Artificial intelligence-enhanced diagnostics for pediatric asthma and respiratory irregularities using deep learning and wearable sensors

AI in immunologic disease diagnosis (AI-IDD)

This article by Khan et al. (2024) reviews the opportunities of AI, including ML and DL, in the context of immunological ailments diagnosis and management. Integrating AI into clinical practice will help improve diagnosis abilities, clinical research, and disease monitoring. The study highlights the revolutionary potential of AI in healthcare, enabling more accurate and faster diagnosis of immune-related diseases.

AI-powered sensor fabrics for real-time health monitoring (AI-PSF)

This study by Sarker (2025) focuses on sensor textiles that incorporate AI to provide continuous, real-time health monitoring. With biosensors, these wearable devices can track vital signs, including heart rate, blood sugar, and respiratory rate. Case studies demonstrate the effectiveness of these strategies in remote patient monitoring, hospital adoption, and patient outcomes, which help enhance healthcare efficiency and reduce hospital readmissions.

AI’s transformative impact on healthcare

Zeb et al. (2024) describe several applications of AI in healthcare, including robotic surgery, medication research, customized treatment, and diagnostics. It raises questions about ethical concerns, such as computational bias, data integrity, patient privacy, and the potential benefits of AI in improving the accuracy of surgical and disease detection. The study highlights the potential impact of AI on medical research and treatment.

AI and wearable sensors in nanomedicine

As this study by Ahmad & Muhmood (2024) suggests, one approach in which AI and ML can enhance nanomedicine is by utilizing advanced sensors for on-site data collection. Using real-time biochemical data, wearable gadgets fueled by AI significantly improve healthcare decision-making. Furthermore, the research also addresses the challenges of overcoming computational blind spots and legal constraints that hinder the implementation of AI-assisted nanomedicine in clinical settings.

Ethical considerations in AI-powered disease diagnosis

Concerns regarding AI-driven disease diagnosis include algorithmic prejudice, data privacy, and consent. The article by Shah (2023) highlights how wearable technology, AI, and ML can enhance early diagnosis while maintaining fair and reliable healthcare systems. The report emphasizes that AI-powered healthcare solutions depend on a compromise between technical development and ethical problems.

AI in diagnostic tools and predictive analytics

This article by Dangi, Sharma & Vageriya (2025) explores how AI technologies, including robotics and ML, improve the precision of surgical operations, predictive analytics, and diagnostics. The article highlights how AI can aid in early disease detection and personalized treatment planning in resource-constrained settings. It highlights how AI can help mitigate shortcomings and enhance healthcare outcomes.

Supervised machine learning (SML) in healthcare

This work by Roy, Meena & Lim (2022) investigates the impacts of SML in healthcare applications, including disease diagnosis, personalized therapy, remote monitoring, and drug discovery. It emphasizes the need for data quality awareness and highlights the importance of understandable AI for biological researchers. The article also addresses potential innovations and ethical issues related to SML in healthcare.

AI-powered mobile health (AIM) for disease management

The current scoping review tracks studies on AI-driven mobile health apps using wearable sensor technologies and smartphones by Bhatt et al. (2022). It highlights how DL and FL facilitate the development of forecasting models for treating chronic illnesses. Particularly during the COVID-19 pandemic, the article emphasizes how AIM solutions are increasingly utilized for handheld health monitoring, which depends on safe data interchange in healthcare.

The literature study highlights how predictive modeling, real-time monitoring, and improved diagnosis accuracy enabled by AI can transform healthcare. Wearable gadgets driven by AI, FL, and DL models enhance patient care and disease detection using our proposed method, AI-PARI, compared to other traditional methods, including AI-IDD, AI-PSF, SML, and AIM.

