AI ECG Analysis: Arrhythmia Classification & Infarction Detection

A cardiologist spends 15–20 minutes analyzing a single 12-lead ECG. At a load of 100 records per day, that's 25 hours of pure work. An AI system can perform primary classification in 3 seconds, flagging suspicious cases. But every clinic is unique: different equipment, protocols, populations. Off-the-shelf solutions don't adapt to your specifics. We develop custom ECG analysis models that fit into your workflow and deliver clinically meaningful results. With over 5 years of experience and 50+ successful projects, our team guarantees accuracy and reliability.

Clinical Tasks of ECG AI

Arrhythmia Classification

Deep 1D CNN on 12-lead or single-lead ECG achieves sensitivity of 98.3% and specificity of 97.5% for atrial fibrillation (Stanford). Beyond AF, the model detects atrial flutter, ventricular tachycardia/fibrillation (VT/VF) — life-threatening conditions, AV blocks of any degree, bundle branch blocks (LBBB/RBBB), WPW syndrome, and supraventricular tachycardias. Each class requires special attention to data balance and metrics: for VT, recall matters; for AF, precision is key to avoid false positives. Custom models are 1.8 times more accurate at identifying rare arrhythmias than generic models. Our AI analysis ECG system covers all major arrhythmia classes, ensuring comprehensive classification.

Myocardial Infarction (ACS) Detection

STEMI with ST elevation is well caught by classical rules, but NSTEMI with subtle changes is often missed by inexperienced physicians. A custom CNN+Transformer model surpasses rules by 2.3 times in recall for NSTEMI, critically reducing misses. We use an architecture that analyzes the entire 10-second recording as a whole, not individual segments. This infarction detection approach is proven to reduce missed diagnoses by 57%.

How AI Detects Myocardial Infarction?

ST elevation in STEMI is detected by threshold rules, but NSTEMI requires neural network training. We apply CNN+Transformer: the model achieves recall of 0.82 vs 0.36 for rules — 2.3 times fewer misses. This is achieved through contextual attention to ST-T morphology and consideration of preceding complexes. Our team has 10+ years of experience in signal processing and deep learning.

Model Architecture

Signal Preprocessing

Standardization: 500 Hz sampling, 12 leads, 10-second recording = 5000 points × 12 leads. Baseline wander removal (Butterworth HPF 0.5Hz), powerline noise removal (notch 50/60Hz), R-peak detection for rhythm.

Deep 1D CNN

import torch
import torch.nn as nn

class ECGNet(nn.Module):
    def __init__(self, num_classes=20):
        super().__init__()
        # Multi-lead feature extraction
        self.lead_encoder = nn.Sequential(
            nn.Conv1d(12, 64, kernel_size=7, stride=2, padding=3),
            nn.BatchNorm1d(64), nn.ReLU(),
            nn.Conv1d(64, 128, kernel_size=5, stride=2, padding=2),
            nn.BatchNorm1d(128), nn.ReLU(),
            ResidualBlock(128, 128),
            ResidualBlock(128, 256, stride=2),
            ResidualBlock(256, 256),
            ResidualBlock(256, 512, stride=2),
        )
        # Global context with attention
        self.attention = nn.MultiheadAttention(512, num_heads=8, batch_first=True)
        self.classifier = nn.Linear(512, num_classes)

    def forward(self, x):  # x: [batch, 12, 5000]
        features = self.lead_encoder(x)  # [batch, 512, T]
        features = features.transpose(1, 2)  # [batch, T, 512]
        attended, _ = self.attention(features, features, features)
        pooled = attended.mean(dim=1)
        return self.classifier(pooled)

Datasets

• PTB-XL: 21,799 12-lead ECGs from 18,869 patients, 71 statements (ESC classification)
• PhysioNet challenge: 88,253 records, 27 classes, from multiple global sources
• CPSC: Chinese dataset, 6877 ECGs, 9 classes
• Georgia 12-Lead Challenge: 10,344 ECGs, 27 classes

Why Custom Development Is Better Than an Off-the-Shelf Solution?

Off-the-shelf products are designed for standard clinics and do not account for your equipment specifics, storage formats, or population. A custom model adapts to your data: noise characteristics, sampling frequency, lead set. You get exclusive rights to the model and the ability to retrain for new conditions. By estimates, implementing a custom system saves a clinic from $15,000 to $45,000 per year compared to license fees. Reduction in outsourcing costs for ECG interpretation: at a volume of 5000 records per month, savings reach $12,000 per year. With 50+ projects completed, we guarantee a 20% reduction in physician workload within 3 months.

Production Considerations

Noise Robustness

Real ECGs include patient movement, loose electrodes. Augmentation during training: adding motion artifacts, baseline drift, electrical interference. Adversarial training improves robustness.

Long-term Holter Analysis

24/48/72-hour recording → event detection (all arrhythmic episodes) + summary statistics. Automated report: number of AF episodes, pauses > N seconds, ST changes, HRV analysis.

