Artificial intelligence (AI) is starting to replace workers in various fields, but radiologists need not be overly concerned for now. Despite a global shortage of specialists, they analyze thousands of medical images each day, and their conclusions are vital for the early detection of cancer. This type of work is well-suited for AI applications.
One study from the Mayo Clinic found that an AI model can detect pancreatic cancer an average of 475 days before it becomes visible on scans. However, AI is not without its errors. Jens Vogel-Claussen, director of the radiology department at Charité University Clinic in Berlin, cited two examples: in one instance, the system mistakenly identified a scar from tuberculosis as a potentially cancerous node, while in another, it overlooked a suspicious sign that later turned out to be lung cancer.
“The software is not perfect, the radiologist is not perfect, but together, a radiologist with the appropriate AI software works more effectively than a radiologist alone,” he noted.
Potential and Limitations of AI in Medicine
Experts believe that AI has enormous potential to transform healthcare. It is already being implemented in various medical specialties, including oncology and pathology, where doctors examine biological samples on laboratory slides. However, this potential will not be fully realized for some time.
The main issue is that AI often draws conclusions with unwarranted confidence. “The model can sound absolutely confident and still be inaccurate, and in healthcare, this is not an abstract risk—it is someone’s diagnosis,” said Brian Anderson, CEO of the Coalition for Health AI, which collaborates with healthcare systems and providers to openly discuss the limitations of such tools.
In addition to the risk of misdiagnosis, AI can also reinforce erroneous conclusions made by doctors—a phenomenon known as confirmation bias. Since models are trained on different datasets, the quality of their interpretations can vary significantly.
Responsibility and the Future of AI
Another challenge is accountability. According to the World Health Organization, only 8% of countries implementing AI in diagnostics have established standards defining who is responsible in the event of a system error. Suzanne Ludwig, an analyst at Bernstein, believes that AI should complement human expertise rather than replace it.
“AI must function properly because, ultimately, patients’ lives are at stake, which significantly reduces tolerance for errors,” she noted.
Source: Bloomberg
