolute relative difference (MARD) of 8.2% (A MARD <15% is considered indicative of accuracy comparable to fingerstick blood glucose testing.). Despite their high accuracy and continuous monitoring capabilities, these devices remain inherently invasive. Sensor patches require periodic replacement, and frequent recalibration with fingertip blood glucose tests is often necessary to ensure reliability.

Various non-invasive methods for blood glucose monitoring have been developed, including biosensor systems and technologies utilizing optical, microwave, and optical-acoustic combinations. Among these, the OptiScanner^® 5000, approved by the FDA in 2020, employs mid-infrared spectral technology to achieve non-invasive monitoring. Other devices include the C8 MediSensors device, which uses Raman spectroscopy, the Gwave device, which utilizes radio frequency technology, and the Glucontrol GC 300, which relies on near-infrared (NIR) technology. Recently, Zhang et al. developed a non-invasive blood glucose monitoring method based on improved Raman spectroscopy, demonstrating excellent monitoring performance (Zhang et al., 2025). While promising, the accuracy of these devices requires further validation in clinical settings. A novel direction in glucose monitoring leverages AI techniques to predict blood glucoses using physiological data such as electrocardiogram (ECG) and photoplethysmogram (PPG) signals (Figure 1). PPG signals contain rich information, with changes in blood glucose concentrations affecting tissue absorption of specific light wavelengths and influencing blood viscosity. Similarly, ECG waveforms have been shown to correlate with blood glucose, with parameters such as heart rate (HR), heart rate variability (HRV), and segments like ST and QT demonstrating a 92% sensitivity for glucose prediction (Tobore et al., 2019) Recent advancements in AI, particularly the application of neural networks such as Convolutional Neural Networks (CNN) and Deep Neural Networks (DNN), have significantly enhanced the accuracy of non-invasive glucose monitoring. For instance, a study involving 21 participants collected 103 days of ECG and PPG signals. The proposed model achieved excellent monitoring performance with a root-mean-square error (RMSE) of 1.49 mmol/L, a mean absolute relative difference (MARD) of 13.42%, and Zone A+B coverage of 99.49% in tenfold cross-validation (Li et al., 2024). Table 2 provides an overview of recent studies exploring non-invasive blood glucose monitoring methods (Krishnan et al., 2020; Kownacka et al., 2018; Habbu et al., 2019; Joshi et al., 2020; Porumb et al., 2020; Tsai et al., 2020; Zhang et al., 2020; Deng et al., 2021; Lee and Lee, 2021; Li et al., 2021; Sempionatto et al., 2021; Nie et al., 2023; Li et al., 2024; Saha et al., 2024; Pors et al., 2025; Zhang et al., 2025; Zhang et al., 2026).

Figure 1

Tissue fluid glucose levels, such as those measured in interstitial fluid (ISF), often lag behind blood glucose. Consequently, most device manufacturers note that ISF glucose monitoring may not accurately reflect rapid changes in blood glucose. Beyond physiological signals like ECG and PPG, bodily fluids such as saliva, sweat, and tears are being explored as alternative mediums for glucose monitoring. However, the correlation between glucose levels in these fluids and blood glucose remains inconsistent (Patel et al., 2015). For example, experimental devices have been developed for saliva glucose estimation, such as one affixed to dried teeth (Tseng et al., 2018), and for tear glucose monitoring using contact lenses (Farandos et al., 2015). Saha developed a sweat-based glucose monitoring device demonstrating promising accuracy (MARD ~10.56%), but further clinical trials are needed to validate its efficacy (Saha et al., 2024). Unfortunately, these technologies are still in the experimental stages, primarily limited to animal studies, and significant challenges persist in ensuring sensor safety and continuous usability.

Despite advancements, non-invasive glucose monitoring methods have yet to find practical application in the operating room. Current FDA-approved minimally invasive glucose monitoring devices meet the requirements for continuous and dynamic perioperative monitoring. However, pre-market clinical trials for these devices have primarily focused on diabetic patients, often excluding those with severe comorbidities or visible skin lesions. Additional limitations include prolonged sensor calibration times (up to several hours) and high costs, which may hinder widespread adoption. ISF-based methods, while minimally invasive and broadly applicable, suffer from reduced accuracy and are highly susceptible to environmental factors. For instance, anesthesia medications that suppress glandular secretion and constant low temperatures in operating room can significantly interfere with measurements, leading to poor precision (Singh and Jha, 2019). Diagnostic research addressing these limitations remains scarce (Kulcu et al., 2003). Blood glucose monitoring methods leveraging AI offer significant potential. Assuming sufficient monitoring accuracy, AI-assisted approaches provide timeliness, continuous and dynamic measurement, non-invasiveness, and ease of deployment. Physiological signals such as ECG and PPG, combined with neural network models, can enable real-time and dynamic blood glucose prediction. These methods impose minimal additional burden on clinical workflows and patients while offering continuous monitoring capabilities. It is worth paying attention to in the future.

Lactate monitoring

For patients without evidence of tissue hypoperfusion, lactate monitoring can help identify other underlying factors, providing a more accurate assessment of the clinical condition. Changes in lactate clearance rates have been identified as independent predictors of mortality risk (De Wever et al., 2004). Consequently, continuous and dynamic lactate monitoring is essential. Existing non-invasive methods can be broadly categorized into two types: Fluid-based approaches and Signal-based approaches (Table 3) (Petropoulos et al., 2016; Lin et al., 2018; Mason et al., 2018; Huang et al., 2019; Budidha et al., 2020; Olaetxea et al., 2020; Yuzer et al., 2022; Huang et al., 2023; Rabost-Garcia et al., 2023; Zi et al., 2023).

