Interview with Dr. Stéphanie Lopez from Université Côte d’Azur.

Interview conducted by Pascale Caron

Between 2018 and today, I led an AI healthcare project, LungScreenAI at Université Côte d’Azur, in partnership with the University Hospital (CHU). This project uses AI to assess the malignancy of pulmonary nodules. My role was to develop an AI tool for radiologists. This included structuring data, recruiting and managing a team, and collaborating with healthcare professionals on integrating these tools into their clinical practice.

After initial funding from the Initiative d’Excellence (IdEx) of Université Côte d’Azur in 2018, we participated in AstraZeneca’s EXPLORE call for projects in 2019. A budget of 200,000 euros was then granted to us to build a multidisciplinary team and deliver a proof of concept in AI directly usable by healthcare professionals.

After six years, this project is integrated into a clinical study led by the Nice University Hospital, but due to lack of necessary funds to move to the next stage, it is currently on hold.

 

What motivated you to include AI in your sector and what changes has it brought?

In healthcare, access to reliable data and rapid diagnoses are crucial. AI offers a unique capability to instantly analyze large databases and predict the malignancy of pulmonary nodules, thus enabling earlier patient care.

With LungScreenAI, we successfully reduced the number of false positives and identified potential cases up to two years before their clinical diagnosis. This technology does not replace radiologists, but complements them by offering a second reading and predictive support.

 

What AI solutions have you adopted?

We used deep learning models, based on convolutional networks, to analyze chest images. By relying on public databases such as NLST, we trained the algorithm with data coupled with histological evidence.

We also collaborated with radiologists to annotate images, which improved the accuracy of predictions. This cooperation was crucial to adjust the models to clinical reality and ensure the tool was intuitive for practitioners.

How did you overcome the human and cultural challenges when integrating AI?

The adoption of AI can raise concerns, particularly in fields as sensitive as healthcare. It was essential to demystify AI by emphasizing its transparency and its role as an assistant, not a replacement.

We organized regular workshops with CHU physicians to understand their needs and concretely show them how AI could improve their routine tasks. These sessions helped create a trust relationship and incorporate their feedback to refine the tool. The interaction with radiologists was particularly enriching: by working alongside them daily, I was able to immerse myself in their practices, which was crucial for adapting our solution to their real needs.

 

What technical obstacles did you encounter?

Access to pseudonymized health data was a real challenge. We collaborated with the CHU’s DPO to comply with regulatory standards, which extended certain stages by three years.

The pseudonymization process, while essential to ensure patient confidentiality, was a considerable task to align legal and technical requirements. Furthermore, developing a robust and generalizable algorithm required constant optimization of models to reduce false positives while maintaining high sensitivity.

The technical ecosystem of the CHU, particularly the interconnections between data platforms, was also an obstacle that we overcame through close collaboration with the IT services of the Nice University Hospital.

 

Why not patent your solutions?

In the field of AI, patenting is often complex, because the tools are mostly based on pre-existing technologies, which limits patentable novelty. Moreover, with technological advances, patents quickly become obsolete.

We preferred to protect our developments through the Agence de Protection des Programmes (APP), which allows us to register priority while remaining flexible. This approach is better suited to a constantly evolving environment like AI.

What positive changes have you observed in your team?

The integration of AI strengthened cohesion among the different project stakeholders. Exchanges between data scientists, physicians, and regulators created a rich and innovative collaborative dynamic.

Radiologists, for example, feel more equipped to make decisions and consider AI as a tool that enhances their expertise. This synergy allowed everyone to better understand the constraints of each profession, thus fostering the emergence of more robust and adapted solutions.

 

What advice would you give to SMEs hesitant to embrace AI?

AI may seem complex and costly, but it represents an immense opportunity to transform processes. Start small: identify a clear use case, work with experts to develop a prototype, and actively involve your teams to ensure gradual adoption.

Finally, if you are in the healthcare field, be prepared to overcome regulatory obstacles, keeping in mind that these constraints are also a guarantee of quality. AI is not a magic solution, but it can transform complex processes if properly implemented.