Traditional clinical methods, such as physical examination, patient history, and manual spirometry, are commonly used to diagnose and manage pediatric asthma. However, these methods have limitations regarding real-time monitoring and detecting subtle, early-stage asthma exacerbations. ML models for pulmonary diagnostics have been the subject of numerous investigations; however, these studies have predominantly focused on static, clinical data or short-term recordings, rather than continuous, real-time data. Additionally, most current solutions are limited in their predictive power and ability to provide real-time feedback, as they rely on symptom-based evaluations or single-modality data, such as imaging or physical examinations. To overcome this limitation, the AI-powered diagnostic system for pediatric asthma and respiratory abnormalities (AI-PARI) utilizes data from wearable sensors, specifically lung sounds, to train a DL model to track the respiratory system in real-time without requiring intrusive procedures. To identify asthma exacerbations and abnormalities early on, AI-PARI records changes in lung acoustics (such as wheezing or crackling) in real-time, unlike conventional diagnostic devices. To further enhance its generalizability and accuracy, the proposed system is trained on a large and diversified dataset using the Pulmonary Sound Dataset from Kaggle. This dataset comprises different and labelled lung sound recordings.

Pediatric respiratory health diagnostic framework powered by AI

Pediactric with asthma and respiratory disorders are greatly benefitted from early detection and immediate care. This facilitation will prevent the condition from becoming worse.

The AI-powered pediatric respiratory health diagnostic model continuously monitors respiratory health of prediatric, because it integrates wearable sensor technology and novel DL models.

Unusual breathing patterns related to asthma attacks and other respiratory conditions are effective detected by the system with the support of CNN and RNN. Then, abnormal patterns are detected earlier to avoid severe risks, so crucial signs like respiration rate, oxygen saturation, airflow patterns, and heart rate variations were detected by the system in real-time. It is illustrated in Fig. 1. Thus, predictive insights are offered by this AI system, as it analyses data. These insights may support the doctors in providing preventive therapy before the condition gets worse. The integration of wearable sensor technology and novel DL models in the suggested AI-powered diagnostic system enhances the monitoring of respiratory health of pediatrics. Improved accuracy, reliability, and applicability are some of the benefits that are offered by these technologies. The complex, and high-dimensional data may result from the signals collected from wearable sensors, and these complicated and high-dimensional data are effectively handled by the potential of DL models. Conventional methods fail to detect those hidden patterns in the data but, these models have the ability to detect such patterns effectively.

DL methods are able to manage large respiratory patterns and health marker datasets, and they offer customized predictions. So, these DL methods are adaptable to the specific health profile of every pediatric patient. The respiratory problems or asthma cases are accurately detected by the system with DL methods, and it also supports in proving immediate treatment. In contrast, wearable sensors enable the continuous monitoring of vital signs, including respiratory rate, lung capacity, oxygen saturation, and cough frequency, through a non-invasive, real-time data collection system. Wearable sensors are superior to conventional diagnostic procedures for children because they can collect continuous, high-frequency data without interfering with the patient’s everyday activities. To intervene quickly before an assault or other catastrophic event occurs, constant monitoring helps identify early indications of anomalies. DL offers robust, predictive insights, while wearable sensors facilitate continuous, real-time data collection. The AI-PARI is an integrated method, that is effective for managing pediatric asthma, and respiratory irregularities. Improvements in model customization is done via the FL framework. The patient’s data is confidentially secured by this FL model.

(1) $$ct\left(ps\right)=\sum _{bp}rp{s}_{ds}\left(ct\right)-2^{-\frac{bp}{2}}\ast \mathrm{\partial }\left(2^{-ms}Sm-ms\right).$$

Determining if a specific pulmonary signal $(ps)$ matches with standard or disordered breathing patterns bp is done by the classification tools $\left(ct\right)$ in the AI models using Eq. (1). The sleep apnea, bronchitis, and asthma are distinguished by their symptoms by these DL models. Then, respiratory conditions are effectively treated by a novel data-driven medical solutions $2^{-ms}$ when $\left(2^{-ms}Sm-ms\right).$ Thus, it analyses the revolutionary possibilities $rps$ of AI-powered diagnostics $ds.$