Fluid-based monitoring approaches: Portable sweat sensors are capable of analyzing metabolites such as glucose, lactate, and electrolytes (Gao et al., 2016). Electrochemical sensors are commonly used to predict lactate levels in sweat. For example, Ruggeri et al. combined colorimetric methods with machine learning techniques to develop a predictive model with excellent performance (Ruggeri et al., 2023). Additionally, sensors have been designed to measure lactate levels in tears (Lin et al., 2018) and saliva (Petropoulos et al., 2016); although these methods lack sufficient diagnostic experimental data to support their clinical use.

Signal-based monitoring approaches: Near-infrared (NIR) spectroscopy has been investigated for its potential to measure lactate concentrations by evaluating light absorption. For instance, Karthik demonstrated a strong correlation between lactate levels and light absorption in sodium lactate solutions, simulating whole blood (Budidha et al., 2020). Similarly, Huang proposed that lactate concentrations might correlate with cardiopulmonary variables and exercise intensity, developing a machine-learning model with excellent fitting performance under low to moderate intensities (Yeragani et al., 1996). Studies also suggest that lactate levels impact myocardial electrophysiological activity, with Zeng employing deep neural networks to analyze twelve-lead ECG data, achieving an accuracy of 80.27% in predicting lactate concentrations (Zi et al., 2023).

Despite these advancements, there is a notable lack of research on non-invasive lactate monitoring in perioperative settings (Figure 1). Fluid-based methods, such as those relying on sweat, are primarily used in sports medicine. The correlation between blood lactate levels and lactate levels in bodily fluids remains inconsistent, with some studies suggesting a relationship (Karpova et al., 2020) while others do not (Klous et al., 2021). A few studies have directly focused on blood lactate concentration as the primary monitoring target, with Rabost directly investigating blood lactate levels, meeting the monitoring requirements for the peri-anesthetic period (Rabost-Garcia et al., 2023). However. factors such as operating room temperature, cholinergic receptor antagonist use, and variability in critical biochemical analytes further limit the utility of fluid-based methods in perioperative care. Signal-based techniques, such as NIR spectroscopy, face challenges related to interference from substances like hemoglobin, which affect absorbance. These limitations highlight the need for innovative approaches. Non-invasive lactate monitoring using neural networks based on ECG data holds promise. Current models show good predictive performance in laboratory settings, impose no additional burden on patients, and are easily integrated into routine ECG monitoring. As neural network algorithms and computational power improve, the accuracy and clinical applicability of these methods are expected to advance.

Electrolytes: potassium

Hyperkalemia is strongly associated with poor clinical outcomes and, in severe cases, can lead to arrhythmias, heart failure, or death. A 2019 study in Japan reported a 6.8% incidence of hyperkalemia in the general population (Kashihara et al., 2019), while studies in China estimate a prevalence of 2–3%, with a misdiagnosis rate as high as 89% (Lai et al., 2021). Despite its clinical significance, the frequency of serum potassium monitoring during the perioperative period remains critically insufficient.

Conventional potassium monitoring relies on invasive blood sampling, which is limited by its inability to provide continuous, dynamic assessments. A portable, continuous, minimally invasive potassium monitoring device (CKM™) has been developed, but robust clinical validation is still lacking. Initial research has explored non-invasive potassium monitoring using PPG for classification (Miller et al., 2023), though most techniques rely on ECG. For example, potassium concentration is mainly related to the amplitude of the T wave, while calcium concentration correlates with the QT interval. This relationship was recognized early and utilized for empirical estimation of potassium levels in patients. However, empirical predictions are highly inaccurate. Approximately 81% of hyperkalemic patients’ ECGs are not correctly interpreted by clinicians (Rafique et al., 2020). One study revealed that the proportion of correctly identified hyperkalemic ECGs was only 34%-43%. Additionally, some normokalemic individuals exhibit ECG features resembling hyperkalemia, and hyperkalemic patients may present atypical ECG patterns, further complicating diagnosis. To avoid these limitations, mathematical models incorporating ECG features—such as T wave height, width, and slope—were developed (Attia et al., 2016), achieving sensitivity and specificity above 80%. However, these models often overemphasize local ECG features, introducing selection bias and reducing predictive accuracy at extreme potassium concentrations (Corsi et al., 2017). Enhancing model parameters, such as applying periodic component analysis, has been shown to improve prediction accuracy, although this significantly increases the computational workload associated with traditional mathematical data analysis methods.

Neural network algorithms offer a promising alternative to address these challenges (Figure 1). These models can process vast datasets and automatically identify critical ECG features for potassium prediction, mitigating issues of selection bias. Bachmann’s study demonstrated the superiority of DNN models over traditional computational models, highlighting their ability to optimize predictions across varying electrolyte concentrations (von Bachmann et al., 2024).

Table 4 provides an overview of techniques for monitoring potassium and calcium disorders, including non-invasive methods for serum potassium assessment (Corsi et al., 2012; Severi et al., 2009; Corsi et al., 2012; Attia et al., 2016; Corsi et al., 2017; Velagapudi et al., 2017; Lin et al., 2020; Regolisti et al., 2020; Kwon et al., 2021; Wang et al., 2021; Kim et al., 2023; Miller et al., 2023; Harmon et al., 2024; von Bachmann et al., 2024; Shang et al., 2025). Early approaches relied on simple mathematical models with limited accuracy. Velagapudi’s study achieved a sensitivity of only 63% using a basic mathematical model (Velagapudi et al., 2017). In contrast, Corsi and colleagues developed models for quantitative analysis of serum potassium levels, achieving a mean absolute deviation of approximately 0.5 mmol/L (Corsi et al., 2012). Compared to these methods, potassium monitoring using neural network-assisted techniques has significantly improved predictive performance. For instance, in Harmon’s study, the AI-ECG method demonstrated sensitivity and specificity exceeding 80%, with a negative predictive value surpassing 99%, highlighting its utility as a preliminary screening tool for electrolyte imbalances (Harmon et al., 2024). Similarly, Attia’s study used single-lead ECG for potassium prediction, achieving a prediction error within 10% of actual values (Attia et al., 2016). These advancements position AI-ECG as a promising technology for clinical applications. While ECG-based potassium level monitoring has already shown potential in managing dialysis patients, its adaptation for perioperative care has yet to be confirmed through dedicated research. Neural network-assisted potassium monitoring using ECG data offers several advantages: 1) Non-invasive and continuous monitoring: Eliminates the need for invasive blood sampling while providing dynamic, real-time assessments; 2) Convenience and cost-effectiveness: ECG monitoring is widely available and requires minimal additional resources. 3) Broad applicability: The flexibility of this approach allows for innovative applications, such as potassium prediction using ECG images captured via smartphones (Kim et al., 2023). Further research is needed to validate these techniques in perioperative settings and explore their full potential in broader clinical applications.

