Around the time data scientist was being heralded as “one of the sexiest jobs of the 21st Century,” Valerio Guarrasi realized that it was also a natural intersection of programming, optimization, statistics, and other disciplines that had lit up his undergraduate mind.
Soon he was enrolled in one of Italy’s first data science graduate programs—at Sapienza University of Rome—and a vision for his career began taking shape.
Today, Guarrasi is an assistant professor at Campus Bio-Medico University of Rome—and one of Computing’s Top 30 Early Career Professionals for 2024.
In the following Q&A, Guarrasi describes:
- How COVID-19’s peak during his PhD program led to his participation in a pivotal project that used AI to predict clinical outcomes in COVID patients
- The connections between medicine’s standard decision-making process and his thesis on multimodal deep learning methodologies
- How his Virtual Scanner project’s ability to use generative AI to translate between imaging modalities produces practical, real-world benefits to patients
- Why, despite his major research contributions, Guarrasi still calls teaching “the best part” of his job
- How he makes complex topics accessible using a layered teaching approach that can be useful for anyone tasked with explaining complex technologies to laypeople
What sparked your interest in data science and artificial intelligence? How did your educational background shape your career path?
During my bachelor’s studies in management engineering, I was particularly drawn to subjects like programming, operations research, optimization, statistics, probability, and databases. These areas gave me a solid foundation in analytical thinking and problem-solving, which I found both challenging and rewarding.
Around that time, data science was starting to gain significant attention—”Data Scientist: The Sexiest Job of the 21st Century” was a phrase that seemed to be everywhere. It piqued my interest, but more importantly, I saw in data science a natural intersection of the disciplines I enjoyed most. That interest led me to enroll in a brand new master’s degree in data science program, one of the first of its kind in my country. It felt like the natural next step to deepen my understanding and start applying these methods to real-world problems.
That same curiosity continues to drive me today. It led me to pursue a PhD in data science, followed by a postdoctoral position focused on AI for precision medicine, and now a role as a researcher. Along the way, my motivation has remained the same: to keep exploring how data and intelligent systems can improve the way we understand and solve complex problems.
Can you share more about your PhD research on “Multimodal Deep Learning for Medical Imaging”? What were the key findings and their implications for the field of precision medicine?
My PhD journey began in an unusual context—during the peak of the COVID-19 pandemic. That first year shaped not only the course of my research but also my perspective on the role of AI in urgent, real-world healthcare challenges. Responding to the immediate need, I joined the AIforCOVID project, an Italian multicenter initiative focused on predicting clinical outcomes in COVID-19 patients using AI on chest x-rays. This early experience gave me the opportunity to work on a high-impact prognostic task, and it laid the foundation for my thesis, which focused on multimodal deep learning methodologies.
Medical decision-making is inherently multimodal: clinicians integrate information from imaging, clinical history, laboratory results, and more. My research aimed to bring that richness into AI models by developing methods capable of learning from multiple data modalities—radiomic features, clinical variables, and beyond. A central theme of my thesis was how to design, fuse, and explain multimodal deep learning systems.
All methodologies were validated in the context of COVID-19, using real patient data and targeting both diagnostic and prognostic outcomes. These contributions have implications beyond the pandemic, especially in the field of precision medicine, where leveraging diverse data sources to create individualized models is key.
As an assistant professor at Università Campus Bio-Medico di Roma, how do you approach teaching complex subjects like AI and machine learning to your students? What do you find most rewarding about this role?
Teaching is truly the best part of my job. For a long time, I imagined that my ideal career would involve only teaching—spending all my time preparing lectures, designing courses, and engaging with students. Over the years, though, I’ve come to appreciate the value of combining teaching with research. Research allows you to stay at the forefront of innovation, and that naturally enriches what you bring into the classroom. It means you’re not only teaching the fundamentals but also showing students how those fundamentals shape and are shaped by cutting-edge developments in AI.
When it comes to teaching complex topics like artificial intelligence and machine learning, I believe in a gradual and layered approach. These subjects often involve a fair amount of math, but I start by focusing on intuition—helping students understand the why behind each algorithm. Once that foundation is set, we dive into the theoretical aspects, and finally, we emphasize hands-on implementation. It’s only through practice that students see how everything connects.
I also believe that AI is a very visual discipline. Concepts like neural networks, backpropagation, or dimensionality reduction can be abstract, but when you complement theory with visualizations, interactive tools, and real examples, the ideas really come to life. Math and visuals are not in opposition—they’re complementary, and together they help students build a deep, well-rounded understanding.
What I find most rewarding is when students go beyond what’s taught in class—when they use the tools and knowledge we’ve discussed to start a project of their own, not because it was assigned, but because they’re genuinely curious. Seeing them take initiative, explore beyond the textbook, and apply what they’ve learned in new ways is the clearest sign that the teaching has had a real impact.
Your research spans various domains, including multimodal deep learning, explainable AI, and digital twins. Can you elaborate on a recent project you are particularly proud of and its potential impact on healthcare?
One project I’m particularly proud of is what we call the Virtual Scanner—a framework that leverages generative AI to translate one medical imaging modality into another. The idea is to simulate scans that would otherwise require additional procedures, radiation exposure, or contrast agents, using already available data. In doing so, we aim to reduce patient burden, support a more holistic diagnostic perspective, and lower costs, particularly for low-resource settings where access to advanced imaging equipment may be limited.
The Virtual Scanner sits at the intersection of digital twins and generative models and is designed to augment, rather than replace, current imaging protocols. By virtually generating complementary imaging modalities, clinicians can gain deeper insights while minimizing invasiveness and operational costs.
We’ve applied this concept across several use cases:
- From chest x-rays to radiology reports (and vice versa): Supporting faster reporting, second opinions, and training junior radiologists with richer, multimodal data.