(2) $$\int {\left|PC\left(rd\right)\right|}^{2}dt=\sum _{ba}{\left|P{C}_{rs}\left(prt\right)\right|}^2{+}^{{\sum }_{wb}ir}.$$

In Eq. (2), the historical data ${\left|PC\left(rd\right)\right|}^2$ and present physiological aspects $ba$ are integrated, and this integration may support in predicting the chances of a pediatric child developing a respiratory disease. For prompt action, and preventive treatment, it assists in offering risk assessments $P{C}_{rs}.$ The integration of DL models and modern wearable biosensors in the AI-PART may facilitate in promoting real-time monitoring and early identification of respiratory irregularities.

(3) $$RD\left(sp\right)=\sum _i{\left|d_{tv}\left(cm\right)\right|}^2+\mathrm{\Phi }\left(2^{-di}ba-sp\right).$$

For respiratory data $RD$, pulse patterns, basic components, and temporal variations $d_{tv}$ are some of significant properties $\left(sp\right)$ are generated by the AI framework, and it is defined in the procedure given in Eq. (3). These features enter the classification model $\left(cm\right)$ to ensure exact diagnosis of breathing anomalies by $\mathrm{\Phi }\left(2^{-di}ba-sp\right)$.

The pediatric respiratory irregularities are detected by the system that utilizes wearable sensors for collecting real-time physiological data, and it is presented in Algorithm 1. It sorts heart rate, oxygen saturation in the airflow, and breathing rate using predefined threshold values. It utilizes sensor data to detect hypoxia, constricted airways, and abnormal heart rates, which can aid in the early diagnosis of asthma.

The first step is to monitor the patient’s breathing patterns, oxygen saturation, lung capacity, and other respiratory parameters in real-time using wearable sensors. After that, the data is safely sent to the system, where a DL model that can identify patterns of children’s breathing problems or asthma episodes processes it. The system’s DL algorithms examine the data for outliers or precursors to anticipate such issues before they escalate. Caregivers and healthcare professionals can intervene proactively and modify therapy by receiving real-time notifications based on these forecasts. Over time, the AI-PARI system improves accuracy and refines predictions by incorporating input from caregivers and healthcare experts.

Wearable biosensor integration for real-time monitoring

Wearable biosensors are crucial in pediatric respiratory health monitoring, as they enable noninvasive and continuous tracking of vital physiological data. Constant health monitoring made possible by these biosensors helps identify respiratory anomalies and asthma episodes early.

Real-time data is securely sent to a remote server, where AI systems identify anomalies, as shown in Fig. 2. This perfect integration of devices raises patient engagement by permitting remote monitoring and reducing the need for frequent hospital visits.

Clinical professionals and caretakers may give an immediate care, and it was facilitated by the application of cloud-based dashboard, as it offers real-time notifications of any variations. This condition is effectively managed; it also reduces the risk of the condition and hospital visits.

(4) $$S^{\prime }{o}_{(mo)i}=h{r}_d(hp\ast ds{)}_{rtp},rtp=1,2,\dots ,fm,mi\in \left\{pd\right\}.$$

In Eq. (4), outputs of many sensors $S^{\prime }o$ are presented. Monitoring oxygen levels $mo$, heart rate $h{r}_d$, and airflow are integrated in $S^{\prime }o$. Thus, a whole health profile $hp$ was facilitated by this integration. Then, real-time $rtp$ physiological data $pd$ are also ensured by this diagnostic system $ds$.

(5) $$ft{\left(gf\right)}_i=s{r}_i\ast ns\left(\prod _i\right)\ast \int \int r{d}_{i}dg,i=1,2,\dots N.$$

Equation (5) explains the technique of filtering $ft$ used to eliminate remnants $e{r}_i\ast ns$ and noise from sensor readings $s{r}_i$. Consequently, $r{d}_i$ the AI model seems more likely to generate a correct diagnosis $dg$ given reliable data. To identify irregular breathing patterns associated with asthma and other respiratory disorders, healthcare providers and caregivers can access real-time alerts and predictive insights through a cloud-based diagnostic dashboard, which processes these data streams.

