Prof Paula Petrone. Image: Lorenzo Raschi Petrone
Prof Paula Petrone, of the Barcelona Institute for Global Health, warns that AI is not without fault and models must be trained to avoid biases and encourage citizens’ trust in scientific research.
“In a sense, I believe all scientists could be seen as data scientists,” says Prof Paula Petrone. “We analyse, uncover patterns and draw conclusions using data to understand the world around us.”
When she began her career, machine learning and data science were not as popular as they are today. Petrone was always fascinated by “coding, physics, biology and math”, and knew she wanted to work with data to uncover information about human health and disease.
Petrone is very active in the science community in Barcelona where she organises the Women in Data Science (WiDS) Barcelona Biomedicine conference, which aims to make the work of women scientists in the field of biology, biomedicine and digital health more visible. She also acts as a digital health start-up consultant, mentor and activist on ethics and diversity issues in STEM. Recently, Petrone was a speaker at University College Dublin’s Human Health, Impact and Technology (HHIT) series.
She spent time working with pharma companies Novartis and Roche, but she is currently in academia. She leads the biomedical data science team at the Barcelona Institute for Global Health (ISGlobal). “Having transitioned from industry to academia, I can confidently say that most of my main responsibilities resemble those of a similar R&D position in industry. As team lead, I focus on ensuring that the team is motivated and that our research is robust, innovative, impactful and of course, well-funded.”
ISGlobal’s research centres around three main themes, says Petrone. One is the analysis of real-world evidence from hospitals to assess disease risk and understand what factors keep people healthy. Another area is biomedical image analysis, ranging from microscopy imaging to non-invasive ultrasound technologies. The scientists on her team are also looking to expand their expertise in analysing what Petrone describes as “unstructured medical histories” using natural language processing techniques.
The trouble with data shortages
A problem that Petrone and her team have encountered is the lack of patient data and the difficulty in obtaining it. Data is often unwieldy at the best of times, but the fact Petrone is working with data for medical research purposes makes it a little bit more complicated.
‘To address the challenges of obtaining patient data in the lab, we have started investigating the potential uses of synthetic imaging data’
She has found a workaround thanks to a technology that everyone is talking about at the moment – generative AI. “To address the challenges of obtaining patient data in the lab, we have started investigating the potential uses of synthetic imaging data. Previously, I was unsure about how synthetic data could enhance predictive models, but with the rapid advancements in generative AI, I now recognise its potential and I believe this will soon be an important aspect of our imaging models,” she says.
When it comes to diagnostics, AI can help recognise things that the expert human eye cannot. Petrone and her colleagues are developing algorithms for the detection of pathology in non-invasive ultrasound imaging. “We observe that AI can sometimes detect the signature of disease even when this is not obvious to a trained expert,” she says, adding that this holds “great promise” for the deployment of diagnostic non-invasive and affordable ultrasound devices in clinical settings that either have a lack of radiologists or a lack of facilities. Petrone says her team is working on applications for infectious diseases, cancer and tuberculosis, and they are open to new clinical collaborations to advance their work.
Citizen engagement matters
But Petrone is not just interested in AI’s potential; she is also passionate about encouraging more people to engage in scientific research. “By actively involving citizens in our research, from the initial design of the project to the interpretation and conclusions of the study, they may be more likely to contribute their information,” she points out. “This approach could help raise a sense of ownership and trust, as well as promote a deeper understanding of scientific research among participants.”
The principle of trust and the importance she places on scientists like her having the trust of citizens is clearly important to Petrone. Trust and ethics are also factors in her ongoing research informed by AI. She is fully aware that AI is not perfect and that many AI models are trained on retrospective data which perpetuates existing biases and inequalities.
As she says, “Such biases can contribute to disparities in healthcare outcomes, particularly for underrepresented populations”. Therefore, healthcare researchers have to be careful to consider differences in things like sex and gender as these variables can have a predictive value.
‘By actively involving citizens in our research, from the initial design of the project to the interpretation and conclusions of the study, they may be more likely to contribute their information’
Petrone is “very excited” about recent developments in explainable AI, which she says is about making underlying AI systems understood by users. This is an important step on the way to demystifying a much-hyped technology.
In the context of biomedical applications, Petrone says she believes that all algorithms used should be accompanied by explanations of how these systems function and why doctors should trust them at the time of making critical medical decisions. That way, research using tech in healthcare can be trustworthy and transparent. This belief is very much a reflection of her assertion that all scientists are data scientists. With bad data comes bad decisions and outcomes.
10 things you need to know direct to your inbox every weekday. Sign up for the Daily Brief, Silicon Republic’s digest of essential sci-tech news.
