Connecting Artificial Intelligence and Healthcare: An Insight into the Biomedical Computer Vision Lab

Are you a person with a passion for AI and a curiosity about its applications in analysing biomedical images? Dive into the fascinating world of the Biomedical Computer Vision Laboratory at the University of Tartu Institute of Computer Science, led by the researcher Dmytro Fishman. In this interview, we explore - the lab's research focus, collaborations with industry partners, and the opportunities available for students interested in joining this innovative lab.

Discover how this lab is making strides in using AI to improve healthcare and explore career opportunities in machine learning, computer vision, and more. Whether you are a student or simply curious about the intersection of technology and medicine, this interview provides valuable insights into the exciting work happening at the Biomedical Computer Vision Laboratory.

Dmytro Fishman is the head of the Biomedical Computer Vision Lab and a machine learning/artificial intelligence lecturer. Originally from Ukraine, he completed his master's degree in Software Engineering at the University of Tartu and subsequently earned a Ph.D. here. He is a passionate scientist who combines professional areas in computer science and biomedicine. Dmytro is also a co-founder and Chief Science Officer at Better Medicine company, which builds AI tools that help radiologists diagnose, detect, and measure cancer lesions.

Could you provide an overview of the Biomedical Computer Vision Lab at the University of Tartu?
The lab is characterised by its flat structure, mainly consisting of a group of students and me. The group mainly consists of master's students, a few PhD students, and, very occasionally, bachelor's students. The lab operates with a simple, non-hierarchical structure, where my role is to assist students in completing their projects and ensure the successful execution of collaborative projects with industry partners. Our primary focus is training and developing AI systems that are able to understand various biomedical images, ranging from microscopy images of cells to images of tissues and organs.

What are the main research projects that your lab is currently working on?
One of our projects is trying to generate synthetic data that would be useful for training AI models. A significant obstacle to using AI in the medical field (especially in imaging) is a lack of data. It's a huge problem because normally, people don't go around with their CT scans saying - “Hey, do you want to use my CT scan? I have a tumour in my lungs.” They try to keep their health issues private - so getting such CT scans is very difficult and requires a lot of regulatory approvals.

To train the Al model, we need to indicate the places that we are interested in. Here arises another problem - finding an expert who can point out where the tumour is on the scans, i.e., a radiologist - is expensive and quite difficult. Instead, we try to build an AI model similar to Midjourney based on a generative adversarial network or stable diffusion. For example, we would request - “Hey. I want a scan of a male patient of about 40 years old with a huge tumour in the left kidney”. And it would generate a scan according to our specifications. And then, you can use this synthetic scan as a basis for training AI models that would look for kidney tumours. You would not need to pay anyone, engage anyone, or wait months until the work is done.

We also have other projects, such as building systems that can detect tumours from CT images. We do a lot of work with microscopy images (on a cellular level) to detect and count cells and help biologists analyse those images.

Can you share a recent success story or breakthrough from your lab's efforts?
I have a PhD student who completed several pretty ambitious projects that have been published in Nature Scientific Reports, and after that, my PhD student got a job offer from a company called Syngenta, which is an international company studying plants. Now, he is one of the few people in the company who actually understands both AI and imaging. I think it's a pretty good example of how, by doing research for your PhD, you can get into high positions in big companies, get pretty good pay, and work worldwide (my PhD student works in the UK now).

Are there specific courses related to your lab's research focus?
I usually expect students to take my Machine Learning course (MTAT.03.227 fall semester) and Neural Networks (LTAT.02.001 spring semester). Currently, I am working on a new course, which is going to be called “Deep Learning for Computer Vision”. Taking this course will definitely be quite useful, too.

What hands-on experiences or lab work can students expect to be involved in?
Everything! They can expect to get involved in absolutely any activity that we ourselves are engaged into. It's a massive spectrum of things. We are not developing the product, but we are the ones who are developing the technology. We are very hands-on and very production-oriented and work with real companies. At the same time, you can do a lot of more theoretical research, like generating scans and seeing what happens. There is a lot of room for freedom. We can do all sorts of intriguing, different, interesting stuff. But at the same time, once we see a value in that, we are also interested in making that useful downstream for companies, people, etc. On the one hand, we generally aspire to create production-ready code and, at the same time, do a lot of research.

Are there any notable collaborations of your lab with industry partners or other research institutions?
The entire lab is about industry collaborations. We are extremely applied. For example, we have one collaboration with a company called Revvity. They are developing microscopy hardware and the software that comes with it. We helped them to integrate AI into their software. When biologists do images of their cell lines, they also want to get the analysis done. Besides the image, I want to know what is in my image and how many things are there. Usually, it is done through classical computer vision algorithms such as edge detection and simple thresholding. Thus, we have assisted them in the integration of the recent AI deep neural networks into those pipelines. This collaboration has been ongoing for seven years, which is a significant duration for an industry partnership, given that companies often change their goals and perspectives. Despite these changes, we have successfully maintained a strong and lasting relationship.

The other collaboration is with my own company, Better Medicine. We help this company build some of the technology to detect cancers, measure tumours, etc. We are also on the way to signing the contract with Syngenta. It will be the third collaboration with the industry.

We also have academic partners such as local Estonian hospitals, including Tartu University Hospital and Tallinn East Central Hospital, where we do things pro bono. For example, we help pathology labs automate their processes because, at the moment, pathology labs do everything manually. This means that people spend their hours counting cells from the images rather than running the AI that can do it in a few seconds. We have a partner from the UK called Wellcome Trust Sanger Institute, where we helped to quantify some of the microscopy experiments (integrating AI into their processes).

How does your lab support and nurture entrepreneurship among students?
I created my own company when I was still a PhD student. My students are perhaps not yet ready to create companies, but I would definitely encourage and support their initiatives. When I conduct my lectures, I always say that we need more companies like Better Medicine. First of all, in Estonia, there are not many of them; secondly, globally, there are not many of them, and we need more because we need more competition. Because competition drives progress, and progress helps us all to live longer.

What career opportunities do students who engage with your lab often pursue after graduation?
Lab work means applying computer vision, machine learning, and AI. Graduating from this lab means heavy experience with machine learning models, deploying and building, etc. Estonia does not yet have much of a market for computer vision experience, but I think it will be growing soon. Nevertheless, machine learning has taken on. So, you can expect yourself to be competitive in Machine Learning, Data Science, and Computer Vision. If you aspire to become a computer vision engineer/machine learning engineer, this is the right way to go.

What specific qualities or skills do you believe are most crucial for students interested in joining your lab?
We expect students with a strong interest and some experience in machine learning (building and training models) who are excited about this field. Basic knowledge of biology is a plus but optional. Our students typically come from the Computer Science or Software Engineering programmes. Our lab primarily focuses on practical applications rather than deep biological research. We aim to teach machines to identify patterns like biologists and radiologists do. If you are excited about the opportunity to contribute to cutting-edge research that bridges computer science and biology, we encourage you to join us on our exciting journey.

Author of the text: Viktoriia Abakumova, Software Engineering student at the Institute of Computer Science
If you are interested in joining the laboratory and further collaboration, you can contact Dmytro Fishman, dmytro.fishman@ut.ee