Integrating artificial intelligence (AI) into wearable health monitoring systems introduces both innovation and regulatory complexity, requiring a standardized approach to ensure patient safety, software robustness, and compliance with international medical device regulations. This paper presents a comprehensive case study of Baby FM, an AI-powered wearable medical device developed for continuous temperature monitoring in pediatric and veterinary care. The system combines real-time sensing, secure cloud connectivity, and anomaly detection through interpretable AI models. We detail a structured approach to achieving high software quality through the implementation of a customized Quality Management System (QMS), in compliance with ISO 13485, ISO 14971, and IEC 62304. The QMS supported modular documentation, traceability, risk control, and test automation throughout the product life cycle. In addition, the study outlines the preparation of technical documentation for CE and ALIMS certification, including verification and validation evidence, clinical benefit justification, and post-market surveillance planning. Beyond the internal development and compliance strategy, this paper provides a comparative overview of standard adoption in multiple industries, including the med-tech, pharmaceutical, and industrial IoT sectors. This case study presents several key findings: Modular QMS implementation enabled incremental regulatory compliance without stalling agile software development, Early adoption of ISO 13485, ISO 14971, IEC 62304, and MDR 2017/745 reduced regulatory friction and aligned documentation with development milestones, Embedding explainable AI techniques (SHAP visualization and audit trails) improved transparency, clinician trust, and regulator acceptance. Comparative analysis confirms that medtech devices require more stringent certification than pharma and health IT but deliver stronger post-market accountability. The scalable architecture supports future extensions in oncology, fertility, and livestock monitoring. The findings illustrate how practices from these sectors can inform the development of intelligent health systems and guide strategic decisions for startups seeking certification. Insights from real-world clinical trials, regulatory interactions, and AI transparency strategies ofer a replicable methodology for innovators looking to balance agility and compliance in the development of AI-driven medical products.
In the last decade, advances in wearable technologies have fundamentally reshaped the way health data
is collected and analyzed [
The momentum behind wearable medical devices is driven by a combination of technical innovation
and systemic transformation in healthcare. Enhanced sensor accuracy, lower power consumption, and
user comfort have made continuous measurement feasible outside traditional clinical settings. At the
same time, advances in wireless protocols—such as Bluetooth Low Energy (BLE) and Narrowband
IoT (NB-IoT)—have made it possible to securely transmit patient data with minimal latency. These
capabilities are increasingly aligned with preventive, personalized medicine models, which rely on
ongoing monitoring rather than episodic clinical visits [
The Baby FM system was created in response to a distinct clinical need: continuous, non-invasive temperature tracking in infants and small animals, particularly in environments where access to care is limited or periodic checks are insuficient. Baby FM comprises a CE-compliant wearable sensor, a mobile application, and a cloud back-end that enables longitudinal monitoring, predictive alerts, and integration with electronic health records. Unlike traditional thermometers or infrared devices, the system allows for round-the-clock data collection, supports pattern recognition, and ofers clinicians traceable, explainable outputs for decision-making.
To navigate the regulatory landscape, the Baby FM team aligned its development approach with the
European Medical Device Regulation (EU MDR), under which the device qualifies as a Class IIa product.
This classification demands strict technical documentation, software validation, and implementation
of a formal Quality Management System (QMS). The process involved compliance with the following
standards:
• ISO 13485, which sets out the requirements for QMS specific to medical device manufacturers
and emphasizes documentation, traceability, and continuous improvement [
To ofer a clearer understanding of how quality and safety standards are applied across diferent industries, we included a comparative table based on sector-specific usage that can be found in Table 1 in Section 3.
The focus is on standards like ISO 13485, ISO 14971, and IEC 62304, which are foundational in the medical device field. Their implementation is also examined in related sectors such as pharmaceuticals, digital health platforms, and industrial IoT. This comparison sheds light on how regulatory priorities and technical expectations vary across domains, despite sharing similar goals related to reliability, risk mitigation, and system integrity.
Startups like Baby FM often face additional complexity due to limited resources and short development
cycles. Yet research has shown that a phased and modular approach to ISO 13485 implementation can
be both feasible and efective in early-stage environments [
This paper presents a structured case study of how Baby FM was designed, validated, and prepared for certification. We explore the intersection of software quality, risk management, and cross-sectoral standardization. Through this example, we aim to contribute actionable insights for technology developers and medical innovators seeking to navigate the evolving field of intelligent health systems while remaining compliant with national and international regulations.