(6) $$Srs=sl\left(\sum _{j}ef\ast CM{\left(nd\right)}_i\right)<\left|\sum _{j}rt\ast srs{\left(ac\right)}_i\right|.$$

Equation (6) manages the secure $sl$ and efficient data transport $ef$ based on gadgets with sensors and sensor reporting system (SRS) to server hardware in the cloud. By continuously monitoring data without delay $CM{\left(nd\right)}_i$, guarantees real-time alerts $rt$ in case of respiratory crisis $rp$ by comparing these two values like $srs\left(ac\right)$.

Combining sensor technology with AI-driven analytics provides a scalable, cost-effective, and efficient approach to pediatric asthma management. The proposed system detects heart rate variability, oxygen saturation, airflow patterns, respiratory rate, and respiration rate in real-time using wearable devices coupled with contemporary biosensors.

The wearable sensors played a crucial role in continuously monitoring the patient’s oxygen saturation, breathing rate, and lung capacity, among other respiratory parameters. These sensors provide non-invasive, continuous monitoring and a wealth of data to identify respiratory abnormalities early. The asthma cases or real-time irregularities are not effectively predicted by the system without sensors. So, the collected data are sent to DL models. Because, the patterns, and irregularities are effectively detected by the DL model, and it also have the potential to predict future occurrences via past histories. When comparing the DL models with conventional models, the conventional methods fail to detect such irregularities, but these DL models effectively detect with data analysis and prediction processes. One can ensure that the prediction is precise by training the algorithm on large datasets, and it supports in customizing those predictions to suit every patient-specific health profile. Then, feedback given by healthcare providers, and professionals via the feedback loop facilitates in improving the accuracy of the system for long time. Thus, the predictive abilities of the system are improved.

DL models for respiratory pattern analysis

For analysing irregularities in pediatric respiratory patterns, this AI model is crucial. The real-time physiological data are collected by wearable biosensors. For analysing these data, RNN and CNN are suggested.

Large dataset is utilized for training a model. The child airflow signals, clinical diagnoses, and environmental factors, including humidity and air quality are all included in this dataset, and it was presented in Fig. 3. The utilization of large datasets supports the AI models in detecting respiratory problems more effectively. DL enables the system to identify minute variations in respiratory patterns, allowing for a proactive response.

(7) $$\mathrm{\partial }(2^{-ms}Sm-ms)=\int {\left|PC\left(rd\right)\right|}^{2}dt:\to cm.$$

Equation (7) allows one to officially identify patterns $Sm-ms$ of visual interest from respiratory data as $\mathrm{\partial }\left(2^{-ms}Sm-ms\right)$ using convolutional neural networks (CNNs) ${\left|PC\left(rd\right)\right|}^2$. In toddlers with asthma, it helps identify irregularities in airflow $dt:\to cm$. A significant step forward in pediatric respiratory treatment, this AI-powered system provides a scalable and cost-effective solution.

(8) $$\sum _{bp=T/cs}rp\left(pr\right):\to BP(tf\ast ds{)}_i+\sum _{j}es\ast bp{\left(aa\right)}_i.$$

In order to detect abnormal breathing patterns $BP(tf\ast ds{)}_i$, an RNN uses a paradigm shown in Eq. (8) to handle sequential respiratory data, including variations over time as ${\sum }_{bp=\frac{T}{cs}}rp\left(pr\right)$. It makes it possible to spot early on the signs that can aggravate asthma * $bp(aa{)}_i$.

(9) $$\int \int r{d}_{i}dg=h{r}_d(hp\ast ds{)}_{rtp}+CM(nd{)}_i.$$

Completing this calculation will help us to select the aim operations that most effectively $(hp\ast ds{)}_{rtp}$ balances the actual values of the respiratory $r{d}_{i}dg$ condition forecasts with their accuracy with Eq. (9). When the training $CM{\left(nd\right)}_i$ is increased, the prediction efficiency of the framework increased. Early symptoms of respiratory problems and accurate detection of asthma symptoms are offered by the integration in this system.