- From low-dose to high-dose CT: Preserving diagnostic quality while minimizing radiation exposure for patients, especially in repeated scans.
- Virtual contrast enhancement in mammography: Enabling the benefits of contrast imaging without the need for contrast agents, making screening safer and more accessible.
- Harmonization of CT scans from different machines: Improving the comparability of scans across institutions, which is crucial for multi-center studies and AI training.
- From whole-body CT to PET imaging: Offering functional imaging insights in oncology without the high cost and logistical challenges of PET scanners.
- From MRI to CT: Facilitating treatment planning and data augmentation, especially where CT data is missing or hard to acquire.
This line of work opens the door to safer, more cost-effective, and more equitable healthcare, especially as we continue to refine the fidelity, robustness, and interpretability of these generative systems. It also strengthens the broader vision of AI-powered digital twins, where patient-specific models integrate multiple data streams to support precision diagnostics and personalized treatment planning.
Given your extensive experience in applying AI to medical data, what do you see as the most promising advancements in AI for healthcare in the next five years?
One of the most promising directions I see is the development of generative, multimodal, any-to-any systems—foundation models capable of understanding and generating across multiple medical data modalities. These models would take in any subset of available patient data—such as an image, a clinical note, or lab results—and be able to generate or simulate missing modalities, effectively “completing” the patient profile.
This kind of system would be transformative for healthcare. First, it addresses a common challenge in real-world clinical settings: missing data. Instead of discarding incomplete records or compromising model performance, AI could intelligently fill in gaps to support diagnosis and treatment. Second, it would empower clinicians with a more holistic and consistent view of the patient—even when only partial data is available. Finally, it has the potential to democratize access to advanced diagnostics, especially in low-resource environments, by simulating high-cost or less accessible modalities.
How do you balance your responsibilities as a researcher, educator, and project coordinator? What strategies do you use to manage your time and ensure productivity across these roles?
Time is one of the most critical resources in this career. Between research, teaching, and coordinating projects, there are often overlapping priorities and objectives that don’t always align. What I’ve learned is that this role works best if approached with an altruistic mindset. If you genuinely care about what’s best for your students, for the researchers you supervise, and for the broader goals of your projects, quality tends to follow naturally.
When it comes to managing people, I believe in a balanced approach: not micromanaging every step, but also not being completely hands-off. It’s important to provide direction, ask the right questions, and set a clear framework—while also giving young researchers the freedom to find their own solutions. If your focus is on helping them grow and succeed, the results will benefit everyone involved, including your own work.
As for time management, I’ve never been someone who postpones tasks. Maybe it comes from my childhood, where I balanced school with competitive swimming and intensive classical piano playing—disciplines that taught me the value of structure and consistency. I tend to tackle things straight away rather than waiting for a deadline. I believe that the things we do need time to breathe. Starting early allows you to revisit your work with a fresh perspective, giving space for improvements and often leading to unexpected insights. Ultimately, productivity for me comes from a mix of discipline, care for people, and respect for the process itself.
What advice would you give to students and professionals who aspire to make significant contributions to artificial intelligence and data science? What skills and experiences do you consider most valuable?
Whether you’re pursuing a career in academia or industry, my first piece of advice is: never stop learning and, just as importantly, never think you’ve learned it all. Being humble with knowledge is essential in a fast-moving field like AI. There’s always something new to discover and always someone you can learn from. If you keep that mindset, you’ll never be out of step with the world, and when you feel the need to change direction, you’ll already have the tools and mindset to do it.
For those focused on AI, I can’t stress enough: don’t overlook the fundamentals. There’s a tendency to jump straight into the latest trending models without truly understanding the core principles behind them. But if you look closely, many of today’s “new” methodologies are simply creative combinations of existing techniques, structured in a novel way. The deeper your understanding of foundational concepts—like optimization, probability, neural networks, or representation learning—the better equipped you’ll be to not only understand these new models, but to build your own.
Another piece of advice: even if your career path pushes you toward specialization, it’s crucial to maintain a broad perspective. Try not to get trapped in a narrow view of your work. Understanding how your niche connects to the bigger picture will make you not only a better researcher or practitioner, but also more innovative, collaborative, adaptable, and impactful in the long run.
Bio: Valerio Guarrasi
Valerio Guarrasi is an assistant professor at Campus Bio-Medico University of Rome, where he works in the Department of Engineering’s Unit of Computer Systems and Bioinformatics. He has a PhD and master’s degree in data science, and a bachelor’s degree in management engineering from Sapienza University of Rome.
His research centers on multimodal deep learning and generative AI, focusing on developing innovative methodologies for biomedical applications. He is particularly interested in how AI can enhance healthcare by improving diagnostics and patient outcomes. Guarrasi has contributed to several high-impact projects at the intersection of AI and medicine, including topics such as AI-driven medical imaging, synthetic data generation, and multimodal fusion strategies—efforts that have advanced the field and earned recognition through wins in international AI competitions.
Guarrasi is also a dedicated lecturer, teaching courses in programming, machine learning and big data, deep learning, and generative AI. He supervises a diverse group of students—ranging from PhD candidates to master’s and bachelor’s students—all within the realm of AI in health and life sciences.
His academic and professional experiences span multiple countries, including South Korea, the United States, Sweden, and Italy, giving him a rich, multicultural perspective that informs his research and collaborations.
Dig Deeper
To learn more about Guarrasi’s professional activities,
Each week over the next few months, Tech News will highlight different Top 30 honorees. For a full list, see Computing’s Top 30 Early Career Professionals for 2024.
In addition to Computing’s Top 30, IEEE Computer Society offers many other awards; to read
about the honors and the honorees—and perhaps nominate an impactful professional in your life—visit the IEEE CS Awards page.
