Usages and perceptions of artificial intelligence among French radiologists

Although French radiologists recognize the potential of AI to enhance precision and personalization in their field and overwhelmingly express a willingness to adopt it, its current implementation is hindered primarily by high costs and insufficient equipment.

The Gist

Imagine the world of medical imaging as a vast, high-stakes library where radiologists are the master librarians. Their job is to scan thousands of books (X-rays, mammograms, CT scans) every day to find the one tiny, hidden typo that could mean a life-or-death diagnosis.

Now, imagine a new, super-smart robot assistant (Artificial Intelligence) has been invented to help these librarians. This robot can spot typos faster than a human eye and never gets tired. But here's the twist: The librarians are ready to hire the robot, but the library owners haven't bought one yet.

This paper is a report card on how French radiologists feel about this new robot assistant. Here is the breakdown in plain English:

1. The Big Picture: Everyone Wants the Robot, But Few Have It

The study asked 100 French radiologists, "Do you use this AI robot?"

• The Good News: Almost no one is afraid of the robot. In fact, 0% of them said, "I will never use this." They aren't worried about the robot stealing their jobs; they see it as a helpful tool. Over 70% said, "Yes, I want to use it!"
• The Bad News: Most of them don't have it yet. While 80% of radiologists in big university hospitals have the robot, only about 1 in 4 or 1 in 3 in regular clinics have it.

2. Why Don't They Have It? (The "Too Expensive" Problem)

If everyone wants the robot, why isn't it on every desk? The paper points to two main roadblocks:

• The Price Tag: The robot is very expensive. It's like trying to buy a Ferrari when you only have a bicycle budget. For many clinics, the cost is just too high.
• The Empty Garage: Many clinics simply don't have the equipment installed. It's not that the doctors are refusing the technology; it's that the technology hasn't been delivered to their doorstep yet.

3. The "Mammography" Special Case

The study focused heavily on breast cancer screening (mammography).

• The Robot's Track Record: In other areas of radiology (like looking at lungs or bones), the robot is already helping 75% of doctors. But in breast imaging, only about 53% are using it.
• Why the lag? The paper suggests that the "rules" for breast imaging robots are stricter and harder to perfect. Also, there are fewer certified robots specifically for breasts compared to other body parts.
• The Workflow Hiccup: Interestingly, doctors using the robot for other body parts said it made their day much faster (85%). But for breast imaging specialists, only 65% felt it helped. Some felt the robot made them spend more time double-checking the robot's work, which slowed them down.

4. What Do the Doctors Actually Want?

When asked what they need from this AI, the doctors didn't say, "I want to be replaced." They said:

• "Help me find the needle in the haystack." (Better diagnosis).
• "Don't let me miss anything." (Fewer errors).
• "Let me go home on time." (Saving time and boosting productivity).

5. The Fear Factor: "Will the Robot Take My Job?"

This is the question everyone asks about AI. The paper has a very reassuring answer: No.

• Only about 8% of the doctors (mostly breast specialists) felt a tiny bit of fear that the robot might replace them.
• The vast majority see the robot as a co-pilot, not a replacement. They want the robot to handle the boring, repetitive scanning so they can focus on the complex decisions and patient care.

The Bottom Line

Think of French radiology as a group of drivers who are all eager to upgrade to self-driving cars. They know the cars are safer and more efficient. But right now, the cars are too expensive, and the roads aren't fully paved with the necessary technology yet.

The Conclusion: The doctors are ready, willing, and able to use AI. They aren't scared of it. The only thing stopping them is the cost and the lack of equipment. If the price comes down and the tools become available, the revolution in medical imaging will happen very quickly.

---

A Note on the Source:
This paper is a "preprint," which means it's like a draft report that hasn't been fully checked by other experts yet (peer review). It's a sneak peek into new research, so while the findings are promising, they should be taken as a strong signal rather than a final law of physics.

Technical Summary

1. Problem Statement

Despite the recognized potential of Artificial Intelligence (AI) to enhance radiology through improved precision, prediction, and personalization (particularly in breast imaging/mammography), its actual adoption in France remains heterogeneous and limited.

• The Gap: While University Hospitals report high adoption rates (80%), general radiology centers lag significantly. Estimates suggest only 20–30% of radiologists use diagnostic AI in mammography, with only 10% of centers equipped as of 2023.
• The Unknown: There is a lack of precise data regarding the specific barriers preventing wider adoption, the actual willingness of practitioners to integrate these tools, and the specific expectations radiologists have for AI solutions in their daily workflow.
• Context: Previous studies (2019, 2024) indicated an appetite for AI, but the current landscape requires an updated analysis of barriers (cost, equipment, trust) and the fear of professional displacement.