When compared to conventional diagnostic methods, the AI-driven approach is scalable, accurate, and effective. While CNNs excel at identifying spatial features in airflow patterns, RNNs analyze sequential data to detect fluctuations in breathing patterns, enabling the detection of respiratory anomalies over time.

Cloud-based predictive analytics and federated learning

The suggested strategy integrates a cloud-based predictive analytics framework to offer real-time access and raise diagnostic accuracy. The cloud platform utilizes data from wearable biosensors to evaluate trends suggestive of respiratory discomfort using AI techniques.

This method guarantees strict security and enhances customization while remaining compliant with healthcare regulations, as shown in Fig. 4. Due to analytics that utilize the cloud and storage, which also help the system remain scalable, it is ideal for large-scale installations. The proposed approach utilizes FL to improve AI models by optimizing diagnosis accuracy and flexibility.

(10) $$AD\left(p{z}^{\prime }-h{i}^{″}\right)=\left(eps-f{d}^{″}\right)\ast iq\left(s-efi{f}^{″}\right).$$

Assessing the trends in one’s airflow data $AD(p{z}^{\prime }-h{i}^{″})$, this Eq. (10) reveals one method to estimate potential hazards for health $ps-f{d}^{″}$. This allows doctors to act before major asthma episodes even begin $iq\left(s-efi{f}^{″}\right)$. Analytical foresight and FL, taken together, increase the safety and efficiency of asthma therapy for children.

(11) $$Vd\left(rd-s{g}^{″}\right)=(f{i}^{\prime }-da(pr-s{r}^{″}))\ast fl(iv-p{d}^{″}).$$

All algorithms trained on various devices create $Vd$ a single global $s{g}^{\prime }$ model without exchanging any raw data $rd$ using this Eq. (11); this is the federated instructional $f{i}^{\prime }$ process defined by this equation. It improves diagnostic accuracy $da$ while ensuring privacy $pr$. $s{r}^{″}$ FL $fl$ allows for the refining of individualized models $iv$ and the protection of personal data $p{d}^{″}$ without sacrificing security. This technology enhances disease care and reduces the need for emergency hospital visits by outperforming conventional diagnostic methods.

(12) $$\sum _{j}es\ast bp{\left(aa\right)}_i\ast \iint r{d}_{i}dg+CM{\left(nd\right)}_i.$$

Equation (12) drives the regular upgrades ${\sum }_{j}es\ast bp{\left(aa\right)}_i$ of cloud-based AI models $\int \int r{d}_{i}dg$ using recently acquired data gathered by wearable devices $CM{\left(nd\right)}_i$. It assures us that it never stops and that the diagnostic precision increases with time.

This AI-powered approach examines processed sensor data using a risk-based scoring system to diagnose respiratory illnesses. Weighing abnormalities such as low oxygen levels, fast breathing, and airflow restrictions helps us understand them using Algorithm 2. The cumulative risk score enables caregivers to obtain an AI-based diagnosis in real-time, allowing for proactive intervention in situations of pediatric asthma.

Predictive analytics informs caregivers and medical professionals about possible medical emergencies, enabling quick response. An essential component of the system is that FL guards patients’ privacy by allowing AI models to be trained across numerous devices without sharing any raw patient data.

Predictive analysis of the proposed system

Evaluation of the proposed AI-driven diagnostic system’s performance will help assess its effectiveness in identifying young respiratory diseases. Using significant performance metrics such as recall, precision, and F1-score, we evaluate the system’s accuracy in identifying bouts of asthma and breathing irregularities. The proposed approach aims to transform pediatric respiratory treatment by continually refining and utilizing scalable AI-powered solutions.

