What is NeuroKit2, the World-Standard Library for Biosignal Processing

Heartbeat, electrodermal activity, respiratory rhythm, muscle contractions, brainwave fluctuations — these biosignals are the most direct reflection of the state of the human body and mind. Yet the raw data is buried under vast amounts of noise and human-specific artifacts, making advanced signal processing indispensable for extracting meaningful information. **NeuroKit2**, an open-source Python library, is becoming the world-standard answer to this challenge. It processes seven biosignal modalities — ECG (electrocardiogram), PPG (photoplethysmography), EDA (electrodermal activity), EMG (electromyography), EEG (electroencephalography), RSP (respiration), and EOG (electrooculography) — through a unified API, enabling clinical research-grade signal processing pipelines to be executed in as few as two lines of Python code. Released under the MIT license, it is downloaded approximately 93,500 times per month and has been cited in over 590 academic papers. For heart rate variability (HRV) analysis, it calculates 124 metrics in a single call; for complexity analysis, it provides 112 nonlinear indices (entropy, fractal dimension, Lyapunov exponents) — a level of comprehensiveness unmatched even by commercial software. The library includes 8 built-in R-wave detection algorithms and achieves an accuracy of 0.9761 on the UoG database. Lead developer Dominique Makowski (Assistant Professor at the University of Sussex and head of the Reality Bending Lab) maintains the design philosophy that "researchers and clinicians without extensive knowledge of programming or biomedical signal processing can analyze physiological data in just two lines of code." In an era where over 200 million Apple Watch users measure ECG and blood oxygen, Oura Ring earns top ratings for consumer-grade HRV accuracy, and Beacon Biosignals raises $86 million in Series B funding to invest in AI-driven EEG brain health biomarkers — NeuroKit2 serves as the foundational infrastructure bridging the raw data output from these wearable devices with the high-quality analytical results required in research and clinical settings. In Japan, TechDoctor raised 12 billion yen in Series B funding to deploy its digital biomarker platform "SelfBase," Chugai Pharmaceutical has integrated objective, continuous physiological data collection via wearable devices into multiple drug discovery projects, and the 2025 Medical Care Act revision formally established online medical consultation in law. This article comprehensively examines the fundamentals of biosignal processing, NeuroKit2's technical features, comparisons with similar tools, application cases, the wearable ecosystem, VC investment trends, challenges and limitations, and developments in Japan.

What is Biosignal Processing — Decoding the Data Emitted by the Body

Biosignals are a collective term for electrical, mechanical, and chemical signals generated by biological systems.

ECG (Electrocardiography) records the electrical activity of the heart. The waveform consisting of the P wave, QRS complex, and T wave is indispensable for diagnosing heart rate rhythm, arrhythmias, and myocardial ischemia. EEG (Electroencephalography) records the brain's electrical activity from the scalp and evaluates cognitive states and sleep stages across frequency bands of alpha, beta, theta, and delta waves. EMG (Electromyography) measures the electrical activity of muscles and is applied in motor control and rehabilitation. EDA (Electrodermal Activity) captures changes in skin conductance caused by sweat gland activity and serves as an indicator of autonomic nervous system arousal and stress responses. PPG (Photoplethysmography) optically measures changes in blood volume and is widely adopted in wearable devices such as the Apple Watch and Oura Ring. RSP (Respiration) records breathing patterns from airflow or chest wall expansion. EOG (Electrooculography) detects eye movements.

These raw data cannot be used as-is. They are buried in noise from electromagnetic interference, power line noise, motion artifacts (18.7% of ECG signal time is lost to motion artifacts), sensor failures, and transmission errors. Without signal processing steps — noise removal, peak detection, feature extraction, frequency decomposition, and nonlinear analysis — it is impossible to extract meaningful information from biosignals.

Traditionally, this processing relied on dedicated commercial software (AcqKnowledge, Kubios, LabChart) and MATLAB-based tools (Ledalab, PsPM). High licensing costs, black-box algorithms, and reproducibility constraints were ongoing challenges. The maturation of the Python ecosystem and the open science movement have enabled the rise of open-source libraries such as NeuroKit2.

