Will AI Replace Radiologists?

Image Citation: [1]

Written by Elise Park ‘28

Edited by Leopold Li ‘28

In 2023, a study of 1035 cases compared the detection rates of avulsion fractures between an optimized artificial intelligence model and practicing radiologists. While both AI and radiologists exhibited similar rates of accuracy, the AI model achieved a detection rate of 57.89%—radiologists identified just 29.82% [2].

While potentially alarming, this study reflects an undeniable trend of AI rapidly sweeping over the field of medicine, revolutionizing clinical workflow and diagnostic capabilities. With its ability to quickly process vast amounts of data and identify subtle patterns that may be overlooked by the human eye, AI algorithms are quickly reshaping healthcare practices, particularly in radiology—a field dependent on imaging, screening, and pattern recognition. Today, approximately one third of American radiologists report using AI regularly in their clinical practice [3]. This influence is only expected to grow, with the market for AI in medical imaging projected to increase tenfold over the next decade, surpassing $500 billion in 2030 [4].

At the core of this AI revolution is deep learning, a subset of machine learning that enables algorithms to independently learn from large datasets without being engineered by humans. The early 2000s witnessed a revolutionary shift with the advent of convolutional neural networks (CNNs) inspired by the layered structure of the human brain. These networks excel at processing visual data, quickly processing hierarchical representations of labelled images to build fast pattern recognition. AI models can quickly recognize and classify tumors, lesions, and other abnormalities—in many cases, outperforming experienced specialists [5].

Within radiology, AI is not only optimizing image interpretation but also image acquisition. Deep learning algorithms can accelerate magnetic resonance imaging (MRI), computed tomography (CT), and positron emission tomography (PET) scanning, enhancing diagnostic speed and accuracy while improving image reconstruction. They can, for example, enhance cardiac imaging by coloring grayscale echocardiography images, allowing clinicians to better visualize anatomical structures and abnormalities within the heart [4]. With more accurate and precise three-dimensional representations, including wall motions and left atrial/ventricular volume changes, AI models show promise in predicting conditions such as atrial fibrillation and myocardial infarction [5].

AI’s extraordinary processing speed also makes it a valuable tool for all clinicians. Algorithms can process biopsies, MRI scans, and blood tests far more rapidly than the human brain. A recent study in intraoperative diagnostics identified this rate as just under 150 seconds, while conventional pathology methods require an average of 20-30 minutes [4].

Apart from cardiovascular applications, breast cancer, which accounts for nearly 15% of all AI applications in medical imaging, remains a primary area of interest for AI developers [3]. AI tools have been developed to assist in breast lesion characterization, density estimation, and image quality improvement in mammography. These models have a 12.3% higher recall rate than that of radiologists, highlighting their potential to improve early detection and reduce false positives in one of the most common cancers [3].

However, with the recent explosion of AI in medicine, new concerns and challenges have emerged. A recent study found that 44% of American medical students report being less likely to specialize in radiology due to the increasing influence of AI [3]. This reflects the growing trend involving medical students fearing career displacement by artificial intelligence, raising concern over projected workfield shortages in the specialty. In response, there have been widespread calls to integrate AI education into radiology residency programs. In fact, 83% of radiology residents support integrating AI training into their curriculum [6]. By collaborating with AI models and developing human-AI hybrid decision making systems, future physicians can learn to harness these tools rather than competing with them.

With the potential to permanently reshape clinical practice, it’s clear that AI will remain a lasting presence in the reading room. However, instead of viewing AI as a threat, medical students and healthcare professionals should approach it as an evolving tool, one that can complement clinical expertise and ultimately work to improve patient outcomes.

References

1. America RS of N. Roadmap for AI in Medical Imaging [Internet]. Research & Development World. 2019. Available from: https://www.rdworldonline.com/roadmap-for-ai-in-medical-imaging/.

2. Liu Y, Liu W, Chen H, Xie S, Wang C, Liang T, et al. Artificial intelligence versus radiologist in the accuracy of fracture detection based on computed tomography images: a multi-dimensional, multi-region analysis. Quantitative Imaging in Medicine and Surgery [Internet]. 2023 Oct 1 [cited 2023 Nov 7];13(10):6424–33. Available from: https://pubmed.ncbi.nlm.nih.gov/37869340/.

3. Mello-Thoms C, Mello CAB. Clinical Applications of Artificial Intelligence in Radiology. Clinical Applications of Artificial Intelligence in Radiology. 2023 Apr 26;96(1150).

4. Transition. AI in radiology: 10 use cases, benefits and examples [Internet]. www.itransition.com. 2024. Available from: https://www.itransition.com/ai/radiology.

5. Najjar R. Redefining Radiology: A Review of Artificial Intelligence Integration in Medical Imaging. Diagnostics [Internet]. 2023;13(17):2760. Available from: https://www.mdpi.com/2075-4418/13/17/2760.

6. Salastekar NV, Maxfield C, Hanna TN, Krupinski EA, Heitkamp D, Grimm LJ. Artificial Intelligence/Machine Learning Education in Radiology: Multi-institutional Survey of Radiology Residents in the United States. Academic Radiology. 2023.