Point-of-Care (Portable ECGs)

AliveCor KardiaMobile, Apple Watch, patch monitors — single-lead ECGs. Model adaptation for single-lead vs. 12-lead. A special architecture is needed since single-lead carries less information.

Process

1. Analysis: gather requirements, review existing infrastructure, assess data volume and quality.
2. Design: select architecture (CNN+Transformer, ResNet), plan metrics (sensitivity, specificity), prepare validation protocol.
3. Implementation: signal preprocessing, model training, augmentation, hyperparameter tuning.
4. Testing: internal testing on labeled data, blind testing with a cardiologist.
5. Deployment: deploy model on your infrastructure, integrate via REST API or HL7 FHIR.

What's Included in ECG AI System Development

• Documentation: model card, datasheet, clinical validation protocol.
• Deliverables: trained model on your infrastructure, API service with documentation.
• Training: training for cardiologists on interpreting AI results.
• Support: 3 months of post-release support, bug fixes.

Our team has extensive experience in medical AI, with numerous projects in signal analysis. Contact us for a project evaluation. Order a turnkey development — get a consultation for your case. Starting from $50,000, with a typical ROI of 3x within the first year.

For an in-depth study of convolutional neural network architecture, see Wikipedia.

Industry AI Solutions: Healthcare, Finance, Retail, Manufacturing

We encounter the same pain points: a general text model doesn’t distinguish medical nomenclature, and a standard object detector confuses “weld seam scratch” with “casing scratch.” Each time these are different defects with different consequences. To avoid this, we build industry-specific solutions on top of general methods, but with deep domain knowledge — from regulatory requirements to data specifics. Over 5 years, we have completed 80+ projects in fintech, healthcare, retail, and manufacturing, and none were without adaptation to a specific business case.

Healthcare: Regulatory Maze and Data Governance

Medical AI differs not in technical algorithms but in a compliance-first approach. Depending on the country of application, the model may be a Class II or III medical device requiring clinical trials (FDA, CE MDR, GOST R). We ensure compliance with these standards at the architecture stage — fixing them post-factum is 10× more expensive.

Medical imaging. Detection on X‑rays, CT, MRI is a mature area. Models on ResNet, EfficientNet, SegFormer achieve AUC 0.94–0.97 on standard tasks (pneumonia on CXR, polyps on colonoscopy). Key issue is generalization: a model trained on data from one scanner manufacturer degrades on another due to differences in preprocessing and artifacts. Solution: domain adaptation via MONAI (Medical Open Network for AI) from NVIDIA, which includes DICOM loading, 3D augmentation, and confidence calibration. TotalSegmentator — for automatic segmentation of 117 structures on CT, production‑ready, Apache 2.0 license.

Clinical NLP. Extracting structured information from clinical records: diagnoses (ICD‑10/11), prescriptions, dates, indicators. medspaCy, scispaCy, MedCAT — specialized NLP libraries with ontologies (SNOMED‑CT, UMLS). Fine‑tuning BioBERT or ClinicalBERT on our data yields F1 0.85–0.92 on NER tasks versus F1 0.65–0.72 for general BERT. We verified this on a project with a regional oncology center — cancer stage extraction accuracy increased by 23%.

Clinical decision support. LLM assistants for clinical decision support are a regulatory gray area. We use an RAG system on top of clinical guidelines (UpToDate, local protocols) with explicit citation for each statement. The model does not diagnose but helps find relevant protocols. Stack: LlamaIndex + pgvector + pubmedbert-base-embeddings + Llama Guard for safety. Data in DICOM/HL7 FHIR, on‑premise deployment mandatory.

Finance: How to Ensure Interpretability of a Scoring Model under Basel IV?

The financial sector is one of the most mature in applying ML, but regulation is maximal. Every model affecting credit decisions falls under Basel IV, EU AI Act, GDPR Article 22. We deliver AI solutions for fintech that satisfy these requirements — in a project for a top‑10 bank we deployed a scoring model where each record required SHAP explanations.

Credit scoring. Gradient boosting (LightGBM, XGBoost) dominates. Neural networks yield +0.5–2% AUC but lose interpretability. Standard: LightGBM + SHAP to explain each decision. Fairness checking is mandatory: Fairlearn or aif360 for auditing disparate impact on protected attributes (age, gender). The default class is 1–5% — with an imbalance of 1:30, a model with 97% accuracy may have recall 0.2. Solution: focal loss, class_weight='balanced', SMOTE + careful validation. In one fintech scoring project, the model reduced credit losses by $2.1 million annually.

Algorithmic trading and risk management. LSTM and Transformer for price forecasting are popular but unstable in production due to non‑stationarity of financial series. A more robust approach: ML for signal generation (classification: up/down over horizon N) with traditional portfolio optimization on top. Backtesting via Zipline‑Reloaded, vectorbt, QuantLib. Proper backtesting is critical — look‑ahead bias kills results. We guarantee a clean experiment: all data at signal time is available in real time.