This paper is structured to lead the audience by way of a logical arrangement of thoughts and findings that are as follows: Section 1 Introduction introduces the integration of AI in wearable medical devices and outlines the objectives and scope of the Baby FM case study. Section 2 Materials and Methods describes the system architecture, development tools, standards applied, and methodological framework used throughout the study. Section 3 Literature Review This section reviews existing work on AI-driven health monitoring, regulatory compliance, and quality management in medtech and related sectors. Section 4 Certification Pathway and Technical File Preparation this section outlines the structured process of preparing technical documentation for CE and ALIMS certification, including risk analysis and verification evidence and details the design and deployment of a modular QMS aligned with ISO 13485, ISO 14971, and IEC 62304 to support development and compliance. Section 5 Results this section presents the outcomes of QMS implementation, certification milestones, and the impact of explainable AI on transparency and acceptance and evaluates the device’s clinical accuracy, market comparison, and internal quality metrics. Section 6 Discussion: Lessons Learned and Strategic Trade-ofs reflects on the practical challenges, key insights, and strategic decisions involved in balancing agile development with regulatory demands. Section 7 Conclusion this section summarizes the case study’s key contributions and ofers guidance for startups pursuing certification of AI-powered medical technologies.
To acquire a thorough knowledge and methodology for the system architecture presented in this document, a structured approach was used to infer links between components based on typical data lfow patterns in IoT and healthcare systems.
While clinical evaluation were carried out in partnership with two university-afiliated hospitals in
Belgrade: The Pulmonology Clinic and the Institute for Infectious and Tropical Diseases. These studies
validated system performance in clinical environments, focusing on accuracy, user comfort, and alert
responsiveness. The findings were compiled into a Clinical Evaluation Report (CER), meeting MDR
Annex XIV requirements [
In our clinical trial, standard hospital gallium thermometer will be used along with our device by
applying comparative analysis. Clinical trial methodology was followed by ISO 14155 for planning and
execution, while all risk analyses were grounded in ISO 14971. Usability evaluations adhered to IEC
62366-1 [
As Baby FM incorporates AI features, additional focus was given to documentation of model
transparency and traceability of used methodology. Collaboration with an experienced Notified Body led to
improvements in explainability of AI logic, training validation, and output interpretation. This was key
for meeting MDR Article 61 and Recital 48, which stressed the importance of justifiable and interpretable
software behavior [
The past decade has seen a swift advancement in the incorporation of intelligent technologies within healthcare systems, particularly with the increasing presence of AI-powered wearable and implantable medical devices.
These technologies now facilitate various clinical functions, including early detection of diseases and ongoing patient monitoring, particularly in fields such as cardiology, oncology, pediatrics, and post-operative care. Although the potential benefits are widely recognized, these innovations also raise current issues regarding transparency, regulatory control, and integration into clinical practices.
To tackle these issues, international standards ofer a systematic basis for the design, validation, and
monitoring of products after they enter the market. ISO 13485 [
Wilmink et al. [
Similarly, Porbundarwala et al. [
To provide a better understanding of how quality and safety standards are implemented across various industries, a comparative analysis has been incorporated on the following Table 1.
It illustrates that, although ISO 13485, ISO 14971, and IEC 62304 are well-established in the medical device industry, other sectors such as pharmaceuticals, health IT, and industrial IoT focus on diferent elements like cybersecurity (ISO/IEC 27001) or data integrity. This diversity highlights the distinctive regulatory objectives and risk profiles of each sector.
The growing interconnection of AI technologies across various sectors has led both regulators and
researchers to reevaluate how healthcare software is classified. A significant development in this area is
the European Union’s Artificial Intelligence Act, which sets forth new requirements for AI systems based
on their purpose and associated risk levels. In the realm of healthcare, in regard to European Union
Artificial Intelligence Act for Healthcare [
2. Supplementary or companion software that aids a medical device without being essential to its operation 3. Independent software that executes a medical function on its own, identified as Software as a
Medical Device (SaMD)
Each of these categories necessitates a specific approach to compliance. Embedded systems need to comply with both the Medical Device Regulation (MDR) and IEC 62304, ensuring traceability throughout their lifecycle and adherence to device-level safety measures. Support software is often required to meet GDPR and ISO/IEC 27001 standards, particularly when it deals with sensitive health information, and must demonstrate usability and risk management through ISO/IEC 62366 and ISO/IEC 25010.
