Alright, folks, we have some fascinating news in the world of artificial intelligence (AI) and machine learning. These tools are booming in various industries, including healthcare. But hold your horses, because a recent study suggests that when it comes to analyzing X-rays, AI just can’t compete with human radiologists.
This study, published in the journal Radiology, gathered 72 radiologists and pitted them against four commercial AI tools. The task at hand? Interpreting 2,040 chest X-rays from older adults, with an average age of 72. It turns out that about a third of these X-rays displayed diagnosable conditions like airspace disease, pneumothorax (collapsed lung), or pleural effusion (aka “water on the lung”).
Now, the AI tools did show some promise. They were reasonably sensitive in diagnosing these conditions, hitting accuracy rates ranging from 62% to 95%. Not too shabby, right? Well, here’s the kicker: these AI tools also produced quite a number of false positives, especially when the diagnosis got more complicated. When it came to pneumothorax, the AI systems had positive predictive values ranging from 56% to 86%. On the other hand, those radiologists nailed it 96% of the time. That’s some impressive human accuracy right there.
The story doesn’t get any better for AI when it comes to pleural effusion. Their positive predictive values were around 56% to 84% accurate, while the radiologists maintained their 96% accuracy. And let’s not forget about airspace disease. AI only got it right in 40% to 50% of cases. That’s not gonna cut it, my friends.
Dr. Louis Plesner, a resident radiologist and lead study author from Denmark, highlighted the challenges AI faces. He stated, “AI systems seem very good at finding disease, but they aren’t as good as radiologists at identifying the absence of disease, especially when the chest X-rays are complex.” Plus, the high rate of false positives from AI could result in wasted time, unnecessary testing, and increased radiation exposure for patients. That’s a costly business, my friends.
Zee Rizvi, the co-founder of Odesso Health, an AI-assisted medical records service, agrees with the findings. He believes that AI should complement human skills, not replace them entirely. We’re just not far enough along in the AI and deep learning space to kick humans out of the equation. It’s as simple as that.
Dr. Fara Kamanger, a dermatologist and the founder of AI skin health tool DermGPT, also chimed in on the study. She praised its robust design, involving multiple AI tools and radiologists to confirm diagnoses. Kamanger sees immense potential for AI in healthcare, from drug development to patient care. She acknowledges that AI won’t replace human physicians anytime soon, as they have the advantage of a comprehensive clinical evaluation and the ability to consider various factors for accurate diagnoses.
In the end, both Rizvi and Kamanger share a common sentiment: AI and human experts in healthcare should team up. Cooperation between the two camps could lead to stronger outcomes than either can achieve alone. So, let’s keep exploring this promising field and find ways for AI and radiologists to join forces for even better patient care.