Electrolytes: calcium

Calcium ions play a vital role in the body, regulating cellular processes through kinase cascade reactions, enzyme activation, and other signaling pathways. Mild or gradual changes in calcium levels are often asymptomatic, with significant gastrointestinal or neuropsychiatric symptoms manifesting only when imbalances are severe or rapid. Both hypercalcemia and hypocalcemia are associated with poor clinical outcomes. Studies indicate that hypocalcemia increases mortality rates threefold in patients with pulmonary embolism (Murthi et al., 2021). Additionally, early calcium monitoring and intervention in trauma patients can significantly reduce transfusion requirements and the incidence of trauma-induced coagulopathy (De Robertis et al., 2015).

Research on non-invasive calcium monitoring remains limited (Figure 1). Recent advancements in ECG analysis, supported by deep learning technologies, have provided a new avenue for non-invasive monitoring. Calcium ion concentration affects the ECG by altering action potential duration and the QT interval. However, the performance of deep learning models for predicting blood calcium levels has been inconsistent (Table 4). For example, in Kwon’s study, the specificity for hypercalcemia prediction was only 52% (Kim et al., 2023). This limitation arises due to the highly concentrated normal range of calcium ions in healthy individuals. During model training, minimal fluctuations in calcium ion levels result in negligible ECG changes, leading to suboptimal fitting performance (Corsi et al., 2012). Additionally, the complex physiological mechanisms involved in calcium regulation may further limit the accuracy of these models.

While current ECG-based deep learning models for blood calcium prediction may currently be better suited as initial screening tools rather than definitive diagnostic instruments, they hold potential for specific applications. Continuous and dynamic monitoring of calcium levels could be particularly valuable in improving outcomes for high-risk populations, such as patients with severe trauma, hyperparathyroidism, or endocrine tumors. Future research should focus on optimizing model parameters, integrating multimodal data sources, such as by employing hybrid models combining ECG and Fluid-based signals and conducting robust clinical validation to enhance the utility of non-invasive calcium monitoring. These efforts could pave the way for more reliable and accessible tools for early detection and management of calcium imbalances.

Limitation

This review assessed several emerging technologies for non-invasive blood gas analysis and identified deep learning–assisted monitoring based on ECG or PPG signals as the most promising direction. However, given the rapid evolution of deep learning technologies, the findings and discussions presented here may soon require updating. A notable limitation in current non-invasive monitoring research is the inconsistent reporting of performance metrics. Although Clarke and Parkes Error Grid Analyses and MARD are the accepted standards for assessing clinical accuracy, reliance on binary metrics (e.g., sensitivity and specificity) persists in some studies. Such metrics are prone to boundary effects and may increase the susceptibility to reporting bias. Currently, a consensus on specific performance metrics for intraoperative monitoring technologies is lacking. Nevertheless, drawing from recent FDA “Breakthrough Device Designations”, we observe that an AUC around 0.9 and a MARD <10% serve as de facto references (U.S. Food and Drug Administration, 2018a; U.S. Food and Drug Administration, 2018b; U.S. Food and Drug Administration, 2021; U.S. Food and Drug Administration, 2022; U.S. Food and Drug Administration, 2024). In high-stakes environments like the perioperative period, sensitivity demands are typically even higher (>90%). The formulation of dedicated standards for these scenarios remains a crucial objective for future studies. Furthermore, some studies included in this review suffer from small sample sizes, making them highly susceptible to overfitting—a challenge particularly pronounced in models utilizing deep learning technologies (Miotto et al., 2018). Because deep learning networks process physiological signals with complex temporal characteristics, calculating a traditional ‘minimum sample size’ as done in conventional clinical research is often impractical. Therefore, solutions must be implemented at the algorithmic level. For models with limited training data, transfer learning can be employed to mitigate the risk of overfitting. Additionally, data augmentation techniques—such as adding noise, introducing baseline wander to the original signals, or utilizing Generative Adversarial Networks (GANs)—can be strategically used to expand the training set (Iwana and Uchida, 2021). During the model training phase, strategies like dropout and early stopping are also highly effective in preventing overfitting and enhancing model generalization. Importantly, due to their inherent ‘black-box’ nature, current deep learning models for non-invasive monitoring often lack clinical interpretability. To facilitate the clinical translation of these technologies and earn the trust of physicians, future studies must prioritize Explainable AI (XAI) techniques (Amann et al., 2020). For instance, Class Activation Mapping (e.g., Grad-CAM and Saliency Maps) can be utilized to generate heatmaps, verifying whether the model’s predictive focus aligns with established medical knowledge—such as explicitly highlighting peaked T-waves on an ECG when predicting hyperkalemia (Kwon et al., 2021). Additionally, SHapley Additive exPlanations (SHAP) can be implemented to quantify the specific impact and contribution of various input parameters on the final prediction, thereby making the decision-making process transparent. In addition, this review primarily focused on metabolic indicators (e.g., glucose and lactate) and electrolytes (e.g., potassium and calcium) based on currently available evidence. Other key internal environmental parameters—such as pH, bicarbonate, and base excess—remain insufficiently studied due to limited data. Future research should prioritize these underrepresented parameters, starting with those that are already well characterized and progressively expanding to a broader spectrum of internal environment components. Most of the studies included in this review evaluated non-invasive monitoring performance within specific populations and under clinical conditions.

