While magnetic resonance imaging (MRI) and computed tomography (CT) are indispensable diagnostic tools in modern medicine, they rely on costly infrastructure that keeps advanced imaging largely inside hospitals. Professor Guo Xiaoqing, Assistant Professor in the Department of Computer Science at Hong Kong Baptist University (HKBU), believes ultrasound offers a path toward greater accessibility in community clinics, remote areas, and other non-hospital settings. Under her leadership, the UltraVision+ Lab is building AI-powered systems that support clinicians throughout the ultrasound scanning process, making this affordable imaging technology easier to adopt at scale.
Professor Guo's path into medical imaging began during her undergraduate studies at Beihang University's School of Biological Science and Medical Engineering, where she first encountered the field and developed a foundational interest in AI for biomedical imaging. She went on to pursue her PhD at the City University of Hong Kong, where she systematically built her expertise in AI for medical imaging, working on endoscopic, CT, MRI, and dermoscopic analysis using open-source data. Yet the more she learned, the more she sensed a gap between technical research and clinical reality. "Toward the later stages of my PhD, I started asking how the models and metrics we developed could translate more directly into real clinical decisions," she recalls.
That uncertainty gave way to a clearer sense of purpose during her postdoctoral work in the Department of Engineering Science at the University of Oxford from 2023 to 2025, in affiliation with Professor Alison Noble’s research group. Its extensive network of clinicians and industry collaborators helped Professor Guo see a concrete way forward: rather than working on well-defined tasks where the problem formulation is already established, she could start from clinical pain points and distil them into research questions that AI can address. The experience reinforced her belief that technical proficiency alone is not enough: understanding real clinical workflows is essential to asking the right questions in the first place.
This clinically grounded mindset now drives Professor Guo’s research group, the UltraVision+ Lab, whose name reflects its mission of combining ultrasound and computer vision. The “Plus” carries a dual meaning : it refers to the multimodal signals that accompany scanning — spoken language, visuals, probe motion, and other spatial information — and to the role of AI as a collaborative partner alongside the sonographer. The goal is to build an intelligent co-sonographer: a system that works with clinicians rather than replacing them. She explains the longer-term vision through a simple analogy: “Blood pressure monitoring once required a clinic visit. Today, people can measure it at any time and in any place with a device bought over the counter. Ultrasound could follow part of that trajectory.”
While ultrasound offers an affordable and portable alternative that can make medical imaging more accessible in different settings such as smaller clinics and community health centres, it also presents unique interpretive challenges. Unlike CT or MRI, which can provide three-dimensional (3D) anatomical views, ultrasound is built from sound-wave-based two-dimensional (2D) frames. As a result, clinicians need experience to interpret a sequence of 2D slices accurately. Tiny movements can also dramatically alter the image and affect the result.
To address these challenges, Professor Guo has set her sights on tailoring AI tools to guide clinicians to administer ultrasound scans for patients at healthcare settings that may not have the resources of large hospitals.
At the heart of this work is Professor Guo’s ambition to build not just a model that reads images, but an AI-powered collaborator that supports the entire scanning workflow — from pre-scan guidance and real-time assistance during acquisition, to post-scan summarisation. “If we only train on open datasets without a solid understanding of the clinical reality, the model may look good on paper but fail in practice,” Professor Guo stresses.
To support clinicians across the full workflow, the UltraVision+ Lab spends significant time in collaboration with practitioners and professors at Prince of Wales Hospital, carefully discussing what data to collect and how. The aim is to capture as many signals involved in the scanning process as possible — video, audio, probe motion, spatial information, and text reports — so that the AI can develop a comprehensive understanding of what happens during a session. However, collecting such multimodal data is far from straightforward. As clinicians cannot always speak while operating the probe, audio and video can easily fall out of sync, and considerable effort is needed to align and validate these signals. “The process of defining research questions and setting valid parameters for results benchmarking is also a complex one, requiring detailed discussions with the clinicians,” Professor Guo adds.
These extra efforts make the difference between a tool that works in one controlled setting and one that withstands the variability of everyday practice.
Professor Guo began her work in linking ultrasound scanning with multimodal learning in a reliable way in January 2023 – a time when large language models (LLMs) and ChatGPT were beginning to reshape public expectations of AI. Given how quickly these tools have become part of daily life, she believes the pace of AI development will accelerate significantly in the coming years. “We are currently working at a slower pace,” she says. “I believe the pace will pick up over time with greater industry involvement.”
Despite taking a more time-consuming approach to ensure AI models are tailored for clinical reality, Professor Guo has been making solid progress. One recent example is the Lab’s work on “Sonomate” – an AI assistant designed to support users during foetal ultrasound examinations, which was published in the acclaimed journal Nature Biomedical Engineering in 2025.
Moreover, she has earned prestigious recognition, including her selection as one of the Top 80 Young Female AI Scholars by Baidu Scholars in 2023, the Rising Star of Women in Engineering in Asia award from the Asian Deans’ Forum in 2024, and the Medical Image Analysis (MedIA) Best Paper Award from the Medical Image Computing and Computer Assisted Intervention Society (MICCAI) in 2025. Yet Professor Guo remains grounded, believing that the most important thing is to keep working in an area that genuinely interests her.
For Professor Guo, sustained effort begins with curiosity. “Interest is the most important thing in research,” she says. “I was energised by medical imaging research from my undergraduate years. My postdoctoral work restored a sense of direction. Now, my focus is on taking the time to understand problems properly, especially through conversations with doctors that often take longer than training a model on open data. It is a worthwhile effort.”
That philosophy also shapes how she recruits students. Rather than selling an idealised vision of AI, she explains the realities of ultrasound research – data collection, clinical coordination, and careful validation – and looks for people who are genuinely curious about those constraints. With the right team, Professor Guo and the UltraVision+ Lab are poised to put ultrasound one step closer to becoming not just a specialist technology, but a more accessible part of everyday care.
