Artificial Intelligence in Medicine: Interpreting Clinical Imaging

The interpretation of medical images is “a task that lies at the heart of the radiologist’s work.”^[i] In fact, the human element—a radiologist’s eyes and brain—is just as important as what a high resolution, 3- dimensional image depicts. Imaging types include x-ray, MRI, CT scan, ultrasound and PET scan, but no matter how advanced the technology, it’s literally useless without the professional reader who interprets what is shown.

That appears to be changing. Radiologists have a new collaborator in the form of Artificial Intelligence (AI). In fact, “use of A I in radiology has shown great promise in detecting and classifying abnormalities on plain radiographs, computed tomographic (CT) scans, and magnetic resonance imaging (MRI) scans, leading to more accurate diagnoses and improved treatment decisions.”^[ii] There are several areas in which AI is already improving the efficiency and accuracy of human abilities:

• It can quantify specific values such as tissue density, anatomy, and even blood flow
• It can speed up imaging preproduction, enhancing features swiftly even as the image appears on a monitor, making a radiologist’s recognition more efficient
• It can prioritize dangerous conditions such as a brain bleed for immediate attention
• It can make predictions of the aggression of a cancer or the outcome of a treatment.

For its development and training, AI depends on informed, correct human judgment. It is, after all, computer experts and clinicians who perform the work of defining an AI task, selecting slides for marking and digitizing, creating a program/algorithm, inputting, and testing the data. You know the expression, “Garbage in, garbage out,” meaning AI output is only as good as how it is programmed and what’s fed into it. So far, so good: in head-to-head comparisons, AI performance appears to be generally competitive with experienced readers.

Radiologists on alert

Indeed, AI is good enough that some radiologists are worried about the future of their jobs. These fears can be calmed if AI is understood as a co-pilot or navigator in the clinical cockpit, rather than the captain of the ship. However, in many hospitals and medical centers where radiologists are the first to see images and write a report for a specialist to read, some AI applications are designed directly for the specialist. This skips the radiological middleman, so to speak. For example, a urologist may submit images of a patient’s prostate biopsy slides to a predictive algorithm, or a cardiologist may be able to evaluate the narrowing of a patient’s aorta without consulting a radiological reader. It remains to be seen to what extent AI programs are adapted for direct use by non-radiological personnel, but many radiologists consider themselves on alert.

Enthusiasm balanced with caution

Needless to say, proponents of AI interpretation of clinical imaging are the rah-rahs on the AI bandwagon. It may be easy for them to overlook potential bumps in the road of real-world medicine. According to one article, these include AI results presented in ways difficult for a clinician to understand for real-life workflows; lack of validation via independent scientific, peer-reviewed results of large randomized trials; inadequate data on AI performance with underrepresented or minority patient subgroups; and potential for failing to adapt as disease patterns or populations shift over time. The article’s authors particularly note:

The current generation of AI models in radiology can handle only a limited set of interpretation tasks, and they rely heavily on curated data that have been specifically labeled and categorized. Although focusing on the image as an isolated model input has some value, it does not reflect the true cognitive work of radiology, which involves interpreting medical-imaging examinations comprehensively, comparing cur-rent and previous examinations,and synthesizing this information with clinical contextual data to make diagnostic and management
recommendations.^[iii]

The Sperling Medical Group, including our Sperling Prostate Center, is proud to be among the estimated 2% of U.S. medical centers that integrate AI into our diagnostic and treatment planning. The world of medicine is poised on the threshold of AI integration in clinical practice, and we are pleased to be pioneers in applying it to our MRI-based services. To learn more, contact us.

NOTE: This content is solely for purposes of information and does not substitute for diagnostic or medical advice. Talk to your doctor if you are experiencing pelvic pain, or have any other health concerns or questions of a personal medical nature.

References

[i] Rajpurkar P, Lungren MP. The Current and Future State of AI Interpretation of Medical Images. N Engl J Med. 2023 May 25;388(21):1981-1990.
[ii] Ibid.
[iii] Ibid.