For Nigerian diagnostic centres, hospitals and HMOs
Take a photo of the paper ECG. Seconds after it arrives, your doctor has a first read to check: the heart rate, the key intervals and the patterns it sees. Your doctor still decides.
01
Nigeria has about 400 cardiologists (heart doctors) for more than 220 million people, according to the Nigerian Cardiac Society. Many ECGs are read by doctors who are not heart specialists, or they wait for one.
02
Printouts sit in a pile until a heart doctor is free. Patients wait for a result or come back another day. The doctor on duty reads alone.
03
You photograph the printout. Seconds after the photo arrives, TraceRead AI has rebuilt the 12 tracings, measured them and listed the patterns it sees, each with a score from 0 to 100.
04
Your doctor. TraceRead signs nothing, and nothing reaches a patient until a doctor has checked and signed it.
Heart doctor numbers: President of the Nigerian Cardiac Society, quoted in The Guardian (Nigeria).
A real first read, replayed
The numbers are TraceRead's own measures of three public research ECGs, printed and photographed by our simulator. They replay as the phone reads each page.
Every step of a real runRun T-0057, from PTB-XL (Germany)
The AI first read
Real runs of TraceRead's code on public research ECGs from PTB-XL. Hover a lead to see how closely the red rebuilt line follows the original recording.
Numbers under each lead: how closely the rebuilt line follows the original recording, where 1.00 is a perfect match. Median for this page: 0.96.
PTB-XL record 15863, a man aged 68. A4 paper, 3x4 with one lead II strip, pulses read 4 of 4.
| Measure | From the photo | Original recording | Reference program |
|---|---|---|---|
| Heart rate (beats a minute) | 63 | 63 | |
| PR interval (ms) | 136 | 137 | |
| QRS width (ms) | 88 | 87 | |
| QT interval (ms) | 406 | 396 | |
| QTc, corrected for heart rate (ms) | 414 | 404 | |
| Heart axis (degrees) | 20 | 19 |
Sinus rhythm
the usual rhythm, started by the heart's natural pacemaker
Left ventricular hypertrophy
the muscle wall of the main pumping chamber may be thicker than usual
Unspecific reduced blood flow pattern
beat changes that can go with reduced blood flow to the heart muscle, without a clear place
Reduced blood flow pattern, front and side
beat changes over the front and side of the heart that can go with reduced blood flow to the heart muscle
ST depression
the middle part of the beat sits lower than usual
Unspecific wide QRS
the signal spreads slowly inside the lower chambers, so the main spike is wide
Left atrial enlargement
the upper left chamber may be larger or under more strain than usual
What the dataset's cardiologists recorded for this ECG
Case studies
What TraceRead does
For each one: what it does, what code checks without relying on the AI, and what a person does.
Stage 1
On the phone, before anything is sent
The phone checks the picture about five times a second for blur, glare, poor light and a page that does not fill the frame, and says what to fix.
Before sending, the technician drags a box over the printed header where a machine prints the patient's details.
Photos wait safely on the phone and send themselves when the network comes back. Each photo is kept near 1.5 MB.
Stage 2
From photo to 12 rebuilt tracings
TraceRead AI finds the four corners of the paper and how it is turned, even in a photo taken at an angle.
TraceRead AI finds the calibration pulse at the start of each row and works out where each of the 12 leads sits on the page.
TraceRead AI follows the ink of every lead and rebuilds it as a digital signal, 500 points a second.
Marks exactly which seconds of each lead the paper shows, because most printouts show each lead for only 2.5 seconds.
When the rebuilt trace looks unreliable, TraceRead gives no first read and asks for a new photo.
Stage 3
From tracings to an AI first read
TraceRead AI checks the tracing for 71 patterns of rhythm, conduction, thickened heart muscle and ST and T changes, each with a score from 0 to 100.
Heart rate, PR interval, QRS, QT, QTc and heart axis, measured on all the leads together.
How it works
Step 1 of 9, case study T-0057
Step 1
The technician photographs the printout with a phone. The phone checks sharpness, glare, light and framing first.
What goes in
What comes out
Step 2
TraceRead AI finds the four corners of the paper and how it is turned, then fixed code measures the printed grid.
What goes in
What comes out
3x4 with one lead II strip, pulses read 4 of 4
Step 3
The calibration pulse at the start of each row tells where each of the 12 leads sits and how tall 1 millivolt is.