NeuroKit2——Clinical Research-Level Signal Processing in 2 Lines of Code

NeuroKit2 is an open-source Python library developed by a research team led by Dominique Makowski (Assistant Professor at the University of Sussex). A paper was published in *Behavior Research Methods* in 2021 (Makowski et al., 2021), and the library is available on GitHub under the MIT license.

Its design philosophy is straightforward:

"Researchers and clinicians without extensive knowledge of programming or biomedical signal processing can analyze physiological data in just 2 lines of code."

In practice, ECG analysis is completed in the following 2 lines:

ecg = nk.ecg_simulate(duration=15, sampling_rate=1000, heart_rate=80)
signals, info = nk.ecg_process(ecg, sampling_rate=1000)

Supported Biosignal Modalities

NeuroKit2 supports 7+ modalities: ECG, PPG/BVP, RSP, EDA/GSR, EMG, EOG, EGG (electrogastrogram, under development), and EDR (ECG-derived respiration). While most competing libraries specialize in a single modality, NeuroKit2's greatest differentiator is its ability to process all modalities through a unified API.

ECG Processing Pipeline

The ecg_process() function performs cleaning, peak detection, and delineation (detection of P-waves, QRS complexes, and T-waves) in a single call. The R-wave detection algorithm includes 8 built-in methods: NeuroKit (default), Pan & Tompkins (1985), Hamilton (2002), Zong et al. (2003), Martinez et al. (2004), Christov (2004), Gamboa (2008), and Elgendi et al. (2010), achieving an accuracy of 0.9761 on the UoG database. 12-lead multi-channel ECG is also supported.

HRV (Heart Rate Variability) Analysis — 124 Metrics

Heart rate variability is an important biomarker reflecting autonomic nervous system activity. NeuroKit2 computes 124 HRV metrics in a single call.

Time-domain: RMSSD, MeanNN, SDNN, SDSD, CVNN, and more. Frequency-domain: ULF, VLF, LF, HF, VHF power, LF/HF ratio, LFn, HFn, LnHF. Nonlinear metrics: SD1, SD2, SD2/SD1, CSI (Cardiac Sympathetic Index), CVI (Cardiac Vagal Index), SampEn (Sample Entropy).

This comprehensiveness rivals even Kubios, the gold standard commercial software for HRV analysis.

Complexity Analysis — 112 Nonlinear Metrics

NeuroKit2's complexity analysis module provides 112 nonlinear metrics, including entropy-based measures (Shannon, Approximate, Sample, Fuzzy, Multiscale, Composite Multiscale), fractal dimensions (Sevcik, Katz, line length), Detrended Fluctuation Analysis (DFA), and Lyapunov exponents.

Makowski et al.'s study "The Structure of Chaos" (2022, MDPI Entropy) empirically compared these 112 metrics and demonstrated that 12 selected metrics explained 85.97% of the total variance across all metrics.

EDA Analysis

The tonic component of EDA (SCL: Skin Conductance Level) and the phasic component (SCR: Skin Conductance Response) are automatically decomposed, with peak detection and amplitude extraction performed automatically. A signal quality assessment function is also built in.

Comparison with Similar Tools — What Makes NeuroKit2 Different

The Python ecosystem for biosignal processing contains several major libraries.

BioSPPy is a multimodal library supporting ECG, RSP, EDA, EMG, EEG, and PPG, and is the most direct competitor to NeuroKit2. NeuroKit2 internally integrates BioSPPy's algorithms as an option, ensuring compatibility for BioSPPy users migrating to NeuroKit2.

MNE-Python is the industry standard for brain signals (EEG, MEG, sEEG, ECoG). It delivers state-of-the-art performance in source localization (estimating signal sources within the brain). MNE-Python is the first choice for brainwave analysis and has a complementary relationship with NeuroKit2.

HeartPy specializes in ECG and PPG and is designed for noisy field data. Its ability to run on Arduino and Teensy microcontrollers is unique, giving HeartPy an advantage for real-time embedded applications.

WFDB is PhysioNet's official library, specialized for data input and output. Developed by MIT-LCP, it is essential for accessing the PhysioNet database.

