See how the Queen of Hearts AI is continuously improving

Written by Willy Frick

A man in his early 40s with BMI 36, hypertension, and a 30 pack-year smoking history presented with three days of chest pain. It started while he was at rest after finishing a workout. He described it as a mild intensity, nagging pain on the right side of his chest with nausea and dyspnea. It woke him the next day and radiated into his back. He was only able to sleep while sitting in a chair. He went to urgent care and had an ECG (not available) which was interpreted as normal, and was sent home. His pain returned, and he went back to the urgent care but was sent to the ER. His ECG is shown:

What do you think?

If this ECG looks familiar to you, you have a great memory. It is ECG 2 from this case a few months back.  At the time, the ECG was interpreted as not OMI with high confidence by the Queen of Hearts.

There was only one AI model available at the time, simply called the Queen of Hearts. But the armamentarium is increasing, and there are now multiple AI models. The version that was first introduced is now referred to as "eOMI." eOMI is shorthand for "entirety of OMI." Her training set included active OMIs as well as reperfused OMIs (e.g. Wellens waves).

The geniuses at Powerful Medical have been hard at work developing newer models to improve diagnostic performance. The second model introduced is known as "aOMI," which stands for "active OMI." The aOMI training set does not include reperfusion tracings, only ECGs which correspond with active, ongoing OMI. It is easy to see how this model will have improved specificity. But interestingly, I have also been seeing a lot of cases where the model shows improved sensitivity for subtle active OMIs.

Here is Queen of Hearts interpretation according to the new aOMI model:

As you can see from the old post, this was in fact TIMI 0 proximal RCA OMI. Here was the timeline for this case:

• 1809: ECG 1 obtained
• 1849: hsTnI 5548 ng/L
• 1922: ECG 2 obtained, diagnostic for active OMI
• 2054: hsTnI 10497 ng/L
• 2248: NTG SL tab
• 2152: Nitro paste
• 2245: Morphine 4 mg IV
• 2248: hsTnI 17809 ng/L
• 2250: Heparin infusion
• 2331: Nitroglycerin infusion
• 0112: Morphine 4 mg IV
• 0445: Morphine 4 mg IV
• 0424: hsTnI 25736 ng/L
• 0518: cath lab activated

TEN HOURS from time of first diagnostic ECG until the patient was discovered to have a completely occluded proximal RCA.

Imagine the myocardium that could have been salvaged if aOMI had been available and paid attention to.

===================================

MY Comment, by KEN GRAUER, MD (12/22/2024):

===================================

There is a phenomenon known as, "intra-observer" variability — in which the same ECG is resent to the same interpreter at a spaced interval, with goal of seeing how much this interpreter's analysis of that same ECG may have changed. In years past — the figure I am familiar with is ~10-20% variability by the same interpreter when asked to interpret the same ECG at a later time (ie, weeks-to-months later — such that the interpreter won't remember the tracing).

• I find this result humbling — with my personal goal of always striving to minimize variability in any of my interpretations for critical findings.
• Of course another reason why providers may interpret a 2nd reading of the same ECG differently that 2nd time around — is that there may have been learning since seeing that tracing the 1st time.

In today's case — it is QOH who has learned, as her next version ( = aOMI ) was successful in recognizing the acute findings in today's tracing, whereas an earlier version was not successful.

• I did not see this case that Dr. Frick previously presented in the March 13, 2024 post of Dr. Smith's ECG Blog. So, this was my 1st viewing ...

MY Thoughts on Today's ECG:

Even without the updated aOMI version of QOH — there are 2 leads in ECG #1 that should immediately capture your attention:

• My "eye" was immediately drawn to the subtle changes in lead aVL (within the RED rectangle in Figure-1). These include ST segment flattening with slight ST depression and terminal T wave positivity.
• Clearly — the tiny size of the QRS complex in lead aVL makes recognition of the above changes challenging. But the point to emphasize — is that it is the shape of the ST segment in lead aVL that cannot be interpreted as "normal" in a patient with new chest pain. Learn to recognize this shape ...
• Support that the ST segment flattening in lead aVL is indeed abnormal — is forthcoming from the flattened ST segment, and near flattening of the T wave in lead V2 (within the BLUE rectangle). As we have emphasized often in this Blog — leads V2 and V3 normally manifest slight, upward sloping ST elevation. The shape of the ST-T wave in lead V2 indicates loss of the normal ST-T wave appearance in lead V2 — which in association with the abnormality in lead aVL in this patient with new chest pain indicates recent/acute OMI until proven otherwise.
• NOTE: Others in our group immediately recognized the subtle ST-T wave abnormalities in today's tracing (as described in detail by Dr. Frick in the March 13, 2024 post). With experience — Your "eye" can learn to recognize this shape!

QUESTION:

The lead II rhythm strip shows regular sinus rhythm (RED arrow P waves) — until beat #9, which is a PVC.

• Did YOU Notice that the PR interval of the next P wave has increased? (ie, the PINK arrow P wave in Figure-1). 
•     — Does this increase in PR interval indicate Mobitz I AV block?

Figure-1: I've labeled the initial ECG in today's case.

ANSWER:

There is no AV block in Figure-1.

• Instead of AV block — the reason the PR interval before beat #10 is increased — is that beat #9 is an interpolated PVC. If the timing of a PVC is "just right" — the PVC may occur sandwiched in between 2 normal sinus beats without the usual compensatory pause. When this happens — there may be "concealed" retrograde conduction from this PVC that delays arrival of the next on-time sinus P wave. (See the April 9, 2020 post for more on interpolated PVCs, including laddergram illustration).
• While patients without heart disease may also have PVCs — the finding of this PVC in today's patient who presents for chest pain, and who manifests abnormal ST-T waves in 2 leads (aVL and V2) — should add to suspicion that an acute event may be in progress.