Existing research on non-invasive monitoring is predominantly segmented: glucose for daily diabetes management, lactate for athletic performance, and potassium for routine clinical care. Few studies target the perioperative phase, where unique confounders exist. Anesthesia, surgical stress, and blood loss can significantly impact cardiac electrophysiology and peripheral perfusion, challenging the signal integrity of ECG and PPG. Similarly, sweat-based lactate sensing is vulnerable to hypothermia and hemodynamic instability. It remains to be proven whether the performance of these technologies in non-surgical settings can be successfully extrapolated to the complex perioperative context. In future clinical practice, it will be essential to account for regional and ethnic differences, variations in baseline health status, differences in data-acquisition devices, and the influence of complex surgical procedures on monitoring accuracy and reliability. By addressing these gaps, the development of comprehensive, real-time, non-invasive blood gas monitoring systems can significantly enhance patient care and clinical outcomes.

Conclusion

This review highlights four critical biochemical parameters—glucose, lactate, potassium, and calcium—by examining current non-invasive monitoring alternatives, comparing the advantages and limitations of various methods, and evaluating their suitability for the perioperative environment. While technologies such as micro-needling, ISF monitoring, and Raman spectroscopy have demonstrated rapid advancements, non-invasive monitoring of critical biochemical analytes using physiological signals from the body surface, such as ECG and PPG, shows significant promise for perioperative applications due to its unique advantages in this context. Future efforts should aim to expand the range of monitored parameters by extracting additional information from ECG and PPG signals, maximizing the capabilities of neural networks in processing large datasets, and adopting multimodal monitoring approaches. This could enable cross-validation and supplementation of key parameters, including those not directly measurable (e.g., pH, sodium levels), to create a more comprehensive monitoring system. However, this approach is not without limitations. High-quality data are essential for effective model training, and significant variability exists in the predictive performance of different models. Challenges also include the complexity of interpreting results, the potential for overfitting, and the need for robust validation in diverse clinical settings. Furthermore, among all FDA-approved AI medical devices, the majority focus on image analysis, highlighting the relative underdevelopment of AI applications in continuous physiological monitoring. Despite these challenges, advancements in peri-anesthetic digital intelligence hold immense potential to transform the field. Continuous, real-time non-invasive monitoring of internal environments could greatly enhance patient safety, minimize the risks associated with invasive procedures, and elevate the standard of perioperative care. By addressing the current limitations and fostering innovation in AI-driven technologies, the future of perioperative monitoring is poised to become more accurate, efficient, and patient-centered.

References

• 1

  AmannJ.BlasimmeA.VayenaE.FreyD.MadaiV. I.Precise4Q consortium (2020). Explainability for artificial intelligence in healthcare: A multidisciplinary perspective. BMC Med. Inform Decis Mak.20, 310. doi: 10.1186/s12911-020-01332-6

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 2

  ArbousM.GrobbeeD.KleefJ. (2001). Mortality associated with anaesthesia: A qualitative analysis to identify risk factors. Anaesthesia56, 1141–1153. doi: 10.1111/j.1365-2044.2001.02051.x

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 3

  AttiaZ. I.DeSimoneC. V.DillonJ. J.SapirY.SomersV. K.DuganJ. L.et al. (2016). Novel bloodless potassium determination using a signal-processed single-lead ECG. JAHA5, e002746. doi: 10.1161/JAHA.115.002746

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 4

  BarassiA.UmbrelloM.GhilardiF.DameleC. A.L.MassaccesiL.IapichinoG.et al. (2015). Evaluation of the performance of a new OptiScanner™ 5000 system for an intermittent glucose monitoring. Clinica Chimica Acta438, 252–254. doi: 10.1016/j.cca.2014.09.008

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 5

  BudidhaK.MamoueiM.BaishyaN.QassemM.VadgamaP.KyriacouP. (2020). Identification and quantitative determination of lactate using optical spectroscopy-towards a noninvasive tool for early recognition of sepsis. Sens (basel Switz.20, 5402. doi: 10.3390/s20185402

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 6

  ChristiansenM. P.GargS. K.BrazgR.BodeB. W.BaileyT. S.SloverR. H.et al. (2017). Accuracy of a fourth-generation subcutaneous continuous glucose sensor. Diabetes Technol. Ther.19, 446–456. doi: 10.1089/dia.2017.0087

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 7

  ChristiansenM. P.KlaffL. J.BrazgR.ChangA. R.LevyC. J.LamD.et al. (2018). A prospective multicenter evaluation of the accuracy of a novel implanted continuous glucose sensor: PRECISE II. Diabetes Technol. Ther.20, 197–206. doi: 10.1089/dia.2017.0142

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 8

  CorsiC.BieJ.MortaraD.SeveriS. (2012) Innovative solutions in health monitoring at home: The real-time assessment of serum potassium concentration from ECG. IMPACT Anal. Solut Chronic Dis. Prev. Manage., 116–123.

  • Google Scholar

• 9

  CorsiC.BieJ.NapolitanoC.PrioriS.MortaraD.SeveriS. (2012). Validation of a novel method for non-invasive blood potassium quantification from the ECG2012, 105–108.