Software as a Medical Device (SaMD) faces the most rigorous regulatory oversight under the AI Act because of its ability to make autonomous decisions. It must be clearly transparent, auditable, explainable, and clinically validated prior to its implementation.
The policy report by AIHTA [
The classification of Baby FM under the European Union Medical Device Regulation (EU MDR) 2017/745
was based on the device’s intended use for continuous temperature monitoring and its potential
diagnostic impact. As outlined in Rule 10 of Annex VIII [
During the early stages of development, the Baby FM team mapped out its regulatory strategy to ensure full alignment with the General Safety and Performance Requirements (GSPRs) defined in Annex I of the MDR. A detailed checklist was developed to collect and track all compliance-related evidence, laying the groundwork for the Technical Documentation needed for CE marking and domestic market authorization via Serbia’s national regulatory body, ALIMS.
The technical documentation was compiled by MDR Annexes II [
Supplementary documentation included:
• A Software Bill of Materials (SBOM) covering all third-party and open-source components
according to NTIA [
This rigorous documentation efort resulted in a fully compliant technical file, enabling Baby FM to achieve both CE certification and national approval in Serbia. The approach demonstrates how structured regulatory planning, proactive documentation, and multidisciplinary collaboration can help startups achieve compliance in complex health technology markets.
Implementing a Quality Management System (QMS) was an essential measure for aligning Baby FM
with medical device regulations and ensuring traceability throughout the development process. From
the beginning, the team chose ISO 13485:2016 [
Recognizing the distinct challenges that early-stage startups encounter, the QMS was crafted to be both flexible and modular. This strategy enabled the team to progressively fulfill necessary regulatory milestones while still adhering to an agile, iterative development approach.
Central to Baby FM QMS was a comprehensive set of documentation, including: • Design and Development Plan (DDP) • Software Requirements Specifications (SRS) • Architecture and Modular Breakdown Document • Verification and Validation Plan (VVP) • Design History File (DHF) • Device Master Record (DMR) • Software Traceability Matrix and Risk Assessment Files
Every one of these artifacts was designed with traceability and readiness for audits in mind. The V-model was utilized to structure development stages, incorporating specific checkpoints that connect design inputs with verification results.
Baby FM closely followed the framework specified in IEC 62304 [
Risk analysis was performed in accordance with ISO 14971 [
Automated unit and integration testing were implemented to aid regression cycles, with coverage
metrics monitored via CI/CD pipelines [
In the following sections, we give the findings from the analysis of the system architecture described in the document. These findings demonstrate the structured interactions and data flows inside the B2C and B2B frameworks, as well as the administrative monitoring capabilities.
Model used in this case is gradient-boosted decision trees applied with XGBoost. Data is from the deidentified 15 pulmonology patients from the pilot study. Model evaluation involves main performance metrics on both training and validation datasets, including accuracy, precision, recall, F1 score, and AUC.
The UML class diagram eficiently depicts the relationships between essential components. This is emphasizing their responsibilities and connectedness in producing a cohesive solution. This results introduction prepares the groundwork for a thorough examination of the system’s functionality and performance, providing insights into its operational eficiency and scalability.
The Baby FM platform is a comprehensive health monitoring system aimed at facilitating high-frequency, non-invasive temperature monitoring for both pediatric and animal patients. Its design highlights the integration of biomedical sensor technology, secure mobile communication, and cloud-based artificial intelligence analytics.
The system architecture is presented on the Figure 1 and comprises three interrelated components: 1. Compact wearable sensor 2. Mobile application serving as a gateway and user interface 3. Cloud-hosted AI engine along with a data repository
Wearable Sensor collectData() sends data sends data
sends data via BLE
Mobile App displayData() sendDataToCloud()
Parent Interface viewChildData()
Sensor Hub/BLE Module transmitData()
Admin Dashboard viewAnalytics() manageSystem() uploads data retrieves data transmits data
accesses data and analytics
Cloud Backend API Database storeData() processRequests() syncs data
returns analytics sends data for analysis EHR Integration syncData()
AI Module Analytics Engine analyzeData() views analytics
The sensor module features a high-accuracy temperature sensor that meets ISO 80601-2-56 standards and is designed to measure core-equivalent skin temperature from one-second intervals with 0.1% accuracy. It utilizes Bluetooth Low Energy (BLE) for wireless communication, ensuring low power usage, which makes it ideal for overnight monitoring in both home and clinical settings. The casing of the sensor is biocompatible and safe for skin contact, while the firmware is equipped with self-check mechanisms and fallback error conditions in line with IEC 60601-1 electrical safety standards.
