Integrating AI ECG Analysis from Wearables into Your Mobile App

Integrating AI ECG Analysis from Wearables into Your Mobile App

We help integrate AI analysis of single-lead ECG from wearable devices—Apple Watch Series 4+ or AliveCor KardiaMobile—directly into your mobile app. The recording lasts 30 seconds at a 512 Hz sampling rate, which suffices for detecting atrial fibrillation and other rhythm disturbances. Your task: capture the signal, process it correctly, and apply the model without false negatives or regulatory pitfalls. Our engineers are certified in medical standards and have 5+ years of experience in mobile health projects. Every decision affects diagnostic accuracy and user safety. We guarantee compliance with FDA and CE requirements.

Developing a medical ECG analysis app involves technical challenges: noisy signals from on-body devices, limited phone computing power, stringent accuracy requirements, and battery life constraints. Without proper processing, even the best deep learning model will produce false positives on every third measurement. We have accumulated experience integrating with HealthKit and BLE devices to avoid these pitfalls.

How to Obtain ECG Data from Wearables?

Apple Watch + HealthKit

Apple Watch Series 4+ writes ECG via HKElectrocardiogram. Access through HealthKit:

import HealthKit

func fetchLatestECG() async throws -> (HKElectrocardiogram, [Double]) {
    let ecgType = HKObjectType.electrocardiogramType()
    let query = HKSampleQuery(
        sampleType: ecgType,
        predicate: nil,
        limit: 1,
        sortDescriptors: [NSSortDescriptor(key: HKSampleSortIdentifierStartDate, ascending: false)]
    ) { _, samples, error in
        // handle
    }

    // For voltage data — separate subquery
    let voltageQuery = HKElectrocardiogramQuery(ecg) { _, result in
        switch result {
        case .measurement(let measurement):
            if let voltage = measurement.quantity(for: .appleWatchSimilarToLeadI) {
                let microvolts = voltage.doubleValue(for: .volt()) * 1_000_000
                voltageData.append(microvolts)
            }
        case .done: break
        case .error(let error): print(error)
        }
    }
    healthStore.execute(voltageQuery)
}

Apple returns raw signal in volts (Lead I), sample rate 512 Hz, about 15360 samples over 30 seconds. The code above demonstrates raw voltage data retrieval. Important: handle errors and ensure the user has granted ECG reading permission.

Third-party BLE Devices + SDK

AliveCor KardiaMobile, Withings Move ECG, and similar devices transmit ECG via BLE. Most provide SDKs: KardiaMobile SDK, Withings SDK. Without an SDK, reverse-engineering the protocol is lengthy and unreliable. We have implemented over 15 integrations with HealthKit and BLE devices, ensuring stable operation. When working with AliveCor KardiaMobile, a common issue is BLE synchronization—we developed a reconnection protocol with exponential backoff.

Why Signal Preprocessing Matters?

Raw ECG signals contain noise: baseline wander, muscle artifacts, 50/60 Hz interference. Without filtering, the model will produce false positives.

A 0.5–40 Hz bandpass filter removes major noise:

from scipy.signal import butter, filtfilt

def bandpass_filter(signal, fs=512, lowcut=0.5, highcut=40):
    nyq = fs / 2
    low = lowcut / nyq
    high = highcut / nyq
    b, a = butter(4, [low, high], btype='band')
    return filtfilt(b, a, signal)

R-peak detection (the Pan-Tompkins algorithm) is needed to compute RR intervals—the primary feature for arrhythmias. On mobile, the filter can be implemented natively: in Swift via the Accelerate framework, in Kotlin via KotlinDL or a port of SciPy code. Beyond filtering, motion artifacts must be removed—we use an adaptive drift estimator based on a median filter with a 200 ms window.

Which Models are Effective for Arrhythmia Detection?

CNN for Rhythm Classification

The standard approach is a 1D CNN on signal windows of 2.5–10 seconds. Input tensor: [batch, time_steps, 1]. Architecture: Conv1D → MaxPool → Dense → Softmax. Public datasets: PhysioNet MIT-BIH Arrhythmia Database (48 two-channel recordings), PTB-XL (21,799 clinical ECGs).

After training, convert to CoreML:

import coremltools as ct

spec = ct.convert(
    torch_model,
    inputs=[ct.TensorType(shape=(1, 2560, 1), dtype=np.float32)],
    compute_precision=ct.precision.FLOAT16,
    compute_units=ct.ComputeUnit.CPU_AND_NE
)
spec.save("ECGClassifier.mlpackage")

FLOAT16 quantization halves the model size without significant accuracy loss. CoreML on Apple Silicon processes ECG 3x faster than TFLite on Android with ML Kit. For model validation on mobile platforms, we use public datasets and our own test recordings. Quality evaluation includes sensitivity and specificity metrics at the 95% level for atrial fibrillation.

Limitations of On-Device Analysis

Single-lead ECG from the wrist is not a clinical 12-lead. Apple's algorithms report AFib sensitivity of 99.3%, but other disturbances require full diagnostics. Therefore, it is important to position the app as a screening tool, not a replacement for physician diagnosis.

What Regulatory Requirements Must Be Considered?

If the app claims diagnostic capability, it is a medical device (CE Class IIa, FDA 510(k)/De Novo). Development requires clinical trials and design control. If positioned as informational, the regulatory burden is lower, but disclaimers are mandatory. Apple itself shows that every ECG result is accompanied by a warning. We help gather the necessary documentation for CE or FDA as part of a preliminary audit.