What goes in
What comes out
Step 4
TraceRead AI follows the ink of every lead and turns it into a digital signal, 500 points a second, drawn back in red.
What goes in
What comes out
Step 5
A printout shows most leads for only 2.5 seconds. The coverage map marks exactly which seconds each lead has.
What goes in
What comes out
The layout match score was 0.981 (a photo is held back when the score is below 0.888) and 4 of 4 calibration pulses were read. The AI's own uncertainty about the five main heart groups was 0.306 (a photo is held back above 0.497). Neither check fired, so the photo is used.
Step 6
A fixed rule checks the pulses, the layout match and cut leads. If the rebuild looks unreliable, it asks for a new photo.
What goes in
What comes out
Step 7
Fixed code measures heart rate, PR, QRS, QT, QTc and heart axis on all leads together.
What goes in
What comes out
Sinus rhythm flagged
Left ventricular hypertrophy flagged
Unspecific reduced blood flow pattern flagged
Reduced blood flow pattern, front and side flagged
ST depression flagged
Step 8
TraceRead AI scores 71 patterns of rhythm, conduction, thickened muscle and ST and T changes, and flags those past their line.
What goes in
What comes out
A doctor checks every line.
The AI first read never reaches a patient on its own. Do not rely on it alone in an emergency. A normal AI first read does not rule out heart disease.
Step 9
The doctor checks the red line against the paper, every number and every flag, then decides. Nothing reaches a patient before that.
What goes in
What comes out
At a glance
Every figure says how many ECGs it was measured on and where they came from: public research ECGs from Germany, China and Brazil, and photos of printed ECGs made by TraceRead's photo simulator.
0.908
Trace match
Median match between the rebuilt trace and the original recording, 500 photos made by computer.
20/29
Patterns from China
Patterns scoring 0.90 or more on 45,152 ECGs TraceRead AI never trained on.
7.85 s
Time to read
95 in 100 ECGs read within this on a 2 vCPU server in AWS Cape Town, 97 photos.
16.8 ms
QTc difference
Average difference from a reference ECG program on 2,198 test ECGs.
Who uses TraceRead
Diagnostic centres
The ECG technician, right after the machine prints
The printout goes into a pile for the heart doctor, who may read it hours or days later. The patient waits, or comes back.
Hospitals
The doctor on duty who is not a heart specialist
An ECG is printed on the ward and there is no cardiologist on site to ask.
HMOs
The medical officer checking a claim that includes an ECG
A claim says the ECG showed a problem, and all you have is a photo of the printout.
Safety and privacy
An AI first read is not the last word.
How patient data is handled
Pricing
Every plan includes every feature on this site. Your own doctors review and sign.
Pay as you go
₦1,500
per AI first read, about $1.10
Centre
₦120,000
a month, about $88
HMOs and groups
Volume price
for many sites, priced per AI first read
Dollar amounts are approximate, at ₦1,370 to the dollar.
Questions
You take a phone photo of a printed 12 lead ECG. TraceRead AI finds the page, rebuilds each of the 12 tracings as a digital signal, measures it and lists the patterns it sees, each with a score from 0 to 100. A doctor then checks that AI first read.
No. TraceRead gives an AI first read for a qualified doctor to review. Nothing reaches a patient until a doctor has checked and signed it. A normal AI first read does not rule out heart disease, and TraceRead must not be relied on alone in an emergency.
Printed 12 lead ECGs from adults (18 and over), at the usual 25 mm a second and 10 mm per millivolt, laid out as 3x4, 6x2 or 12x1, on A4 or US Letter paper.
No. The capture app runs in the phone's web browser. It checks each photo before it is sent and keeps photos on the phone until the network is back.
From public research ECGs from Germany (PTB-XL), China (Chapman Shaoxing Ningbo) and Brazil (CODE-15%), and from photos of printed PTB-XL ECGs made by TraceRead's own photo simulator. Each number on this site says how many ECGs it was measured on.
No. No generative AI writes clinical text. Every pattern name comes from a fixed list, and every measurement is a number from fixed code.
Patient data stays in the AWS Cape Town region, and it is never sent to an outside AI service. TraceRead needs only the patient's age and sex, never a name. TraceRead staff cannot see patient data by default.
₦1,500 per AI first read, or ₦120,000 a month for 100 reads and ₦1,200 for each read after that. HMOs and groups with many sites get one volume price.
We show you TraceRead reading public research ECGs, step by step, and talk through how your team prints, photographs and reads ECGs today.