  • Google Scholar

• 10

  CorsiC.CortesiM.CallisesiG.De BieJ.NapolitanoC.SantoroA.et al. (2017). Noninvasive quantification of blood potassium concentration from ECG in hemodialysis patients. Sci. Rep.7, 42492. doi: 10.1038/srep42492

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 11

  DengH.ZhangL.XieY.MoS. (2021). Research on estimation of blood glucose based on PPG and deep neural networks. IOP Conf Ser: Earth Environ. Sci.693, 12046. doi: 10.1088/1755-1315/693/1/012046

  • CrossRef
  • Google Scholar

• 12

  De RobertisE.Kozek-LangeneckerS. A.TufanoR.RomanoG. M.PiazzaO.Zito MarinosciG. (2015). Coagulopathy induced by acidosis, hypothermia and hypocalcaemia in severe bleeding. Minerva Anestesiol.81, 65–75.

  • Google Scholar

• 13

  De WeverO.NguyenQ.van HoordeL. (2004). Tenascin-C and SF/HGF produced by myofibroblasts in vitro provide convergent pro-invasive signals to human colon cancer cells through RhoA and rac. FASEB J : Publ Fed Am. Soc. Exp. Biol.18, 1016–1018. doi: 10.1096/fj.03-1110fje

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 14

  DonyP.DramaixM.BoogaertsJ. (2017). Hypocapnia measured by end-tidal carbon dioxide tension during anesthesia is associated with increased 30-day mortality rate. J. Clin. Anesth.36, 123–126. doi: 10.1016/j.jclinane.2016.10.028

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 15

  FarandosN.YetisenA.MonteiroM.LoweC.YunS. (2015). Contact lens sensors in ocular diagnostics. Adv. Healthc Mater.4, 792–810. doi: 10.1002/adhm.201400504

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 16

  FinneyS. J. (2003). Glucose control and mortality in critically ill patients. JAMA290, 2041. doi: 10.1001/jama.290.15.2041

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 17

  GaoW.EmaminejadS.NyeinH. Y. Y.ChallaS.ChenK.PeckA.et al. (2016). Fully integrated wearable sensor arrays for multiplexed in situ perspiration analysis. Nature529, 509–514. doi: 10.1038/nature16521

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 18

  GargS. K.KipnesM.CastorinoK.BaileyT. S.AkturkH. K.WelshJ. B.et al. (2022). Accuracy and safety of dexcom G7 continuous glucose monitoring in adults with diabetes. Diabetes Technol. Ther.24, 373–380. doi: 10.1089/dia.2022.0011

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 19

  GleichS. J.WongA. V.HandlogtenK. S.ThumD. E.NemergutM. E. (2021). Major short-term complications of arterial cannulation for monitoring in children. Anesthesiology134, 26–34. doi: 10.1097/ALN.0000000000003594

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 20

  HabbuS. K.JoshiS.DaleM.GhongadeR. B. (2019). “ Noninvasive blood glucose estimation using pulse based cepstral coefficients,” in 2019 2nd International Conference on Signal Processing and Information Security (ICSPIS). Dubai: IEEE, 1–4. doi: 10.1109/ICSPIS48135.2019.9045897

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 21

  HarmonD. M.LiuK.DuganJ.JentzerJ. C.AttiaZ. I.FriedmanP. A.et al. (2024). Validation of noninvasive detection of hyperkalemia by artificial intelligence–enhanced electrocardiography in high acuity settings. CJASN19, 952–958. doi: 10.2215/CJN.0000000000000483

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 22

  HeQ.TanZ.ChenD.CaiS.ZhouL. (2023). Association between intraoperative hyperglycemia/hyperlactatemia and acute kidney injury following on-pump cardiac surgery: A retrospective cohort study. Front. Cardiovasc. Med.10, 1218127. doi: 10.3389/fcvm.2023.1218127

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 23

  HuangS.CasaburiR.LiaoM. (2019). Noninvasive prediction of blood lactate through a machine learning-based approach. Sci. Rep.9, 2180. doi: 10.1038/s41598-019-38698-1

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 24

  HuangS.LeeC.HsuC. (2023). Prediction for blood lactate during exercise using an artificial intelligence-enabled electrocardiogram: A feasibility study. Front. Physiol.14, 1253598. doi: 10.3389/fphys.2023.1253598

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 25

  IwanaB. K.UchidaS. (2021). An empirical survey of data augmentation for time series classification with neural networks. PloS One16, e0254841. doi: 10.1371/journal.pone.0254841

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 26

  JoshiA. M.JainP.MohantyS. P.AgrawalN. (2020). iGLU 2.0: A new wearable for accurate non-invasive continuous serum glucose measurement in IoMT framework. IEEE Trans. Consumer Electron.66, 327–335. doi: 10.1109/TCE.2020.3011966

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 27

  KarinkaS. A.BrazgR. L.CastorinoK. N.LiljenquistD. R.KipnesM.LiuH. (2022). 76-LB: performance of freeStyle libre 3 system. Diabetes71, 76–LB. doi: 10.2337/db22-76-LB

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 28

  KarpovaE. V.LaptevA. I.AndreevE. A.KaryakinaE. E.KaryakinA. A. (2020). Relationship between sweat and blood lactate levels during exhaustive physical exercise. ChemElectroChem7, 191–194. doi: 10.1002/celc.201901703

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 29

  KashiharaN.KohsakaS.KandaE.OkamiS.YajimaT. (2019). Hyperkalemia in real-world patients under continuous medical care in Japan. Kidney Int. Rep.4, 1248–1260. doi: 10.1016/j.ekir.2019.05.018

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 30

  KimD.JeongJ.KimJ.ChoY.ParkI.LeeS.-M.et al. (2023). Hyperkalemia detection in emergency departments using initial ECGs: A smartphone AI ECG analyzer vs. Board-certified physicians. J. Korean Med. Sci.38, e322. doi: 10.3346/jkms.2023.38.e322