(13) $$Ac\left(r{i}^{\prime }-bfp\right)=Un\left(pt-rt\right)\ast cs\left(A{C}^{″}-bfp\right).$$

Equation (13) gauges the appliance’s accuracy $Ac$ in spotting respiratory infections $r{i}^{\prime }$ by restricting both false negatives $bfp$ and unintended positives $Un\left(pt-rt\right)$. This helps one to compare the model $\left(A{C}^{″}-bfp\right)$ with accepted clinical standards $cs$.

(14) $$p{r}_f-st\left(rc-p{r}^{\prime }\right):\to ml\ast dl\left(pr.dd{l}^{″}\right).$$

The approach described above combines $p{r}_f$ recall $rc$ alongside preciseness into one statistic $st$ to provide a more realistic view of the performance $p{r}^{\prime }$ of the AI model $ml\ast dl$ with this Eq. (14). It ensures the procedure will attain a dependable diagnosis level as $(pr.dd{l}^{″})$ the DDL.

(15) $$Ne{v}^{\prime }\left(rd-ys\right)=da\left(gt-se(h{r}^{\prime }\right))\ast [i{f}^{\prime }-uyq].$$

Equation (15) above defines the numerical evaluation $Ne{v}^{\prime }$ of the AI model’s performance on the unavailability of respiration datasets $rd$. It guarantees it can adapt to several types of surroundings $uyq$ and many sorts of youngsters $ys$. The initial findings $i{f}^{\prime }$ reveal that the detection accuracy $da$ rate is far greater than $gt$ the outcomes of conventional diagnostic techniques $\left(gt-se(h{r}^{\prime }\right))$.

This system’s ability to outperform conventional diagnostic procedures has reduced reliance on emergency room visits and enhanced illness management. Future upgrades will optimize the framework’s usefulness and scalability in pediatric cardiopulmonary healthcare, including combination multiple sensors, model abstraction, and large-scale clinical trials.

A novel method is created by integrating AI and wearable biosensors with DL. This novel method is effective in detecting pediatric asthma cases and respiratory problems. RNN and CNN are suggested for analysing those real-time physiological data. The simulation was conducted by comparing AI-PARI with current methods. The AI-PARI attains 95% detection accuracy and it is superior when compared to conventional diagnostic methods.

Computing infrastructure

The computer runs with multiple operating systems like Windows 11 Professional and Ubuntu 22.04 Long-Term Support was executed in this analysis. NVIDIA RTX 3060/3070 graphics processing unit (GPU) with CUDA support for accelerated DL research, 32 gigabytes of DDR4 RAM, and an Intel Core i7 11th Generation central processing unit (CPU) or an AMD Ryzen 7 5800X were included in this hardware configuration. The efficacy of the system is improved by all these components. The storage system was upgraded with a 512 GB NVMe solid-state drive (SSD) to handle large amounts of model weights and sensor data efficiently. This was done to maintain the system working smoothly. Python 3.8 was used to build all the ML processes. Libraries such as TensorFlow 2.8, NumPy 1.21, Pandas 1.3, Scikit-learn 1.0, and SciPy 1.7 were utilized.

Data preprocessing

Synthetic multi-sensor data have been designed to represent four vital physiological characteristics, replicating juvenile respiratory function in a real environment. Some of the criteria in this evaluation are breathing rate, oxygen saturation, heart rate variability, and airflow patterns. Every sample has four features and 100 steps added to it. Before the model received instructions, every single sensor input was subjected to a low-pass Butterworth filter with an order of three and a cutoff frequency of 3 Hz. It was intended to eliminate ambient artifacts and high-frequency noise, making the data on breathing patterns as accurate as possible. Data normalisation came next to bring all sensor values within the range of [0, 1]. This led to the standardization of input across several signal types. The data was then standardized. A split ratio of 80:20 divided the last dataset into training and testing sets. Stratified sampling helped to achieve this. This maintains a harmonic balance between regular and abnormal breathing patterns.

