Can Artificial Intelligence Interpret a Heartbeat?
As a general practitioner, I am increasingly seeing patients coming in with reams of paperwork on their symptoms, medical history and what online chatbots like ChatGPT think the diagnosis could be. Whilst I am all for patient education and empowerment, I must give warning about privacy. Where is your precious data being stored and what is it being used for?
This is why I use Diagnosis Pad — an offline AI medical assistant.
Privacy concerns aside, I also have concerns about the accuracy of such sites. I previously did a study on the dangers of using ChatGPT and found some alarming responses when I presented it with some common clinical presentations.
Often, these sites can also be politically motivated. Recently, I gave ChatGPT a simple set of symptoms a typical patient with ADHD (Attention Deficit Hyperactivity Disorder) would present with. The only change I made was the gender. ChatGPT said it wouldn’t be ADHD if it was a man, but definitely ADHD if it was a woman. Shocking.
Diagnosis Pad
I wanted to test out just how good Artificial Intelligence (AI) can be when it comes to interpreting sensor information (such as a heart beat) and whether this changes its thought process. For this mini experiment, I’m using an offline application called Diagnosis Pad. Being offline, there is no patient data being sent or processing to servers or clouds, maintaining the security and privacy of sensitive information. The application itself acts as a tool to the doctors bag. It uses ambient audio to listen to the conversation between the doctor and patient and analyses the information using its medical AI. With this, it then presents the user, in real-time, a list of possible differential diagnoses and recommended questions to further diagnose.
Case Study — Mr Gregory
The case that I chose was a 64 year old man, Mr Gregory, who had been experiencing symptoms of shortness of breath and occasional dizziness. Now these two symptoms on their own could amount to a number of different differential diagnoses. But for this particular study I was aiming to get a diagnosis of Atrial fibrillation, a condition where the heart beats fast and out of sync.
From the image above, you can see the transcript that was provided as well as the initial list of differential diagnoses. The AI came up with a primary differential of Pulmonary Embolism (PE), whilst this was not what I was after, I was still pleasantly surprised. A Pulmonary embolism is where a patient has a clot in the lungs causing them to be short of breath and can lead to death if not treated rapidly. It may also present as dizziness, as a result of being short of breath. So this in itself is a viable diagnosis. Interestingly, Atrial fibrillation did appear as the second most possible diagnosis and thirdly Chronic Obstructive Pulmonary Disease (COPD). So even though the AI was presented with very limited information, it came up with some reasonable diagnoses.
The recommended questions would also have been useful in a real life situation. In the case of a presumed Atrial fibrillation, naturally an Electrocardiogram (ECG) would be ordered to assess the rhythm of the heart. By giving the practitioner a prompt in the right direction it may help those who are less experienced.
Sensor Information Interpretation
I wanted to see what would happen when I input sensor information, such as a heart beat. How will it change the differentials?
After role-playing a stethoscope examination, I presented my findings to the ‘patient’ and AI.
Reassuringly, it updated its differential diagnoses and the primary was that of Paroxysmal Atrial fibrillation.
Now that we were both in sync on the diagnosis, I wanted to see why the AI thought of this particular condition. The app has an inbuilt medical terminology database, so if the clinician is unsure why the AI has thought of something, it’s easy to find out using the contextually aware explanation.
Although the AI got the diagnosis correctly, it still alerted me, as a the clinician, to proceed to a full examination and other investigations such as an electrocardiogram (ECG) or echocardiogram. Of course, these would be done in this clinical scenario before such a diagnosis was made.
Conclusion
The ability of the app to interpret the sensor information and change the differential diagnosis in real-time was impressive. This opens the pathway for the interpretation of other sensor information such as blood pressure monitors, blood glucose monitors and even ECG interpretations. This could then be useful in areas where medical aid is inaccessible or limited.
However, for now, artificial intelligence is still exactly that, artificial. It is therefore vital to still see a doctor in person for an appropriate check up for any ailments. But it may be something that doctors can incorporate in their care to help to improve patient outcomes.
What do you think about AI in healthcare, is it a fad or will it be around to stay? Let me know below.
I hope you have found this article helpful.
Take care and stay healthy,
Dr Nora x
Try the application for free and if you have any questions or comments, drop a line to the developers here.
Please note that the contents of this article are not intended to be a substitute for professional medical advice and should not be relied on as health or personal advice. Always seek the guidance of your doctor or other qualified health professional with any questions you may have regarding your health or medical condition.
Links to products often include an affiliate tracking code which allow us to earn fees on purchases you make through them.