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 31

  KlousL.De RuiterC. J.ScherrerS.GerrettN.DaanenH. A. M. (2021). The (in)dependency of blood and sweat sodium, chloride, potassium, ammonia, lactate and glucose concentrations during submaximal exercise. Eur. J. Appl. Physiol.121, 803–816. doi: 10.1007/s00421-020-04562-8

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 32

  KotagalM.SymonsR. G.HirschI. B.UmpierrezG. E.DellingerE. P.FarrokhiE. T.et al. (2015). Perioperative hyperglycemia and risk of adverse events among patients with and without diabetes. Ann. Surg.261, 97–103. doi: 10.1097/SLA.0000000000000688

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 33

  KownackaA. E.VegelyteD.JoosseM.AntonN.ToebesB. J.LaukoJ.et al. (2018). Clinical evidence for use of a noninvasive biosensor for tear glucose as an alternative to painful finger-prick for diabetes management utilizing a biopolymer coating. Biomacromolecules19, 4504–4511. doi: 10.1021/acs.biomac.8b01429

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 34

  KrishnanS.VinuprithaP.KathirveluD. (2020). “ Non-invasive glucose monitoring using machine learning,” in 2020 International Conference on Communication and Signal Processing (ICCSP, Vol. 2020. 780–783. doi: 10.1109/iccsp48568.2020.9182434

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 35

  KulcuE.TamadaJ. A.ReachG.PottsR. O.LeshoM. J. (2003). Physiological differences between interstitial glucose and blood glucose measured in human subjects. Diabetes Care26, 2405–2409. doi: 10.2337/diacare.26.8.2405

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 36

  KwonJ.JungM.KimK.JoY.ShinJ.ChoY.et al. (2021). Artificial intelligence for detecting electrolyte imbalance using electrocardiography. Noninvasive Electrocardiol.26, e12839. doi: 10.1111/anec.12839

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 37

  KwonS.ThompsonR.DellingerP.YanezD.FarrohkiE.FlumD. (2013). Importance of perioperative glycemic control in general surgery: A report from the surgical care and outcomes assessment program. Ann. Surg.257, 8–14. doi: 10.1097/SLA.0b013e31827b6bbc

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 38

  LaiX.XuC.LiG.WangL. (2021). Prevalence of hyperkalemia and its effects on mortality in hospitalized patients: A single center 10-year retrospective analysis. Natl. Med. J. China.101, 3472–3477. doi: 10.3760/cma.j.cn112137-20210506-01062

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 39

  LeeE.LeeC. Y. (2021). PPG-based smart wearable device with energy-efficient computing for mobile health-care applications. IEEE Sensors J.21, 13564–13573. doi: 10.1109/JSEN.2021.3069460

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 40

  LiJ.MaJ.OmisoreO. M.LiuY.TangH.AoP.et al. (2024). Noninvasive blood glucose monitoring using spatiotemporal ECG and PPG feature fusion and weight-based choquet integral multimodel approach. IEEE Trans. Neural Netw. Learn Syst.35, 14491–14505. doi: 10.1109/TNNLS.2023.3279383

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 41

  LiT.YeS.LiJ. Z. (2021). High accuracy blood glucose monitoring based on ECG signals. Chin. J. Eng.43, 1215–1223. doi: 10.13374/j.issn2095-9389.2021.01.12.009

  • CrossRef
  • Google Scholar

• 42

  LinC. E.HirakaK.MatloffD.JohnsJ.DengA.SodeK.et al. (2018). Development toward a novel integrated tear lactate sensor using Schirmer test strip and engineered lactate oxidase. Sensors Actuators B: Chemical.270, 525–529. doi: 10.1016/j.snb.2018.05.061

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 43

  LinC.LinC.FangW. (2020). A deep-learning algorithm (ECG12Net) for detecting hypokalemia and hyperkalemia by electrocardiography: Algorithm development. JMIR Med. Inf.8, e15931. doi: 10.2196/15931

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 44

  LongC. A.FangZ. B.HuF. Y.AryaS.BrewsterL. P.DugganE.et al. (2019). Poor glycemic control is a strong predictor of postoperative morbidity and mortality in patients undergoing vascular surgery. J. Vasc. Surgery.69, 1219–1226. doi: 10.1016/j.jvs.2018.06.212

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 45

  MasonA.KorostynskaO.LouisJ.Cordova-LopezL. E.AbdullahB.GreeneJ.et al. (2018). Noninvasive in-situ measurement of blood lactate using microwave sensors. IEEE Trans. BioMed. Eng.65, 698–705. doi: 10.1109/TBME.2017.2715071

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 46

  MearaJ.LeatherA.HaganderL. (2015). Others. Global surgery 2030: Evidence and solutions for achieving health, welfare, and economic development. Lancet (lond Engl).386, 569–624. doi: 10.1016/S0140-6736(15)60160-X

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 47

  MillerF.MurrayJ.BudhotaA. (2023). Evaluation of a wearable biosensor to monitor potassium imbalance in patients receiving hemodialysis. Sens Bio-Sens Res.40, 100561. doi: 10.1016/j.sbsr.2023.100561

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 48

  MiottoR.WangF.WangS.JiangX.DudleyJ. T. (2018). Deep learning for healthcare: Review, opportunities and challenges. Briefings Bioinf.19, 1236–1246. doi: 10.1093/bib/bbx044

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 49

  MuehlematterU.BluethgenC.VokingerK. (2023). FDA-cleared artificial intelligence and machine learning-based medical devices and their 510(k) predicate networks. Lancet Digit Health5, e618–e626. doi: 10.1016/S2589-7500(23)00126-7