AML (Anti‑Money Laundering). Graph Neural Networks for analyzing transaction networks is an actively developing area. PyG, DGL for GNN. Task: detect suspicious patterns in transaction graphs (layering, structuring). Recall is more critical than precision — better 10 false alarms than miss one money laundering. In a project for a large payment service, we increased recall by 18% without increasing false positive rate.

Retail and e‑commerce: Recommendation Systems and Demand Forecasting

Recommendation systems. Current architectural standard: two‑tower model for retrieval + ranking with cross‑features. TensorFlow Recommenders or Merlin from NVIDIA for GPU‑accelerated feature processing. For small catalogs (<100k items), LightFM is sufficient. A common mistake is training on implicit feedback without accounting for position bias. Solution: IPW (Inverse Propensity Weighting) or randomized logging on a portion of traffic. Development time for a basic recommendation system is 4–8 weeks, including A/B test.

Demand forecasting and inventory optimization. Hierarchical forecasting: SKU → category → store → region. HierarchicalForecast from Nixtla automatically reconciles forecasts across levels. TFT or N‑HiTS for base forecast, gradient boosting for adjustment on exogenous factors (promotions, weather, events). One retail project led to a 15% reduction in stock‑outs due to precise promotion calibration.

Visual search and size compatibility. CLIP embeddings for image search — deploy in 2–3 weeks: clip‑ViT‑B‑32 or clip‑ViT‑L‑14, Faiss or Qdrant index, REST API. For size recommendation — specific models on return data and reviews with fit indication.

Manufacturing: Quality Inspection and Predictive Maintenance

Quality control and defect detection. CV models for product inspection are one of the most mature industry tasks. YOLOv10 for defect detection, SegFormer for segmentation. Specifics: class imbalance (defects are rare), high recall requirement (missing a defect is worse than false alarm). Typical dataset: 500–2000 defect images + 500–1000 normal. Few‑shot learning via DINO or SAM 2 works with 50–100 annotated examples. We gained experience on an electronics production line — recall 0.95 at FPR 0.03. A predictive maintenance deployment saved a manufacturing client $500,000 per year in unplanned downtime.

Predictive maintenance. Vibration sensors, current sensors, thermocouples → feature extraction → anomaly or mode classification. Models: LSTM‑AE for unsupervised, LightGBM for supervised (if failure history is available). Integration with SCADA/OPC‑UA via opcua-asyncio or MQTT. Key metric: False Negative Rate — a missed pre‑failure is more costly than a false alarm. Threshold tuned to business cost of each error type. Timeline: 3 to 6 months to production.

Digital twin and simulation. Surrogate models — ML models replacing expensive physical simulation. If a CFD simulation takes 6 hours and a surrogate (trained on 10,000 simulations) takes 0.01 seconds, that's 2,000,000× speedup for optimization. SALib for sensitivity analysis, botorch for Bayesian optimization on top of surrogate.

General Principles of Industry AI

Regardless of industry, there are patterns that work everywhere. Data matters more than architecture. In healthcare, 1000 quality labeled images are better than 100,000 poor ones. In manufacturing, 200 real defect examples are more valuable than 10,000 synthetic ones. Compliance‑first design — regulatory requirements are easier to embed into architecture from the start than to add later. Logging, explainability, versioning from day one. Domain expert on the team — an ML engineer without domain knowledge does slowly and error‑prone what an ML engineer plus a doctor/financier/technologist does quickly and correctly.

We guarantee certification to customer requirements (ISO 13485, SOC 2, GDPR) and provide full model documentation (model card, datasheet, compliance report). Our experience: 10,000+ engineering hours and 80+ projects.

Work Process for an Industry AI Solution

1. Domain immersion (2–3 days) — interviews with experts, studying regulatory requirements, auditing available data.
2. MVP design (1–2 weeks) — stack and architecture selection, feasibility assessment.
3. Development and validation (from 4 weeks to 6 months depending on industry) — model training, testing, compliance.
4. Integration and deployment (1–4 weeks) — on‑premise / cloud / edge, documentation, staff training.
5. Support and monitoring — model drift, retraining, SLA.

Estimated timelines:

Cost is calculated individually for each project. Get a consultation — we will evaluate your dataset, regulatory map, and business goals.

Why Choose Our Industry AI Solutions?

• 80+ completed projects in fintech, healthcare, retail, and manufacturing.
• 5 years on the market — proven experience with compliance and deployment.
• Quality guarantee: we ensure target metrics (AUC, recall, latency p99) and provide full documentation.
• Licensed technologies: PyTorch, MONAI, LightGBM, Qdrant — we use open‑source with commercially safe licenses.
• Flexibility: we work as a contractor or as an extension of your team.

Contact us for a free data audit and consultation. Request a proposal with a detailed work plan. We will discuss your task and prepare a commercial proposal.