Amber Sun, MSIV

The leadership of the ACR prioritizes connecting with members-in-training, including medical students, to keep the radiology pipeline open to the best and brightest candidates. Students, trainees and rising radiologists offer their own unique perspectives on the future of the specialty. Their thought processes and contributions to the field are vital as the College develops and implements its strategic plan. To advance the ACR’s commitment to better engage medical students and provide more volunteer opportunities, the Medical Student Subcommittee was elevated to the Medical Student Section (MSS). The Section reflects the value medical student members bring to the College’s entire membership.

The MSS, representing over 3,600 ACR members-in-training medical students, is led by a nationally representative steering committee of medical students and an RFS advisor. The goal of the MSS is to develop resources for students exploring careers in diagnostic or interventional radiology, nuclear medicine, radiation oncology, or MRI physics. This is the second article in a new Bulletin series authored by medical students, who share their perspectives on the evolving radiology landscape.

Medical Student Perspectives

The integration of AI in medical imaging has brought revolutionary changes, particularly in mammography. One of the most promising applications is AI as a second reader (ASIR), a tool that complements radiologists in interpreting screening mammograms. ASIR has been shown to demonstrate favorable performance in the interpretation of screening mammograms in the U.S. and the U.K.

Mammography has long been a cornerstone in the early detection of breast cancer. However, screening mammography is complex and subject to variability depending on the reader. Furthermore, with increasing workloads, even experienced breast radiologists may encounter difficulties in maintaining precision over time. The introduction of ASIR could offer a potential solution to these challenges.

AI Enhances Mammography

AI algorithms have been developed to analyze mammographic images with remarkable speed and consistency. These systems are trained on large datasets containing thousands of mammograms, allowing them to learn imaging patterns associated with malignant lesions, calcifications, lymph nodes and more. 

The two-reader model, traditionally used in many screening programs outside the U.S., involves double-reading by two radiologists. When deployed as a second reader, a 2023 retrospective study found that their AI system — using data representative of real-world deployments with more than 275,000 mammograms — had non-inferior recall rate, sensitivity, specificity, and positive predictive value with a reduction of workload up to approximately 45%. Moreover, the study found that their AI system can provide comparable accuracy to human radiologists, particularly in identifying subtle lesions.

AI algorithms have been developed to analyze mammographic images with remarkable speed and consistency.

A recent study from 2024 found that, within their AI system, ASIR actually improves sensitivity and decreases specificity when compared to routine dual screening. When used properly and effectively, ASIR has the potential to reduce human reader workload in two-reader models while maintaining or improving the standard of care. Beyond simply identifying lesions, advanced AI systems can prioritize high-risk cases for quicker review, optimizing workflow for radiologists. In resource-limited environments, AI plays a crucial role in optimizing the use of available resources to ensure efficient and timely care.

AI is not meant to replace radiologists; instead, it serves as a powerful assistive tool. In a collaborative workflow, radiologists can combine their clinical judgment with AI to improve overall decision-making and efficiency. As a second reader, AI can identify areas of concern that may prompt further investigation, compelling radiologists to take a closer look at potential abnormalities.

Moreover, AI offers the ability to manage large datasets with relative ease — analyzing thousands of images rapidly while maintaining high sensitivity and consistency. This capability is critical in screening programs with high volumes of cases and wherever efficiency and accuracy have become or are becoming a concern.

Potential Challenges and Ethical Considerations

While AI offers significant benefits, its adoption comes with potential pitfalls. Integrating AI into clinical practice requires careful consideration of ethical, legal and technical challenges. Questions about algorithmic transparency, bias in AI models and the interpretability of AI-generated decisions remain ongoing concerns. Additionally, radiologists need to be trained to effectively interact with AI outputs, integrating these tools into their diagnostic process without becoming overly reliant on automated assessments. An article from the British Journal of Cancer provides one detailed review of the ethical constraints of AI in breast imaging. 

Despite these challenges, the continued development and validation of AI in mammography points to a promising future. The integration of ASIR is just the beginning. Researchers and developers are exploring AI’s potential in more advanced imaging modalities, risk prediction and personalized screening protocols. Soon, AI could help tailor screening strategies to individual patient risk profiles, further optimizing the balance between early detection and minimizing unnecessary interventions.

AI’s role in imaging will continue to evolve, but it is already proving to be a game-changer in screening mammography in different countries. By improving accuracy, efficiency and consistency, ASIR represents a significant leap forward in the fight against breast cancer. As technology advances, the collaborative partnership between radiologists and AI will play an integral role in clinical practice.

For radiologists, embracing AI is about enhancing, not replacing, their expertise. As AI becomes more integrated into clinical workflows, radiologists can avoid human reader-burden and focus more on nuanced cases and clinical decision-making. The constructive collaboration between human expertise and machine learning will most definitely shape the future of breast imaging.