Skip to main content

PainCubeTurning pain into measurable data for earlier clinical decisions.

StellarCube is developing PainCube, a clinical decision support solution that uses multimodal biosignals and AI to quantify and predict pain.

Joint research withJohns Hopkins Medicine · Korea University Anam Hospital

PainCube is at the stage of developing and validating its algorithms on biosignal and medical-record data collected with its research partners.

Explore the technologyView research collaboration

SOLUTION

PainCube solution in detail

  • 01

    Quantifying pain and predicting it ahead of time

    Unlike assessment that rests on what a patient can describe, PainCube analyses biosignals in real time to quantify pain, with the goal of predicting it 30–60 minutes before onset.

    BiosignalsPain index, early prediction

  • 02

    Integrated analysis of multiple biosignals

    Integrates electrocardiography (ECG), photoplethysmography (PPG), oxygen saturation (SpO₂), respiratory rate, heart rate and body temperature to keep watching for pain even in patients who are unconscious or unable to communicate.

    Multiple biosignalsContinuous observation

  • 03

    From reactive to preventive care

    PainCube moves pain management away from responding after pain has set in, toward predicting pain and caring for it in advance.

    Pain predictionPreventive pain care

WHY PAINCUBE

Pain has been left unmeasured

Intensive care still has no objective way to measure pain, and no precise standard for when and how much analgesic to give.

  • No objective way to measure pain, and no precise standard for analgesic dosing
  • Pain measured from the patient's own subjective report, against imprecise criteria
  • Analgesics prescribed on clinical intuition

Postoperative pain not adequately managed

80%

Source · Gan, J Pain Res 2017

A dimly lit intensive care bed beside a vital-sign monitor
Postoperative pain is reported to be inadequately managed in more than 80% of patients in the US (Gan, J Pain Res 2017), with rates varying by type of surgery, the analgesic or anesthetic method used, and time elapsed after surgery.Illustrative imageone icon = 1 in 10 patients

HOW IT WORKS

How PainCube works

Biosignals are gathered, noise is stripped out, pain is quantified and predicted, and the result reaches the care team. The four stages below carry the same numbers as the sections that follow.

PainCube runs on two AI models. DDCAE uses an encoder–decoder structure to detect and remove noise that can be mistaken for pain, such as ECMO operation or body movement, and TCAtt-PainNet learns biosignal patterns preceding pain onset and pain-reduction patterns after analgesic administration to quantify and predict pain.

  1. 01. Sensor

    A sensor worn where each patient allows — chest, upper arm or wrist

  2. 02. DDCAE

    Removal of intensive care noise

  3. 03. TCAtt-PainNet

    Pain anticipated from biosignal and pain patterns

  4. 04. CDSS

    Dashboard-based pain alerts and administration support

Step 01

SENSOR + DATA

Multimodal biosignals and clinical data

PainCube collects six or more biosignals — electrocardiography (ECG), photoplethysmography (PPG), oxygen saturation (SpO₂), respiratory rate, heart rate and body temperature — together with six or more types of medical-record data, including age and sex, diagnosis, surgical history, pain scores (NRS), medication and analgesic administration records.

Biosignal data

  • Electrocardiography (ECG)
  • Photoplethysmography (PPG)
  • Oxygen saturation (SpO₂)
  • Respiratory rate
  • Heart rate
  • Body temperature

Medical record data

  • Age and sex
  • Diagnosis
  • Surgical history
  • Pain scores (NRS)
  • Medication
  • Analgesic administration records

Biosignal acquisition

Acquiring

ECGmV

PPGa.u.

RESPa.u.

Also measured

HR60bpm
SpO₂97%
TEMP36.8°C

Wearing position

  • 1Chest patch
  • 2Upper-arm patch
  • 3Wrist band
Concept screenInterface mock-up. The values shown are not real patient data or verified performance figures.
Step 02

SIGNAL DENOISING AI

DDCAE

Extracting the pure pain signal with a noise-removal AI algorithm

  • Encoder–decoder based, using how regular pain is against how irregular noise is
  • Detects and removes noise mistaken for pain, such as ECMO operation and body movement
  • Extracting the pure pain signal resolves the false-alarm problem

Signal denoising

Denoising

Raw signal

ECGmV

Powerline interference

PPGa.u.