BrainFlow provides a unified SDK for biosensor hardware. It features a C++ core with Python, C#, Java, MATLAB, and Rust bindings, and excels at connecting with devices such as OpenBCI.

pyEDA specializes in EDA/GSR and has achieved 97% stress detection accuracy on the WESAD dataset.

NeuroKit2's uniqueness lies in its multimodal support capable of processing 7 modalities through a unified API, 112 complexity metrics, built-in comparison of 8 R-peak detection algorithms, and a low barrier to entry with as few as 2 lines of code. Having been benchmarked against competing libraries such as BioSPPy, HeartPy, Systole, and nolds, its comprehensive design philosophy — which incorporates algorithms from existing tools as options — underpins its status as a global standard.

Application Domains——The Future of Digital Health Opened by Biosignals

Emotion Recognition and Affective Computing

ECG, EDA, and PPG are reliable indicators of emotion. EEG has achieved 88.86% recognition accuracy across 4 emotions, while a multimodal approach combining ECG + EMG + biosignals recorded 79.3% across 4 emotional states, and ensemble deep learning achieved an F1 score of 90.96%. NeuroKit2 is widely used in these studies for biosignal preprocessing and feature extraction.

Digital Biomarkers and Remote Patient Monitoring

Digital biomarkers enable continuous physiological measurement outside clinical settings. The U.S. CMS (Centers for Medicare & Medicaid Services) accelerated adoption starting January 2026 by relaxing the minimum day requirement for Remote Patient Monitoring (RPM) from 16 days to 2–15 days. The role of standardized open-source tools like NeuroKit2 is growing in quality control and feature extraction of raw data collected from wearable devices.

The Wearable Health Ecosystem

Apple Watch (with over 200 million users) features FDA-approved ECG, atrial fibrillation history, blood oxygen, sleep apnea detection, and hypertension alerts (FDA-approved September 2025). Hypertension detection is achieved through optical heart rate sensors and ML, validated in over 2,000 participants. It is expected to notify more than 1 million previously undiagnosed users in its first year.

Oura Ring (Gen 4) has earned the highest HRV accuracy rating among consumer wearables and introduced the "Cumulative Stress" biomarker in the second half of 2025. WHOOP offers 26Hz heart rate capture and "Healthspan" metrics, with ECG and atrial fibrillation detection available on the MG variant. Ultrahuman is building an integrated ecosystem combining a ring, environmental monitor, blood testing, and AI interpretation.

NeuroKit2 is used by researchers as an analysis tool for PPG and accelerometer data output from these wearable devices to validate wearable algorithms. It serves as the infrastructure bridging raw consumer device data with research-grade analysis results.

Clinical Applications

Beacon Biosignals has enabled at-home neurological diagnostics with its FDA-approved Waveband EEG device. EpiWatch (a Johns Hopkins spinout) obtained FDA 510(k) clearance for seizure monitoring using Apple Watch. Behind these clinical AI applications, signal processing pipelines for EEG and ECG are essential, and open-source tools like NeuroKit2 provide the foundation for research and development.

VC Investment Trends — Capital Flowing into Digital Health

Key Investment Cases

Beacon Biosignals raised $86 million in its Series B in November 2025, bringing total funding to over $121 million. Innoviva, Google Ventures, Nexus NeuroTech Ventures, S32, Catalio Capital, and Takeda Pharmaceutical participated in the round. The company specializes in AI-driven EEG brain health biomarkers.

Japan's TechDoctor raised ¥12 billion in its Series B in May 2025, bringing cumulative funding to ¥18 billion. JAFCO, NVCC, and Sumitomo Mitsui Capital participated. The company operates "SelfBase," a digital biomarker development platform.

Macro-Level Investment Trends

a16z Bio + Health allocated $700 million in 2025, with 50% of a16z's AI projects in the healthcare sector. a16z and Eli Lilly established a biotech ecosystem fund of up to $500 million.

US digital health startups raised $14.2 billion in 2025 (up 35% year-over-year). Global digital health funding reached $22.3 billion in 2025 (up 19% year-over-year), with an average deal size of $20.3 million (up 29%).

The biosignal acquisition and processing systems market is projected to grow from approximately $2.8 billion in 2024 to $5.2 billion in 2033 (CAGR 7.1%). The wearable medical device market is expected to expand from $103 billion in 2025 to $505.3 billion in 2034 (CAGR 20%).

