Choosing A Medical Specialty In The Age Of Artificial Intelligence

Choosing A Medical Specialty In The Age Of Artificial Intelligence

Artificial intelligence (AI) is everywhere—like OpenAI’s Chat-GPT—and increasingly used in medicine. AI could someday augment or even replace some tasks doctors do.

Medical students should consider how AI could change the work physicians do (and get paid for) when choosing specialties. This is vital given the large time investment in specialty training and high barriers to switching.

Similar concerns of “replacement” arose with the industrial revolution (1760-1840) that workers might become irrelevant with the invention of machines like James Watt’s steam engine in 1775 and Eli Whitney’s cotton gin in 1793. Ultimately, some jobs were eliminated. But other jobs were created as humans were needed to mind and fix the machines.

AI’s revolution may take a similar path. Yet it will be exponentially faster. As AI changes the practice of medicine, it may also change the labor needs of the physician workforce in even more dramatic ways than the industrial revolution.

Here are specialties where AI could eventually replace a substantial piece of doctors’ work in the future.

Diagnostic radiology

Radiology uses imaging technologies: X-rays, CT scans, MRIs, and ultrasounds to diagnose disease. AI algorithms are great at detecting image patterns and analyzing digitized data.

In a 2019 study, stand-alone AI with no radiologist input was compared with reads from 101 radiologists in 2,652 mammograms. It found that AI reds were just as good as radiologists. AI-assistance in reading chest x-rays improved the sensitivity of detecting specific x-ray abnormalities for pneumothorax (where air is outside the lung) by 26%, consolidation (which indicates infection), and pulmonary nodules (which can be an early sign of a tumor) by 9%.

In the near term, AI will be a powerful collaborator to radiologists. Yet in the long-term, there may be less need for diagnostic radiologists as AI becomes increasingly independent.

Diagnostic pathology

Pathology involves making diagnoses through examining tissues, cells, and bodily fluids and using laboratory-based tools. Like radiology, AI-powered algorithms analyze digitized pathology slides enhancing cancer detection, tumor classification, and biomarker quantification.

A 2022 study demonstrated that AI models significantly improved diagnostic accuracy of pathological reports for deep myxoid soft tissue lesions which are notoriously difficult to diagnose. The AI model had an accuracy of 97% compared to 70% for human pathologists, and reduced their error rate by 90%. A 2024 study found that a stand-alone AI model was more accurate in interpreting cytology for thyroid fine-needle aspiration, 95% v. 89% compared to expert cytopathologists.

Today AI can augment the accuracy, speed, and consistency of pathologists. Yet as AI advances, some of the work in reading diagnostic pathology images could become entirely automated.

Dermatology

Dermatology involves assessment of rashes and skin lesions, commonly as referrals from generalist physicians and in patients seeking care directly. AI models trained on large datasets of skin images are able to identify cancer and diagnose chronic skin conditions.

A recent study found that AI support significantly improved the sensitivity and accuracy of dermatologists in classifying dermoscopic images of melanoma (i.e. cancer) and nevi (i.e. moles). The sensitivity increased from 60% to 75% and the accuracy from 65% to 73%. Another study found that an AI-assistant improved the accuracy of non-expert physicians in diagnosing skin conditions with an accuracy of the AI-assisted group of 54% v. 44% for the unaided group.

AI algorithms for skin diagnoses are being used in practice and will continue to improve. Some—like Skin Vision and Mole Mapper—diagnose and track skin conditions with no human input. Expanded use of this technology could increasingly shift some skin diagnoses and management to non-specialists as well as directly to patients.

Internal Medicine & Pediatric Non-Procedural Specialists

Cardiologists, endocrinologists, gastroenterologists, rheumatologists, and infectious disease physicians are experts in treating complex diseases in their niche. Training takes several years beyond internal medicine or pediatric residency.

In the future, AI may lower the need to consult these specialists for advice and management. Instead, consulting physicians may rely increasingly on AI, who can help interpret lab findings, electrocardiograms, and make evidence-based recommendations. Generalists—like physicians in primary care, pediatrics, emergency medicine and hospital medicine—may become more comfortable managing complicated conditions with the AI’s guidance, ultimately expanding their scope of practice.

Additionally, AI will likely impact non-procedural work faster than procedural work, like surgical care. Today’s AI in medicine is primary designed to analyze data and is unlikely to operate an independent, surgical robot any time soon.

How fast or even whether these changes will happen in these specialties is unknown. Healthcare is notoriously slow in adopting innovation. Resistance by specialty groups will be strong, particularly when it impacts what physicians get paid to do and referral patterns.

Nevertheless, considering the potential future is important for medical students who are deciding on a long career at a time where AI tools are just starting to change clinical practice.

link

Leave a Reply

Your email address will not be published. Required fields are marked *