Advantages

- Expert-Level Accuracy: Built on an AI model trained with high-quality supervised data strictly labeled by sleep research experts. Enables ultra-high-precision stage classification that virtually matches visual scoring by experienced specialists.
- Natural Classification Results: Equipped with a novel algorithm for analyzing EEG complexity alongside a proprietary correction algorithm that catches pathological stage transitions. Delivers high accuracy while reliably detecting abnormal sleep architecture.
- High Hardware-Independent Versatility: Resistant to device-specific noise and artifacts. Delivers stable classification across diverse measurement environments regardless of the manufacturer or type of amplifiers, sensors, or devices used.

Background and Technology

In sleep research and drug discovery, scoring sleep stages (REM sleep, NREM sleep, and wakefulness) based on EEG and EMG data is essential. However, the field still relies heavily on visual scoring by experienced experts, presenting challenges in efficiency and objectivity due to human error, inter-scorer variability, and heavy workloads. Although automated scoring technologies have recently emerged, questionable results often force researchers to resort to manual verification. To address these challenges, Professor Takeshi Sakurai and his team at the International Institute for Integrative Sleep Medicine (WPI-IIIS), University of Tsukuba, developed "S.A.C." (Sleep Analyzer Complex)—a sleep stage classification AI. S.A.C. was trained on high-quality supervised data compiled through years of research, combined with feature extraction via "complexity analysis". It automatically classifies sleep stages with exceptional accuracy rivaling expert visual inspection. This dramatically lightens the burden on analysts while maintaining high analytical quality, promising to accelerate sleep research.
Key technical features of S.A.C. include:
- Complexity Analysis: By applying complexity analysis (*) to EEG data split into five frequency bands, S.A.C. extracts irregularity and self-similarity in EEG signals as sleep stage features—aspects overlooked by conventional methods. (*Multiscale Entropy for irregularity and Detrended Fluctuation Analysis / Detrended Fractal Analysis for self-similarity)
- Catching Pathological Abnormalities: Equipped with a proprietary correction algorithm, S.A.C. refrains from forcibly "normalizing" unnatural stage transitions caused by disease or disorders. This enables accurate, high-precision capture of pathological sleep architecture.
- Validated Training Data: Utilizes a highly reliable dataset rigorously evaluated and labeled by experts at WPI-IIIS, University of Tsukuba.

Current Stage and Key Data

**Current Stage**
Research software has been developed and is currently utilized in study protocols using laboratory animals (mice and rats).
**Key Data**
- In evaluations using EEG and EMG data from wild-type mice, the AI achieved an exceptionally high concordance rate of over 98% compared to expert visual scoring.
- REM sleep is notoriously difficult to classify automatically and has long posed a challenge for existing AI models. S.A.C. dramatically improved REM sleep recall by approximately 2.3 percentage points compared to conventional models (e.g., SPINDLE-like models) where performance had plateaued.
- In tests using disease-model mice with disrupted sleep architecture (such as narcolepsy models), S.A.C. achieved a breakthrough by accurately identifying abnormal direct transitions from wakefulness to REM sleep (SOREM) with high precision.
- For a 24-hour dataset from a single mouse, the analysis time—which previously required hours of manual labor—was drastically reduced to 90 minutes of automated processing plus 15 minutes of visual check.
- Analysis of EEG data recorded from multiple measurement hardware manufacturers showed virtually no variance in classification results, confirming the broad hardware compatibility and versatility of this technology.

Expectations

The University of Tsukuba is seeking corporate partners interested in commercializing this technology. Potential business and product applications include:
- Integration into EEG Hardware and Analysis Software for Animal Research: Embedding this AI into existing biosignal acquisition systems or software suites adds automated, highly reliable sleep stage scoring, significantly increasing product value.
- Application in Sleep-Related Pharmaceutical R&D: Automating and streamlining large-scale EEG data analysis in drug discovery accelerates development timelines, cuts labor costs, and eliminates human error.

Patent and software licensing options are available for companies interested in commercialization. During implementation and product development, the University can provide necessary expertise and consultation through joint research agreements. Please feel free to contact us for further details or preliminary discussions.

Principal Investigator

Prof. Takeshi Sakurai (International Institute for Integrative Sleep Medicine [WPI-IIIS], University of Tsukuba)

Patents and Publications

Patent Application: WO2025/013824A1 (PCT/JP2024/024606)
Publication: Utility of complexity analysis in electroencephalography and electromyography for automated classification of sleep-wake states in mice (https://www.nature.com/articles/s41598-024-74008-0)