Challenges and Constraints

Motion Artifacts

Motion artifacts are the greatest challenge for biosignals collected by wearable devices. 18.7% of ECG signal time is lost to motion artifacts, leading to false tachycardia detection. Improving noise tolerance in everyday environments requires continuous improvement on both the algorithm and hardware fronts.

Real-Time Processing

NeuroKit2 is primarily designed for offline/batch analysis and is not optimized for real-time embedded applications. HeartPy (Arduino/Teensy compatible) and BrainFlow (C++ core) are better suited for real-time/embedded use cases.

Clinical Validation

Open-source tools often lack the formal clinical validation required for medical device status. Many digital biomarkers fail to advance from the research stage to regulatory approval. AI/ML Software as a Medical Device (SaMD) demands large, high-quality labeled datasets, and continuous learning systems face challenges with safety and efficacy regulations.

Privacy

Biosignals can enable re-identification of individuals. EEG data may disclose sensitive information. GDPR restricts secondary use of health-related data, and constraints on international data transfers make research collaboration difficult.

Japan's Trends — Digital Biomarkers and the Super-Aged Society

Japan faces structural challenges of the world's fastest aging population, accompanied by a rapid surge in healthcare costs and uneven distribution of regional medical services. Biosignal processing and digital biomarkers represent a promising answer to these challenges.

TechDoctor raised 12 billion yen in Series B funding and uses its digital biomarker development platform "SelfBase" to continuously collect and analyze biometric data (sleep, activity, pulse) from wearable devices. The company has a track record of joint research with over 100 research institutions and companies, and has partnered with IQVIA Japan on clinical research solutions.

Chugai Pharmaceutical has introduced objective and continuous physiological data collection via wearable devices across multiple drug discovery projects, advancing its digital biomarker program. Hitachi is developing portable, app-based, and wearable digital biomarkers, with a focus on health promotion and disease prevention in the elderly.

The 2025 revision of the Medical Care Act, which formally established online medical consultations in law, marks an important milestone that will accelerate the adoption of digital biomarkers integrated with remote patient monitoring.

A KAKENHI-funded project (2024–2029) is studying frailty assessment using digital biomarkers; the University of Tokyo has reported AI-based ECG analysis for detecting cardiac function decline; and the 2025 JSAI conference featured a presentation on myocardial infarction estimation via tensor ECG analysis.

Japan's wearable medical device market is projected to grow from approximately $2.2 billion in 2025 to $8.6 billion in 2034. The digital therapeutics (DTx) market is expected to expand from $150 million in 2026 to $1.13 billion in 2035 (CAGR 23.6%).

Future Outlook——The Day Biosignal Processing Becomes Standard Infrastructure

2026–2027: FDA-approved features in wearable devices such as Apple Watch, Oura Ring, and WHOOP are expected to expand, driving an explosive growth in consumer biosignal data. Demand for standardized open-source processing pipelines like NeuroKit2 will accelerate. The FDA's 2026 Wellness Device Update will clarify the boundary between fitness trackers and medical devices, establishing a more defined regulatory environment.

2028–2030: Multimodal biosignal analysis (simultaneous processing of ECG + EDA + PPG + RSP + EEG) will be integrated with AI foundation models, making real-time continuous monitoring of individual health status a practical reality. Digital biomarkers will be incorporated into drug development pipelines as FDA-approved companion diagnostics, and digital endpoints for clinical trials will become standardized.

Beyond 2030: Biosignal metadata will become standard accompanying data for digital content — much like subtitles and timecodes for video. NeuroKit2's 112 complexity metrics and 124 HRV metrics will be embedded directly into wearable device firmware and computed in real time on edge devices. Biosignal processing will transform from a "specialized technical skill" into "standard infrastructure."

Just as Hugging Face CEO Clement Delangue declared that "the era of local AI has arrived," biosignal processing is likewise migrating from the cloud to personal devices. NeuroKit2 is one of the foundational pillars supporting that transition, and its design philosophy of "clinical research-grade signal processing in two lines of code" stands as a symbol of democratization in this field.