Deep learning model

A hybrid architecture has been employed to create the core of the AI diagnostic system. The system consisted of a RNN and a CNN. The layers of the CNN have been assigned tasks to obtain spatial features from the filtered sensor inputs. To do this, it was necessary to collect localized patterns that indicated respiratory anomalies. After completing the feature extraction approach, long short-term memory (LSTM) layers were used to capture the temporal correlations inherent in sequential breathing data. These layers were included following the feature extraction procedure. The model’s architecture consists of two convolutional layers, each with 64 and 128 filters, respectively, followed by max-pooling operations and a single LSTM layer with 64 hidden units. The model was then trained using the model. The last output of the dense layers produced was a sigmoid activation function, employed for binary classification (normal versus disordered breathing). The model was trained for 20 epochs using the Adam optimizer by employing a batch size of thirty-two. The loss function in this work represented a binary cross-entropy value.

Federated learning framework simulation

An FL system was implemented to simulate distributed learning environments and safeguard people’s privacy. Five simulated consumers shared the training dataset, each receiving a distinct fraction of the data. Without sharing any raw sensor data with the central server, each client autonomously trained a local instance of the CNN-RNN model on its private data. After completing each training cycle, the server received the modified model weights; these were then aggregated and updated using federated averaging. A global model was produced by repeating this procedure. This approach safeguarded patient data confidentiality and mimicked the restrictions in the actual world, where wearable gadgets spread data throughout the sector. There was no direct data transfer between the clients and the central node; therefore, the confidentiality of the data was maintained during the simulation.

The evaluating approach

Several distinct criteria—including accuracy, precision, recall, and F1-score—were used to evaluate the effectiveness of the AI-driven diagnostic system. The model was assessed once the training procedure finished using the testing dataset set aside to gauge its predictive ability. The predictions produced by the model were compared with the ground-truth labels using typical classification assessment methods. A categorization report covered every class’s accuracy, recall, and F1-scores. Understanding and validating the consistency and stability of a federated learning (FL) system depends on analysing the mean and standard deviation of data across simulated federated clients.

Analysis of signal patterns

Wearable sensors record respiratory signals; our work aims to extract pertinent data from these signals using AI-PRAI, achieving an output of 93.1% (Fig. 5). Signal processing approaches enable the identification of variations in oxygen saturation, respiratory rate, and airflow using cloud-hosted diagnostic dashboard (C-DB), with an output of 93.8%. The irregularities that are related to asthma cases are effectively detected by the patterns obtained. It provides early detection and effective clinical care.

(16) $$\left(SP-f{c}^{″}\right)=rd\left(b{p}^{\prime }-si\right)\ast rf\left\{an:\to a{s}^{\prime }=0\right\}.$$

The frequency components from respiratory data $rd$ was extracted by the application of signal processing algorithm $\left(SP-f{c}^{″}\right)$, and it may support in detecting aberrant breathing patterns $b{p}^{\prime }-si$ in Eq. (16). To detect asthma-related anomalies $an:\to a{s}^{\prime }$, DL methods transform the chaotic sensory information $rf$ into a more realistic form.

Analysis of AI model performance

Testing the performance of the AI model ensures the system’s accuracy in identifying respiratory issues, as shown in Table 2. Essential steps include precision, recall, and F1-score, which evaluate detection accuracy, minimizing false positives and negatives with an optimized output. The normal respiratory pattern and abnormal respiratory pattern are effectively distinguished by these DL models with two crucial features like precision consistency, and robustness, and it was demonstrated by this analysis.

(17) $$\left[i.ef\left(a^{\prime }-m.r{c}^{″}\right)\right]=c{e}^{\prime }-Rp\left(pp-x{i}^{″}\right).$$

In Eq. (17), correct evaluation $c{e}^{\prime }$ is ensured by assessing the accuracy $a^{\prime }$, recall $m.r{c}^{″}$, and F1-score by the application of an identification effectiveness $i.ef$. Thus, the capacity of the AI model to identify pediatric respiratory problems $Rp$ is also ensured. The ratio of accurate forecasts $\left(pp-x{i}^{″}\right)$ to those that turn out to be incorrect can be found using this equation.