Our Work Process

What's Included

• Integration and regulatory documentation
• Demo app with analyzer
• Model training for your team
• Post-launch support (3 months)

Timeline and Investment Estimates

A basic AFib detector with CoreML and HealthKit — 3–5 weeks (investment discussed individually). An extended system supporting multiple arrhythmias — 2–3 months (investment depends on complexity). A full medical device is a separate project. On-device analysis can save up to 40% compared to cloud processing.

Contact us to evaluate your project—we'll propose the optimal architecture and timeline. Order turnkey AI ECG analysis development with quality guarantees and standards compliance. Get a consultation on integrating an AI model into your app.

Machine Learning in Mobile Apps: CoreML, TFLite, and On-Device Models

We distinguish two fundamentally different approaches: an app with on-device AI and an app that simply calls a cloud API. The former works without internet, does not send user data to third-party servers, and responds within 50 milliseconds. The latter depends on network latency and pricing plans. Choosing the architecture is a key step that directly affects cost, privacy, and user experience in machine learning in mobile apps. Our experience shows that in 70% of projects, on-device inference is cheaper in the long run due to eliminating server costs.

How to Choose Between CoreML and TFLite for On-Device Inference?

CoreML — Apple's native framework for running ML models on device. Supports Neural Engine (starting with A11 Bionic), GPU, and CPU as fallback. Models are converted to .mlmodel format via coremltools from PyTorch, ONNX, or TensorFlow. Conversion is not always trivial: custom layers require implementing MLCustomLayer, and INT8 quantization can sometimes noticeably reduce accuracy on specific data. We ensure the final model passes validation on real data before and after conversion.

TensorFlow Lite — cross-platform alternative for Android and Flutter. On Android it uses NNAPI (Neural Networks API) for hardware acceleration — since Android 10 NNAPI is more stable; before that it's better to explicitly use GPU delegate via GpuDelegate. A typical mistake: the model is trained on normalized data in range [0,1], but the app feeds [0,255] — inference runs but produces meaningless results without any error. We include an automatic input data validation module in the SDK.

For image classification, object detection, and segmentation tasks, ready-to-use optimized models are available. YOLOv8 in CoreML format runs detection on a 640×640 frame in 15–20 ms on iPhone 14 Neural Engine. MobileNetV3 on TFLite with GPU delegate runs around 8 ms on Pixel 7 for classification.

What If You Need Text Generation On-Device?

Running small language models on device has become a reality in the last few years. Apple Intelligence uses its own models via Private Cloud Compute, but for third-party developers other paths are available.

llama.cpp with Metal backend on iOS is a working approach for phi-3-mini (3.8B parameters, 4-bit quantization, ~2.3 GB). Inference: 15–25 tokens/second on iPhone 15 Pro. For integration in Swift, use the Swift Package llama.swift or a wrapper via C interface llama.h. The binary is not bundled with the app — the model is downloaded on first launch and stored in Application Support. Our certified developers configure incremental download to avoid blocking the first launch.

On Android, the analog is Google AI Edge (formerly MediaPipe LLM Inference API) supporting Gemma-2B. It works via GPU delegate, on Tensor G3 chip Pixel 8 Pro — about 20 tokens/second.

Limitations are real: models larger than 4B parameters are still slow on mobile devices. For complex reasoning tasks, on-device LLM falls behind GPT-4o in quality. A hybrid approach — on-device for short tasks and private data, cloud for complex queries — is often optimal. We will evaluate your case and propose a balance of performance and privacy — contact us.

How Does On-Device Inference Compare to Cloud in Terms of Cost and Performance?

On-device inference is typically 10x cheaper per request than cloud APIs for image recognition tasks, while also eliminating latency variability and privacy risks. The table below summarizes the trade-offs.

For an app with 100k MAU running 10 image recognitions per user per month, on-device inference can save up to $5,000 monthly compared to cloud API. Get a free consultation on your ML architecture today.

Integrating OpenAI API and Other Cloud Models

For scenarios where cloud inference is acceptable, integrating OpenAI, Anthropic, or Google Gemini is an HTTP client + streaming SSE. In Swift, AsyncThrowingStream is convenient for streaming responses. In Kotlin, use Flow.

Critically: API keys must never be stored in the app bundle. Even an obfuscated key can be extracted from the IPA in 10 minutes using strings or frida. Correct architecture: mobile app → your own backend → OpenAI API. The backend controls rate limiting, logs requests, and protects the key.

What Is Included in the Work (Deliverables)

• Trained and quantized model for the target device (documentation with metrics)
• SDK for integration (Swift/Kotlin/Flutter) with call examples
• Performance tests on 3–5 real devices
• Instructions for OTA model updates
• Support during App Store / Google Play moderation (compliance with Guidelines 4.2, 5.1)
• 2 weeks of technical support after release

Typical Project Pipeline

1. Task analysis — measure latency, privacy, size, supported devices.
2. Model prototyping — in Python, evaluate accuracy on target data.
3. Conversion and quantization — for CoreML/TFLite with validation.
4. Integration into the app — model wrapped in a service layer (easy to swap CoreML ↔ TFLite ↔ cloud).
5. Testing — on real devices, measure FPS, RAM, battery.
6. Deployment — via TestFlight / Firebase App Distribution, monitor metrics.

Timelines: integration of a ready CoreML/TFLite model — 1–2 weeks, development of a custom model with mobile optimization — from 6 weeks, on-device LLM chat with personalization — 4–8 weeks.

Why We Take on Complex Cases?

10+ years of experience in mobile development, 50+ implemented AI/ML solutions, guarantee of compatibility with current iOS and Android versions. All projects undergo code review and load testing. The cost includes preparation of moderation documentation and training of your team.

Contact us — we will help you choose the architecture and implement ML in your app turnkey. Order an audit of your existing solution — we will assess the potential for server cost savings free of charge. In some projects, savings can reach significant amounts per month.