Impact on the Industry

First, NeuroKit2 is establishing itself as the "jQuery" of biosignal processing — a standard library that abstracts complex low-level processing and makes it accessible to everyone. This is supported by 93,500 monthly downloads, over 590 academic citations, and adoption at major universities worldwide including Duke, the University of Washington, and the University of Auckland.

Second, the rapid growth of the wearable health device market (from $103B in 2025 to $505B in 2034, CAGR 20%) is structurally driving demand for biosignal processing libraries. The analytical needs generated by PPG, accelerometer, and heart rate data from Apple Watch's 200M+ users, Oura Ring, WHOOP, and Ultrahuman cannot be met without open-source tools like NeuroKit2.

Third, VC investment is concentrating in digital health. a16z's $700M Bio + Health allocation, Beacon Biosignals' $86M Series B, and TechDoctor's ¥12B Series B demonstrate that enormous capital is flowing into the intersection of biosignals and AI. NeuroKit2, which provides the standard tooling for biosignal processing, is growing in importance as foundational infrastructure for this ecosystem.

Fourth, digital biomarkers and remote patient monitoring offer structural solutions to Japan's healthcare challenges in its super-aged society — rapidly rising medical costs, geographic disparities, and a shortage of caregiving personnel. The 2025 Medical Care Act revision, which legally established online medical consultation, provides institutional support for this direction.

Fifth, open-source biosignal processing tools guarantee research reproducibility and transparency. The "publication and verification of algorithms" — impossible with commercial black-box tools — is made possible through NeuroKit2's GitHub repository and MIT License. This is also a significant contribution from the perspective of scientific methodology.

References: Makowski, D. et al. "NeuroKit2: A Python toolbox for neurophysiological signal processing" (Behavior Research Methods, 2021, DOI: 10.3758/s13428-020-01516-y), GitHub: neuropsychology/NeuroKit (2,200+ stars, MIT License), PyPI Stats: neurokit2 (~93,500 monthly downloads), Semantic Scholar: ~590 citations, Dominique Makowski (University of Sussex, Reality Bending Lab), NeuroKit2 ECG R-peak detection 0.9761 accuracy (UoG database), NeuroKit2 HRV 124 metrics, NeuroKit2 Complexity 112 indices, Makowski et al. "The Structure of Chaos" (MDPI Entropy, 2022), BioSPPy (GitHub: PIA-Group/BioSPPy), MNE-Python (mne.tools, BSD-3), HeartPy (paulvangentcom/heartrate_analysis_python), WFDB Python (PhysioNet), BrainFlow (brainflow-dev/brainflow), pyEDA (HealthSciTech/pyEDA, 97% stress detection WESAD), Kubios HRV (commercial gold standard), Apple Watch 200M+ users / ECG / blood oxygen / sleep apnea / hypertension FDA cleared (Sep 2025), Oura Ring Gen 4 HRV accuracy, WHOOP 26 Hz heart rate, Ultrahuman Ring + Blood Vision, Beacon Biosignals $86M Series B (Innoviva, Google Ventures, Takeda, Nov 2025), TechDoctor ¥12B Series B (JAFCO, NVCC, May 2025), a16z Bio + Health $700M, a16z + Eli Lilly $500M Fund, US digital health $14.2B (2025, +35%), Biosignal acquisition market $2.8B→$5.2B 2033 (CAGR 7.1%), Wearable medical devices $103B→$505B 2034 (CAGR 20%), Motion artifact 18.7% ECG signal loss (Nature Reviews Bioengineering, 2024), FDA 2026 wellness device update, CMS RPM minimum reduced 16→2-15 days (Jan 2026), EpiWatch FDA 510(k) (Johns Hopkins), Chugai Pharmaceutical digital biomarkers, Hitachi digital biomarkers, TechDoctor SelfBase / IQVIA Japan partnership, 2025 Medical Care Act revision legalizing online medical consultation, KAKENHI digital biomarker frailty assessment (2024-2029), University of Tokyo AI ECG cardiac function decline detection, Japan wearable medical $2.2B→$8.6B (2034), Japan digital therapeutics $0.15B→$1.13B (CAGR 23.6%)