Pingfan Song

Brilliant Minds


How do we harness the use of artificial intelligence (AI) in medicine to make it work more efficiently and accurately, while at the same time following ethical guidelines? These profoundly important questions are some of the issues being tackled by Dr Pingfan Song, Senior Research Associate at the Department of Engineering, who is working to develop the next generation of AI systems for medical imaging and healthcare.



After completing his batchelor’s and master’s degrees in China, Dr Song went to study at University College London (UCL) for his PhD degree where he worked on signal and image processing, combined with machine learning.

One of his research projects was to speed up the MRI process using machine learning and compressive sensing technology. This can achieve more efficient data acquisition and compression without compromising image quality. In this way MRI machines can be used more quickly and scan more patients. He said, “For example, with such smart sampling and reconstruction strategy, we can make the imaging speed 10 times faster. That is a very promising way to treat many more patients in the same amount of time.”

Following his PhD study, Dr Song went to Imperial College where he was a research associate working on an interdisciplinary project combining machine learning and signal processing with neural imaging. He said, “We collaborated with the bio-engineering department to develop two-photon light-field microscopy, a type of advanced 3D microscopy imaging system which can provide fast 3D imaging of brain cells.” The technology has been applied to a variety of virtual reality and augmented reality systems.


“The beautiful idea is to integrate advanced computation algorithms with novel imaging modalities to develop unprecedented scientific research tools.”


With light-field imaging Dr Song and his team can record all of the 3D information needed with a single image. The system uses a micro-lens array with thousands of tiny lenses to record the angles of light rays, and with the angles the depth of the 3D image can be captured with accuracy and speed using machine learning methods.


Deep Learning

Deep learning is an iconic machine learning method that has been widely used for developing incredibly powerful AI systems. It provides a highly effective way to model and analyse the complex world. But one of the issues with deep learning is that it is largely used like a black box that lacks interpretability and transparency, and therefore can be difficult to understand. This can affect reliability and cause unexpected errors.

Dr Song explained, “Sometimes the deep learning models – especially the large and complex ones – have types of behaviour that we have not fully understood and may be beyond our control. That’s one of the motivations behind why we want to develop more trustworthy AI technology with better interpretability, especially for high-stake applications, such as in the medical and healthcare domains.”

The idea is that, thanks to advances in technology, AI-based medical tools will be able to do more of the heavy lifting in the near future, and take a large burden off doctors’ shoulders, so that they can focus on more valuable and important tasks.



Dr Song is also applying trustworthy AI technology to the early diagnosis of dementia. “People start to develop symptoms of dementia maybe five to 10 years before a doctor’s diagnosis. We want to develop AI tools that use a combination of different data sets to perform a prediction of dementia. With good accuracy the model could give benefits to patients by alerting them to the illness earlier.”

Using advanced MRI and PET imaging an analysis can be made of the toxic proteins such as beta amyloid and tau, which form tangles and interfere with brain signalling, found in those suffering from dementia. The patterns that these proteins make can initially be too faint to be seen by the human eye, but Dr Song hopes that new AI systems will be able to glimpse these patterns earlier, while also helping to understand exactly how the proteins work in the progression of dementia.


Ethics and Privacy

Another area that Dr Song is also very much involved in is the conversation around machine learning and ethics. He said, “Many people have fears about AI, whether it will invade their privacy and have a malign influence on their lives, like we see in science fiction films.

“It is understandable that people have such fears, considering that we are facing a new AI era. AI tech is developing fast while we are not sure whether we are fully prepared. On the other hand, we should be rational, neither overly optimistic nor overly pessimistic. We are still at the early age of AI even though AI has accomplished things that we didn’t think possible.”

“Current AI models can be manipulated to hack our private information, and can also be fooled by malicious interference to give wrong results with high confidence. That is why we are devoted to developing trustworthy, responsible, safer tools to tackle those issues, and make AI interpretable, robust, causal, and privacy-preserving.”


“Step by step we solve problems as they arise, and in this way, we can improve technology and direct it in a better way.”


Working in Cambridge

“I enjoy my research work and life in Cambridge. This is a very inclusive environment. Everyone is friendly, and we discuss new ideas and topics and help each other. We are free to explore inter-disciplinary research we are interested in. The city provides a beautiful environment to work in. For instance in the college gardens we can consider problems and chat with friends. It is a different way to enjoy life compared with where I was before in London.”


Living in Eddington

As a member of the Cambridge Postdoc Society committee, Dr Song finds the Postdoc Centre at Eddington the perfect location to connect postdoc researchers and build a better community: “We can arrange seminars and workshops to meet other scholars to share useful and interesting life tips in Eddington, as well as discuss important topics. We are currently planning a set of activities and events, like the 2022 Showcasing Research event, pub nights, and culture talks which will take place during the next few months.

“I enjoy meeting scholars with similar and different research backgrounds to broaden my knowledge boundary, expand and strengthen my network, and help with future collaboration.”


“I like Eddington; the community is friendly, and there are beautiful green spaces and lakes.”


Dr Song added, “My family really enjoys cycling or walking around, which is good for the environment. Many postdocs live here, so it’s easy to find peers to socialise with. Eddington has a supermarket and a lovely nursery and primary school that are just a few minutes’ walk from home. These amenities make life here convenient and enjoyable.”

With all of the questions and anxieties that have arisen over the use of AI and its future direction, it is reassuring to know that there are University of Cambridge scientists like Dr Song paving the road for the development of this technology, to help rather than hinder us in the future.


Toggle navigationToggle search