How AI Is Transforming Medical Diagnostics

One of the most promising applications of AI is its potential to improve disease diagnosis. A study in the American Journal of Roentgenology suggests that, on average, radiologists misinterpret roughly 4 percent of the scans they see each day.

AI models trained on vast ­datasets can reduce errors by identifying patterns that might escape even experienced specialists. A 2023 study published in Radiology found that an autonomous AI system exhibited greater sensitivity than radiologists in identifying critical abnormalities on chest x-rays (99.8 percent versus 93.5 percent).

An AI tool can alert a radiologist to look at a specific part of the lung where it locates an anomaly. It can also compare a patient’s new scan with an older one to look for subtle changes. “It doesn’t supplant the radiologist from making the final decision, but it helps to prevent an error,” notes Lloyd Minor, MD, dean of the Stanford University School of Medicine, on The Rich Roll Podcast.

“The best-case scenario is
one in which AI enhances,
rather than replaces,
human expertise.”

Taofic Mounajjed, MD, a pathol­ogist with Hospital Pathology Associates, says AI can also provide clearer prognoses. Whereas human diagnosis relies on predefined classifications, such as tumor histology and stage, AI can analyze vast amounts of data to uncover subtle image characteristics that offer clues about a disease’s likely course. Those characteristics might include spatial relationships between cells or other statistical features of tumor cells, such as size, variability, and shape.

“There’s so much on a slide that we as human pathologists don’t see or report,” explains Mounajjed. “AI can look at hundreds of metrics and variables and say, ‘Based on our model, you have an 80 percent chance of recurrence in 10 years,’ which is a completely new way of looking at pathology.”

Treatment plans can also be refined with the assistance of AI tools. While doctors have long used biomarkers, like hormone receptor status, to individualize a patient’s treatment, AI can integrate multiple factors, such as tumor microenvironment and DNA mutations, to identify the most viable and effective treatment options for a particular patient.

Challenges ­remain. AI-based diagnostics can suffer from biases in the data used to train it, making it inaccurate and potentially harmful for underrepresented populations. Overreliance on AI might lead doctors to overlook their own clinical judgments or dismiss rare, complex cases that don’t align with algorithmic predictions.

The best-case scenario is one in which AI enhances, rather than replaces, human expertise and its benefits are accessible — and applicable — to all.