The application is intended for both healthcare professionals and caregivers. It shows real-time
readings, trend charts, and alert notifications triggered by AI-detected anomalies. The application
utilizes on-device encryption with AES-128 before sending any data. Additionally, it incorporates
anonymization procedures and consent workflows that comply with GDPR standards. Clinical setups
facilitate the integration with Electronic Health Records (EHRs) via a secure RESTful API and modules
compatible with HL7 FHIR [
The AI component operating on the backend utilizes a cloud infrastructure to interpret temperature
patterns in real time. The system for detecting anomalies combines supervised learning, which
involves clinician-validated labels, with unsupervised clustering methods to diferentiate between normal
variations and significant medical deviations. A key aspect of this system is its interpretability layer,
which depicts the AI’s decision-making process through SHAP (SHapley Additive exPlanations) values
and audit logs. This not only enhances clinical confidence but also meets regulatory requirements for
transparency in AI systems, as specified in the EU AI Act [
As illustrated in Figure 2, quantitative findings from a controlled clinical study conducted under ethical approval comparing Baby FM with a gallium thermometer, Baby FM achieved ±0.1°C accuracy in 91.2% of the cases.
We have incorporated clinical carer feedback gathered during pilot trials in clinical settings at hospitals afiliated with the University, even though formal veterinary usability studies are still in progress and will be part of future work. The main concerns raised by the feedback were automated alert trust, comfort during night time monitoring, and ease of use. Due to Baby FM’s continuous and non-invasive measurement, more than 80% of participating carers preferred it over conventional oral or axillary methods, according to the overwhelmingly positive informal feedback.
In Table 2, we can see the comparison between Baby FM to industry-leading commercial temperature monitoring systems like TempTraq®.
• Unit test coverage: 94% • Requirement-test traceability: 100% • Avg. change request resolution: 2.5 days • Monthly risk review updates since Q3 2024 These metrics help quantify our QMS eficiency and demonstrate adherence to continuous quality monitoring protocols.
The development of Baby FM, an AI-powered medical device for continuous body temperature monitoring, reveals a set of important insights for health technology innovators working in highly regulated environments. This case exemplifies how early integration of regulatory principles, quality assurance standards, and clinical usability requirements can serve not only as compliance measures but also as drivers of product maturity and stakeholder trust. The Baby FM team faced a number of significant obstacles throughout the development and certification process: • To integrate DevOps pipelines with ISO 13485 requirements, formal branching, tagging, and audit mechanisms must be established within CI/CD workflows. • AI transparency: Designing visualisations (SHAP plots, confidence scores) to reveal model logic was necessary to ensure clinical interpretability. • Risk analysis: It was necessary to map cloud delays, BLE dropouts, and false temperature alerts into risk tables that complied with ISO 14971. • Documentation control: Git-based QMS tools and digital signatures were used to manage document versioning and traceability because internal stafing was limited.
Integrating artificial intelligence into a medical device adds a distinct layer of complexity. It is not
suficient for the algorithm to perform accurately—regulators and clinicians alike now expect
transparency, traceability, and explainability. In response to this, the Baby FM development included built-in
interpretability tools that translated model behavior into clear clinical signals. Visualization dashboards
were used to display input-output relationships, anomaly confidence scores, and historical signal context.
This approach addressed concerns outlined in the MDR [
Instead of viewing compliance as a final step, regulatory planning was integrated into every development phase. The team utilized the General Safety and Performance Requirements (GSPRs) from the outset as a guide for aligning design activities, verification tests, and risk evaluations. The technical documentation was developed according to MDR Annexes II and III, while the use of the IMDRF Table of Contents ensured compatibility with international submission formats. Clinical validation eforts were grounded in ISO 14155 and ISO 14971, ensuring a continuous connection between pre-market data and post-market expectations.