Motion artifact

RESPa.u.

Baseline drift

DDCAE

Denoised signal

ECGmV

PPGa.u.

RESPa.u.

Noise in the raw signal

  • Powerline interference
  • High-frequency noise
  • Motion artifact
  • Baseline drift
Concept screenInterface mock-up. The values shown are not real patient data or verified performance figures.
Step 03

PAIN QUANTIFICATION & PREDICTION AI

TCAtt-PainNet

Pain quantification, and prediction 30–60 minutes before onset

  • Learns biosignal patterns preceding pain onset, and pain-reduction patterns after analgesic administration
  • PainCube expresses the result of its biosignal analysis as a pain index (CPI) on a 0–10 scale.
  • PainCube analyzes biosignals in real time to quantify pain, and is being developed with the goal of predicting pain 30–60 minutes before onset.
  • Secures the golden hour for prescribing and administering analgesics

Pain index forecast

Forecasting
Current pain index
1.8
Predicted pain index
7.4
1050
  • Observed
  • Forecast
  • Alert level

The further ahead the forecast reaches, the wider its range.

Concept screenInterface mock-up. The values shown are not real patient data or verified performance figures.
Step 04

CDSS

Clinical decision support system

Data-grounded pain management and analgesic administration through monitoring and decision support

Clinical decision support

Action needed

Jiwoo HanF / 62Fictional patient

Bed12

ProcedureLaparoscopic colectomy

CoursePOD 1

Bed list

  • Bed 033.2
  • Bed 072.7
  • Bed 12Alert1.8

Pain increase predictedBed 12

7.4/ 10

Predicted pain indexCurrent pain index 1.8

A rising pain pattern appears in the denoised biosignals. This patient cannot self-report pain.

Probability of this forecast91%

Basis

  • Heart-rate variability pattern
  • Falling pulse amplitude
  • Rising respiratory rate

Pain index trend

No intervention

-30 minNow+30 min

Risk level
High
Recommendation
Consider preemptive IV analgesia

Vitals

HR
60bpm
NIBP
128/74mmHg
SpO₂
97%
RESP
15rpm
TEMP
36.8°C

Recent events

  • 09:41Pain increase alert raised
  • 09:32Rising pain index observed
  • 09:18Denoising applied
  • 09:05Monitoring started
Concept screenInterface mock-up. The values shown are not real patient data or verified performance figures.

  • Alerts 30–60 minutes before pain onset

    Secures the golden hour to respond to pain

  • Data-grounded support for analgesic decisions

    Including the type and timing of administration

  • Real-time patient monitoring dashboard

    Pain management tailored to each patient

CLINICAL IMPACT

Streamlining the whole pain-management process

Published figures put pain management at 34 minutes per patient per 12-hour shift (SCCM PADIS 2018; Gelinas et al, 2006). By streamlining routine rounds, pain assessment and administration, PainCube is estimated to reduce this to about 7 minutes — roughly 80% less.

Pain management time per patient

34min

Before

7min

After

80%lessEstimate

Conditions
per 12-hour shift · per patient
Baseline source
SCCM PADIS 2018 · Gelinas et al, 2006

JOINT RESEARCH

Joint research

PainCube is being developed through joint research with Johns Hopkins Medicine and Korea University Anam Hospital.

Data collection covers 32 beds at Johns Hopkins Medicine and 24 beds at Korea University Anam Hospital, spanning more than 12 types of biosignal and medical-record data.

  • Johns Hopkins Medicine

    Johns Hopkins Medicine

    Data collection32 beds

  • Korea University Anam Hospital

    Korea University Anam Hospital

    Data collection24 beds