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 50

  MurthiM.ShakaH.El-amirZ.VelagapudiS.JamilA.WaniF.et al. (2021). Association of hypocalcemia with in-hospital mortality and complications in patients with acute pulmonary embolism: results from the 2017 Nationwide Inpatient Sample. BMC Pulm Med.21, 410. doi: 10.1186/s12890-021-01784-0

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 51

  NieZ.RongM.LiK. (2023). Blood glucose prediction based on imaging photoplethysmography in combination with machine learning. BioMed. Signal Process Control.79, 104179. doi: 10.1016/j.bspc.2022.104179

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 52

  Norring-AgerskovD.MadsenC. M.AbrahamsenB.RiisT.PedersenO. B.JørgensenN. R.et al. (2017). Hyperkalemia is associated with increased 30-day mortality in hip fracture patients. Calcif Tissue Int.101, 9–16. doi: 10.1007/s00223-017-0252-9

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 53

  OlaetxeaI.ValeroA.LopezE.LafuenteH.IzetaA.JaunarenaI.et al. (2020). Machine learning-assisted Raman spectroscopy for pH and lactate sensing in body fluids. Anal. Chem.92, 13888–13895. doi: 10.1021/acs.analchem.0c02625

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 54

  PatelB.DaveB.DaveD.KarmakarP.ShahM.SarvaiyaB. (2015). Comparison and correlation of glucose levels in serum and saliva of both diabetic and non-diabetic patients. J. Int. Oral. HEALTH : JIOH.7, 70–76.

  • Google Scholar

• 55

  PetropoulosK.PiermariniS.BernardiniS.PalleschiG.MosconeD. (2016). Development of a disposable biosensor for lactate monitoring in saliva. Sensors Actuators B: Chemical.237, 8–15. doi: 10.1016/j.snb.2016.06.068

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 56

  PorsA.KorzeniowskaB.RasmussenM. T.LorenzenC. V.RasmussenK. G.InglevR.et al. (2025). Calibration and performance of a raman-based device for non-invasive glucose monitoring in type 2 diabetes. Sci. Rep.15, 10226. doi: 10.1038/s41598-025-95334-x

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 57

  PorumbM.StrangesS.PescapèA.PecchiaL. (2020). Precision medicine and artificial intelligence: A pilot study on deep learning for hypoglycemic events detection based on ECG. Sci. Rep.10, 170. doi: 10.1038/s41598-019-56927-5

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 58

  Rabost-GarciaG.ColmenaV.Aguilar-ToránJ.Vieyra GalíJ.Punter-VillagrasaJ.Casals-TerréJ.et al. (2023). Non-invasive multiparametric approach to determine sweat–blood lactate bioequivalence. ACS Sens.8, 1536–1541. doi: 10.1021/acssensors.2c02614

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 59

  RafiqueZ.AcevesJ.EspinaI.PeacockF.Sheikh-HamadD.KuoD. (2020). Can physicians detect hyperkalemia based on the electrocardiogram? Am. J. Emergency Med.38, 105–108. doi: 10.1016/j.ajem.2019.04.036

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 60

  RegolistiG.MaggioreU.GrecoP.MaccariC.ParentiE.Di MarioF.et al. (2020). Electrocardiographic T wave alterations and prediction of hyperkalemia in patients with acute kidney injury. Intern. Emerg. Med.15, 463–472. doi: 10.1007/s11739-019-02217-x

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 61

  RuggeriE.MatzeuG.VergineA.NicolaoG.OmenettoF. (2023). Paper-based wearable patches for real-time, quantitative lactate monitoring. Adv. Sens Res.3, 2300141. doi: 10.1002/adsr.202300141

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 62

  SahaS.Cano-GarciaH.SotiriouI.LipscombeO.GouzouasisI.KoutsoupidouM.et al. (2017). A Glucose Sensing System Based on Transmission Measurements at Millimetre Waves using Micro strip Patch Antennas. Sci. Rep.7, 6855. doi: 10.1038/s41598-017-06926-1

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 63

  SahaT.KhanM.SandhuS. (2024). A passive perspiration inspired wearable platform for continuous glucose monitoring. Adv. Sci. (weinh Bad-württ Ger.11, e2405518. doi: 10.1002/advs.202405518

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 64

  SempionattoJ. R.MoonJ. M.WangJ. (2021). Touch-based fingertip blood-free reliable glucose monitoring: personalized data processing for predicting blood glucose concentrations. ACS Sens.6, 1875–1883. doi: 10.1021/acssensors.1c00139

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 65

  SeveriS.CorsiC.HaigneyM.DeBieJ.MortaraD. (2009). Noninvasive potassium measurements from ECG analysis during hemodialysis sessions2009, 821–824.

  • Google Scholar

• 66

  ShangH.YuS.WuY.LiuX.HeJ.MaM.et al. (2025). A noninvasive hyperkalemia monitoring system for dialysis patients based on a 1D-CNN model and single-lead ECG from wearable devices. Sci. Rep.15, 2950. doi: 10.1038/s41598-025-85722-8

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 67

  SinghA. K.JhaS. K. (2019). Fabrication and validation of a handheld non-invasive, optical biosensor for self-monitoring of glucose using saliva. IEEE Sensors J.19, 8332–8339. doi: 10.1109/JSEN.2019.2920296

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 68

  TejchmanK.DomanskiL.SienkoJ.SulikowskiT.KaminskiM.RomanowskiM.et al. (2007). Influence of perioperational acid-base balance disorders on early graft function in kidney transplantation. Transplant. Proc.39, 848–851. doi: 10.1016/j.transproceed.2007.03.037