2. Methodology

• Study Design: A cross-sectional survey study conducted via an online Google Form (GDPR-compliant).
• Sample Size: 100 radiologists practicing across France, including overseas departments and territories (DOM-TOM).
• Recruitment: Conducted via professional networks (LinkedIn), professional groups, and solicitation of radiology centers.
• Demographics & Bias:
  • Geography: Heavily skewed toward Île-de-France (48%), reflecting the higher density of radiologists in Paris but introducing a selection bias.
  • Experience: The sample is senior-heavy; 89% have more than 10 years of experience, and 38.9% have 20–30 years of experience. Younger radiologists (<5 years) were underrepresented (6.9%).
  • Specialization: 72% of respondents specialize in mammography.
• Data Collection: The questionnaire contained 10 questions, distinguishing between users and non-users of AI, and between mammography specialists and other radiologists. Questions covered usage status, reasons for non-adoption, workflow impact, fear of replacement, and expectations.
• Analysis: Results were analyzed as percentages based on respondent counts for each specific question.

3. Key Contributions

• Quantification of Adoption Barriers: The study identifies that the primary barrier is infrastructure/equipment (lack of AI tools in the center) rather than a lack of interest or trust.
• Differentiation by Specialty: It highlights a significant divergence between mammography specialists and other radiologists regarding workflow improvement and perceived value.
• Psychological Profile: It provides data on the "fear of replacement," showing it is negligible among French radiologists.
• Market Insight: It reveals a strong demand for free or accessible AI solutions, suggesting cost is a major friction point for non-adopters.

4. Key Results

A. Adoption Rates

• Mammography Specialists: 52.8% currently use AI.
• Other Radiologists: 75% currently use AI.
• Note: The higher rate among non-mammography specialists is attributed to a higher availability of CE-marked algorithms for general radiology (90%) compared to breast imaging (10%).

B. Barriers to Adoption (For Non-Users)

• Lack of Equipment: The dominant factor. 94.1% of non-using mammography specialists and 71.4% of others cited that their imaging centers are not equipped with AI.
• Cost: 35.3% of mammography specialists and 28.6% of others cited high costs.
• Lack of Trust/Value: Only 8.8% of mammography specialists expressed a lack of trust. However, 42.9% of other radiologists felt AI offered no added value to their specific practice.

C. Workflow and Impact

• Workflow Improvement: Among current users, 65.8% of mammography specialists reported improved workflows, compared to 85.7% of other radiologists.
  • Context: The lower satisfaction in mammography aligns with literature suggesting AI can increase deliberation time on doubtful cases in breast imaging.
• Fear of Replacement: Extremely low.
  • Mammography specialists: 8.8% fear replacement.
  • Other radiologists: 0% fear replacement.

D. Future Intentions and Expectations

• Willingness to Adopt: 79.4% of non-using mammography specialists and 71.4% of others expressed a desire to adopt AI if available. 0% of all respondents declared a refusal to use AI.
• Primary Expectations:
  1. Aid to Diagnosis: The top expectation for both groups.
  2. Error Reduction: Reducing false negatives/positives.
  3. Time/Productivity: Saving time and improving efficiency.
• Cost Sensitivity: 79.4% of non-using mammography specialists indicated they would want a free AI solution.

5. Significance and Discussion

• Readiness vs. Accessibility: The study concludes that French radiologists are psychologically and professionally ready to adopt AI. The bottleneck is not resistance but accessibility (equipment availability) and affordability (cost).
• Specialty-Specific Challenges: The lower workflow satisfaction among mammography specialists suggests that current AI tools in breast imaging may not yet be fully optimized for clinical integration, potentially creating "alert fatigue" or requiring excessive human verification time.
• Trust and Governance: The authors emphasize that for adoption to scale, AI developers must provide transparent governance, explainability, and regulatory compliance (e.g., EU AI Act) to build trust.
• Policy Implication: To accelerate the "preventive paradigm" in radiology, stakeholders must focus on subsidizing AI deployment in private centers and developing tools that demonstrably save time without compromising diagnostic accuracy.

Limitations: The authors acknowledge the small sample size (N=100) and the selection bias (overrepresentation of Île-de-France and senior radiologists), which may skew the results toward a more optimistic view of AI adoption than the national average.

Conclusion: The integration of AI in French radiology is hindered primarily by economic and infrastructural constraints rather than cultural resistance. Radiologists view AI as a collaborative tool for precision and efficiency, with negligible fear of obsolescence.