Given the nature of Baby FM as a connected device handling personal health data, cybersecurity and
privacy were not treated as add-ons but as design imperatives. The system architecture incorporated
AES-128 encryption, device-level authentication, and GDPR-compliant anonymization policies from
the prototyping stage. These elements aligned with ISO/IEC 27001 [
Compared to other segments of the health technology field, such as pharmaceutical software tools or digital health services, the path to compliance for medical devices is significantly more demanding. These devices are expected to meet strict, layered requirements that span from initial design to long-term use in clinical environments. While some industries can apply quality and safety standards selectively, medical device developers must demonstrate full regulatory alignment across engineering, software safety, and patient risk management.
As illustrated in Figure 3, medtech remains the most consistent adopter of standards like ISO 13485, ISO 14971, and IEC 62304, reflecting a high level of traceability, system validation, and readiness for global approval pathways.
Medical Devices Health IT Pharma Industrial IoT 5
3 2 1 ISO 14971 5
The Baby FM experience confirms that small teams, even with limited resources, can navigate complex certification pathways when compliance is built into the product strategy—not retrofitted. The decision to invest early in quality systems, regulatory intelligence, and human-centered design significantly reduced rework and accelerated engagement with clinical partners and regulatory reviewers. As the company moves toward broader applications (e.g., oncology, fertility, veterinary use), the scalable foundation established through these strategies will support both AI life cycle governance and regulatory expansion.
The development of Baby FM illustrates how startups in digital health can successfully navigate regulatory and technical challenges by planning for compliance from the earliest stages.
Instead of separating quality assurance from design, the team adopted a unified development model that made safety, traceability, and documentation a natural part of the engineering workflow. In terms of technical performance, Baby FM achieved a temperature measurement accuracy of ±0.1°C, with a fever detection precision of 91.2% with the data extracted from the ongoing registered clinical study.
Baby FM shows that responsible development - grounded in standards, focused on usability, and aligned with regulatory logic - can help emerging health technologies succeed even in complex environments.
By relying on established international standards, the team was able to build a device that met both clinical and regulatory expectations. This foundation made certification processes more eficient and helped reduce delays during validation and review. Collaboration with medical professionals during early testing also ensured that the system addressed real clinical needs.
One of the system’s distinctive features is its use of interpretable artificial intelligence. Rather than relying on opaque algorithms, Baby FM provides clear logic paths and visual tools that help users understand how and why the device issues alerts. This transparency played an important role in the building of trust between healthcare professionals and regulatory reviewers.
Looking to the future, the core technology behind Baby FM is being extended to support new use cases. These include temperature monitoring in oncology patients, integration with reproductive health tools, and applications in animal care. Due to its modular structure, the system can be adapted to each setting with minimal redesign, allowing faster rollout and easier validation.
The company is also building a cloud-based service to connect its devices to hospital infrastructure. Using established data exchange protocols, this platform will enable health institutions to incorporate Baby FM into existing workflows, expanding its value as both a device and a data tool.
This case highlights the importance of integrating quality thinking into innovation, especially in AI-driven medical systems. As regulation becomes more demanding, particularly in areas involving machine learning, the ability to document and explain system behavior will be key to clinical acceptance and international growth.
In further projects, Baby FM will further integrate software engineering, cloud-based AI, secure mobile infrastructure, and biological sensing. Quality management systems will be in line with these procedures. The platform will then be layered with architecture that ensures clinical safety, durability, and regulatory compliance—demonstrating how software innovation for medical devices can be both agile and compliant.
In compliance with GDPR Articles 5 and 35, all patient data utilized in the creation and assessment of Baby FM was anonymized. Temperature information is encrypted and saved locally on the user’s mobile device when used at home. In clinical settings, information is sent to a hospital-controlled server via secure channels for a de-identified assessment of comfort, safety, and efectiveness. Ethical Committee of the Republic of Serbia granted ethics approval, and all clinical trials collected informed consent. Due to patient confidentiality and regulatory compliance, data are not made publicly available; however, synthetic datasets for research purposes may be made available upon request.
We thank Idvorski Laboratory for supporting EMC testing under IEC 60601-1 and Axiom International for providing preliminary precision reports showing 0.1°C accuracy. Clinical Trial data is being collected under NCT06447337 (https://clinicaltrials.gov/study/NCT06447337)