Tamponade is an unusual word. The radiologist was looking at the images, he said “I wish they would have told me he had tamponade.”
Cardiac tamponade happens when the fluid sac around your heart fills with blood or other fluid, putting pressure on your heart. Because of the pressure, your heart can’t fill up with blood, reducing how much blood your heart can pump and ultimately causing a drop in blood pressure. Trauma and certain diseases can cause cardiac tamponade. If not treated, it is always fatal.
What happens in cardiac tamponade?
Inside your chest, your heart is contained within a sac called the pericardium. The pericardium has two layers and holds your heart in place, and it also contains pericardial fluid. The fluid helps cushion your heart from outside force and stress.
The chambers of your heart must expand to fill up with blood before pumping blood out. Cardiac tamponade – which is sometimes called pericardial tamponade – happens when the pericardium fills with fluid (especially blood). Because the fluid has nowhere to go, your heart runs out of room and can’t expand enough to fill effectively.
SYMPTOMS AND CAUSES
What causes fluid to surround and compress the heart?
Several different types of injuries can cause cardiac tamponade. For example, it can happen with blunt trauma, such as a fall or a car crash, or penetrating trauma, like a stab wound from a knife.
It can also happen as a result of another injury, such as an aortic dissection. This happens when a tear forms in the wall of the aorta, a major artery in the center of your body. Blood can accumulate in between layers of the aorta’s walls and cause the artery to rupture, filling the pericardium with blood.
The AI system was trained to look for aortic dissection and aneurysms. We all knew that the large provider of reading services had AI that would find and diagnose AAA. This patient with tamponade presented in the emergency department and was quickly sent to CT for a AAA study.
The technologist responded quickly and sent the urgent message to the radiologist that the study was for AAA. This would mean the textual classification as an AAA would elevate the study to the top of the worklist and result in a STAT interpretation.
The patient went down hill fast. The emergency department called the radiologist urging them to read the exam immediately. This call was not received by the radiologist. It was taken by a non-medically trained “facilitator” who took the doctor’s name and number with the promise to call them back.
The patient post CT scan was in the emergency department without an interpretation and died waiting for care.
The results of the study were provided two hours later. A loop of bowel had herniated into the heart area. This bowel was pushing on the heart and slowly causing the demise of the patient.
Why did the AI not save this patient?
Why did the CT technologist do everything right for the patient and still have to feel like she somehow let the patient die under her care? She entered the correct information to cause an urgent/immediate reading of the CT.
The answer is the AI is trained on the images on the CT screen not on the textual information.
If the image showed an AAA then the exam would be prioritized for immediate reading.
It did not!
Artificial intelligence answers the question correctly and tirelessly – if you ask it the right question and you understand its limitation and how to use it.
This is the challenge in imaging right now. We have it but are we doing it well? Has it been deployed well? Is there an ROI for solving this problem?
As this tamponade patient’s heart slowly failed because of the force the bowel loop was placing on it we should think about the failure of the system that was designed to help.
We should think about the expectation we have for AI in the future.
We should understand that to use AI in imaging we must know the opportunity and limitations.
We must ask the correct AI questions; patients lives depend on it. •
Mark Watts is an experienced imaging professional who founded an AI company called Zenlike.ai.