Analysis of sensor accuracy

In real-time, reliable data was offered by the precise wearable sensors, and it is crucial for monitoring asthma in pediatrics. Calibration processes take into account ambient factors like temperature and humidity using the method C-DB, that produces an output of 95.2% and ensures consistent physiological results for the suggested AI system, which has an output of 94.1% in Fig. 6. The analysis of sensor dependability utilizing error detection and repair procedures improves data quality, and it supports in preventing misdiagnoses caused by sensor drift or external disturbances.

(18) $$s\left\{as.lm{e}^{″}\right\}=d{r}^{″}<pm-rtc{c}^{″}>\ast \left\{dt\left(hr-r{h}^{\prime }\right)\right\}.$$

Data reliability $s\left\{as.lm{e}^{″}\right\}$ and measurement errors $d{r}^{″}$ are reduced by using a sensor assurance in Eq. (18). The physiological measures $⟨pm-rtc{c}^{″}⟩$ collected in real-time are consistent, and this is ensured by this equation, with a consideration of environmental variations. The metrics that are used to assess the efficiency of models, are recall, accuracy, and F1-score. When compared to the traditional diagnostic techniques $dt$, the AI model’s efficiency is assessed. From the outcomes, it is clear that the AI models effectively manage respiratory health $\left(hr-r{h}^{\prime }\right)$ with a high detection accuracy of more than 95%.

Analysis of FL efficiency

Without impacting the privacy of patient data, this FL supports in customizing model. When AI models are trained across multiple devices, efficiency analysis evaluates data integration, computational costs, and convergence rate (Table 3). Since the system adapts to various patient conditions while maintaining safe and distributed learning, AI-driven diagnostics are scalable and privacy-preserving.

(19) $$\left[i{f}^{\prime }-uyq\right]\cdot st\left(rc-p{r}^{\prime }\right)+\int \int r{d}_{i}dg.$$

The optimal Eq. (19) for shared learning drives $\left[i{f}^{\prime }-uyq\right]$ updates to prototypes on distributed devices $st\left(rc-p{r}^{\prime }\right)$, guaranteeing data privacy. Consequently, autonomous learning systems converge effectively without sharing raw data and AI model customization is improved. DL models can identify respiratory system anomalies with high degree of accuracy by effectively digesting physiological inputs from wearable biosensors $\int \int r{d}_{i}dg$. The AI system sends real-time alerts to healthcare professionals and caregivers that surpass traditional diagnosis methods.

Analysis of clinical validation

Clinical validation is the only way to verify the effectiveness of the AI system in real-world healthcare settings, using the proposed method AI-PARI, with an average value of 95.5% (Fig. 7). By comparing the outcomes of expert clinical assessments with those of AI-generated diagnoses, statistical validation techniques assess the accuracy and consistency of the expert clinical assessments. This article confirms that the proposed AI-based diagnostic approach complies with medical standards, as demonstrated by C-DB, with an output of 95.7%, fostering trust and adoption in pediatric respiratory care.

(20) $$rf\left\{an:\to a{s}^{\prime }=p\left(pp-x{i}^{″}\right)\right\}:\to s\left\{as.lm{e}^{″}\right\}.$$

Contrary to hypothesis testing $rf$, a statistically proven Eq. (20) is applied in opposing AI forecasts and as $an:\to a{s}^{\prime }$ true ground truth clinical diagnosis $p\left(pp-x{i}^{″}\right)$. This equation gauges the relevance $s\left\{as.lm{e}^{″}\right\}$ of AI-driven diagnosis to ensure that model recommendations complement expert medical evaluations.

According to the findings, the AI-powered method for diagnosing childhood asthma is accurate and efficient. The performance evaluation results of the system AI-PARI confirm the system’s reliability, as its F1-score exceeds industry standards.