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 69

  TierneyM. J.TamadaJ. A.PottsR. O.JovanovicL.GargS. (2001). Clinical evaluation of the GlucoWatch^® biographer: a continual, non-invasive glucose monitor for patients with diabetes. Biosensors Bioelectronics.16, 621–629. doi: 10.1016/S0956-5663(01)00189-0

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 70

  ToboreI.LiJ.KandwalA.YuhangL.NieZ.WangL. (2019). Statistical and spectral analysis of ECG signal towards achieving non-invasive blood glucose monitoring. BMC Med. Inf Decis Mak.19, 266. doi: 10.1186/s12911-019-0959-9

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 71

  TsaiC. W.LiC. H.LamR. W. K.LiC. K.HoS. (2020). Diabetes care in motion: blood glucose estimation using wearable devices. IEEE Consumer Electron Mag.9, 30–34. doi: 10.1109/MCE.2019.2941461

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 72

  TsengP.NapierB.GarbariniL.KaplanD. L.OmenettoF. G. (2018). Functional, RF-trilayer sensors for tooth-mounted, wireless monitoring of the oral cavity and food consumption. Advanced Materials.30, 1703257. doi: 10.1002/adma.201703257

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 73

  U.S. Food and Drug Administration (2018a) in De novo classification request for acumen hypotension prediction index feature (DEN160044) (Center for Devices and Radiological Health). Available online at: https://www.accessdata.fda.gov/cdrh_docs/reviews/DEN160044.pdf (Accessed March 1, 2026).

  • Google Scholar

• 74

  U.S. Food and Drug Administration (2018b) in De novo classification request for dexcom G6 continuous glucose monitoring system (DEN170088) (Center for Devices and Radiological Health). Available online at: https://www.accessdata.fda.gov/cdrh_docs/reviews/DEN170088.pdf (Accessed March 1, 2026).

  • Google Scholar

• 75

  U.S. Food and Drug Administration (2021) in 510(k) substantial equivalence determination decision summary for acumen hypotension prediction index (HPI) feature software (K203224) (Center for Devices and Radiological Health). Available online at: https://www.accessdata.fda.gov/cdrh_docs/pdf20/K203224.pdf (Accessed March 1, 2026).

  • Google Scholar

• 76

  U.S. Food and Drug Administration (2022) in 510(k) substantial equivalence determination decision summary for dexcom G7 continuous glucose monitoring system (K213919) (Center for Devices and Radiological Health). Available online at: https://www.accessdata.fda.gov/cdrh_docs/reviews/K213919.pdf (Accessed March 1, 2026).

  • Google Scholar

• 77

  U.S. Food and Drug Administration (2024) in 510(k) substantial equivalence determination decision summary for stelo glucose biosensor system (K234070) (Center for Devices and Radiological Health). Available online at: https://www.accessdata.fda.gov/cdrh_docs/reviews/K234070.pdf (Accessed March 1, 2026).

  • Google Scholar

• 78

  VelagapudiV.O’HoroJ. C.VellankiA.BakerS. P.PidikitiR.StoffJ. S.et al. (2017). Computer-assisted image processing 12 lead ECG model to diagnose hyperkalemia. J. Electrocardiology.50, 131–138. doi: 10.1016/j.jelectrocard.2016.09.001

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 79

  von BachmannP.GedonD.GustafssonF. (2024). Evaluating regression and probabilistic methods for ECG-based electrolyte prediction. Sci. Rep.14, 15273. doi: 10.1038/s41598-024-65223-w

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 80

  WangK.ZhangN.DengD.QiuY.LinY.JinS. (2018). Clinical analysis of perioperative electrolyte imbalance in 999 patients undergoing gastrointestinal surgery. Chin. J. Gastrointest Surg.21, 1427–1432.

  • Google Scholar

• 81

  WangC. X.ZhangY. C.KongQ. L.WuZ.-X.YangP.-P.ZhuC.-H.et al. (2021). Development and validation of a deep learning model to screen hypokalemia from electrocardiogram in emergency patients. Chin. Med. J. (Engl.134, 2333–2339. doi: 10.1097/CM9.0000000000001650

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 82

  YeraganiV. K.SrinivasanK.BalonR.BerchouR. (1996). Effects of lactate on cross-spectral analysis of heart rate, blood pressure, and lung volume in normal volunteers. Psychiatry Res.60, 77–85. doi: 10.1016/0165-1781(95)02860-9

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 83

  YuzerE.DoganV.KilicV.SenM. (2022). Smartphone embedded deep learning approach for highly accurate and automated colorimetric lactate analysis in sweat. Sens Actuators B.371, 132489. doi: 10.1016/j.snb.2022.132489

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 84

  ZhangQ.JiangD.SunZ.ShiX.ChenX.LiQ.et al. (2026). An ant-nest-structured wearable sensor with pH calibration for reliable monitoring of sweat glucose. ACS Sens.11, 384–393. doi: 10.1021/acssensors.5c03084

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 85

  ZhangG.MeiZ.ZhangY.MaX.LoB.ChenD.et al. (2020). A noninvasive blood glucose monitoring system based on smartphone PPG signal processing and machine learning. IEEE Trans. Ind. Inf.16, 7209–7218. doi: 10.1109/TII.2020.2975222

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 86

  ZhangY.ZhangL.WangL.ShaoS.TaoB.HuC.et al. (2025). Subcutaneous depth-selective spectral imaging with mμSORS enables noninvasive glucose monitoring. Nat. Metab.7, 421–433. doi: 10.1038/s42255-025-01217-w

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 87

  ZiZ.ShuangW.ZizhuL.QingC.KangC. (2023). An artificial intelligence algorithm for hierarchical prediction of arterial blood lactate level based on electrocardiogram images. Chin. J. Card Arrhyth.27, 287–294. doi: 10.3760/cma.j.cn113859-20230616-00091

  • Pubmed Abstract
  • CrossRef
  • Google Scholar