Picture a world where diagnoses are swift, treatments are precise and healthcare is personalised to an unprecedented degree. For clinicians and patients alike, advanced artificial intelligence (AI) and robotics tools stand to transform every aspect of modern medical practice.
Today, hospitals are looking to AI to automate administrative tasks like staff scheduling, inventory management and patient records, freeing healthcare workers to focus on engaging with and caring for patients. Meanwhile, inside operating rooms, robotic assistants are helping surgeons perform complex procedures with new levels of accuracy and efficiency. These AI-driven machines can aid minimally invasive surgeries through button-sized incisions, reducing the risk of human error and infections, as well as speeding up recovery times.
With a passion for biomedical AI and robotics, former A*STAR National Science Scholar Gaurav Manek is taking on the challenge of building responsible AI-driven algorithms and hardware that will improve quality of life for all stakeholders in the healthcare ecosystem. Manek is among a community of scientists that aims to bring out the best of these new technologies, which need to be reliable and cost-effective, as well as be able to navigate complex healthcare settings.
Manek’s current work focuses on exploring biomedical AI solutions that can streamline tedious workflows in medical diagnostics and provide surgeons with real-time, precision-enhancing insights. In this interview with A*STAR Research, Manek reflects on his lifelong love for computers; shares details of his work on building ‘smart’ microscopes; and offers his perspectives on approaching scientific inquiry with care.
Q: What drew you to computer science and biomedicine?
My interest in computers goes way back; I started learning programming when I was eight years old, and it was a hobby I enjoyed throughout my school years. A pre-university internship at A*STAR’s Institute for Infocomm Research (I2R) sparked my passion for computer vision—the use of AI to enable machines to interpret visual information.
From there, I was inspired to pursue research in robotics as an undergraduate student and later—as a National Science Scholar (PhD)—to work on reinforcement learning (RL). In RL, AI agents interactively learn to make a series of decisions to best achieve a long-term goal. This is applicable in many areas: games like chess and Go, automated driving, industrial control and so on. It’s an important training method to create systems that can emulate human intelligence.
During my PhD, I also had the chance to build a small AI Software-as-a-Service (SaaS) startup, which ignited my entrepreneurial spirit. Now, at A*STAR’s Institute of Molecular and Cell Biology (IMCB), I’m working on AI and robotics for medical microscopy—integrating hardware, software and AI models into a commercial product.
For me, the main draw of biomedical AI and robotics is its interdisciplinary nature; I enjoy translating ideas from one field to another. I’m also attracted to the challenge of building AI models and hardware that we can trust. If we succeed, I believe they will rapidly improve the quality of patient care when implemented in real-world settings.
Q: What can AI and robotics bring to biomedicine?
While AI isn’t quite ready to take over medical decision-making, it doesn’t need to do so to be a significant help to us. By automating the more repetitive tasks involved in skilled jobs, it can allow healthcare workers to be more productive.
Imagine a microscope capable of scanning tissue samples for unusual cells and highlighting regions of interest. It would let pathologists focus on interpreting the images, vastly reducing workflow times. Many such opportunities for better efficiency are waiting to be explored.
It’s also worth noting that AI integration into biomedicine presents challenges that require a deep understanding of models, rigorous training and explainable decisions—elements that are essential for building trust. There are also many hurdles involved in using robotics in microscopy, but by applying newer techniques in optics and mechanical engineering to complement those in robotics, we can make microscopes cheaper, better and more autonomous.
Q: Tell us about your current work at IMCB.
We’re developing an advanced microscope for intraoperative margin evaluation during cancer resection surgery. This means that surgeons operating on tumours will be able to check—in real time—whether all cancerous tissue has been removed during a procedure, and continue removing more if needed.
Such tests are currently performed by freezing and finely slicing tissues, which is a relatively complex process with many pitfalls. Our device would bypass the freezing step, reducing the time needed to obtain an image. Surgeons can therefore make more precise, conservative cuts that preserve as much healthy tissue as possible. For operations involving certain tissues such as breast, prostate and colorectal, this could significantly improve patients’ quality of life post-op.
I’m working on the hardware, software and AI for this project alongside a brilliant and motivated team. We have filed for patents for the hardware, and developed new software approaches for it; our goal is to form a spinoff company to commercialise the technology.
Q: What other interesting projects in AI have you worked on?
As mentioned, my PhD thesis focused on the theory of RL; this kind of AI deals with learning to achieve long-term goals by interacting with dynamic environments. It’s a lot like how pet dogs learn what we want them to do based on when we give them treats.
While RL was intriguing to me, I found myself casting about for some real-world applications, and I happened across what’s known as the ‘bin-packing problem’ in combinatorial optimisation. In this, we try to arrange as many objects as possible into some fixed space. While the problem’s name suggests we’re packing physical objects, it’s actually analogous to problems encountered when scheduling meetings, assigning resources and many similar applications.
By applying some insights from classic computer vision, we redefined the way we represented these problems and greatly reduced the time to solve them; this formed part of the AI SaaS product that I referred to before. Although I’m no longer managing the company, the product is still being used for scheduling meetings and events.
Q: What advice would you give those pursuing research as a career?
Approach your scientific endeavours like a hedge fund portfolio: objectively assess each project for its risks and potential returns.
Scientific inquiry is fraught with risk. Some problems may be unsolvable, others might be solved more efficiently by others, and some solutions might not be useful in practice. Each problem you take on is, in a way, a gamble; you’re betting that you can beat others to the best solution, and that that solution will be helpful to others. Therefore, it’s crucial to carefully evaluate the resources—both of time and money—that you invest in each project, and balance your efforts across high- and low-risk ventures.
One of the hardest things to do is to walk away from a difficult problem you’ve invested years into, especially if you’re convinced the solution is right around the corner. Keeping a bunch of low-risk projects running can help ease this transition; it gives you a sense that you have something else to show for your dedicated efforts.
