Part seven of our AI Researcher Profile series invites Associate Professor Fusheng Wang of the Department of Computer Science and of the Department of Biomedical Informatics, to speak on his profound research efforts which are currently making inroads toward the ongoing opioid epidemic.

AI: What types of research interest you, and what inspired you to get involved with this research?

FW: I work on a couple of AI-related research efforts. One of these research areas is 3D digital pathology. 3D Pathology is an imaging technique which allows for more in-depth analysis of human tissues. We tried to build 3D pathology technologies beyond 2D so that we can have a better representation of the knowledge with respect to spatial structures of 3D spaces. We try to integrate tissue images of different types at different scales that can be oriented into different ways while pairing different types of information, then we can understand the images. This understanding allows for more apt predictions surrounding disease progression, diagnosis, treatment response, and so on. Another major area of my research is AI-based opioid epidemic research. The opioid epidemic is a huge problem for the United States. The epidemic causes a lot of overdose cases, especially during COVID-19, where overdose statistics reached a historic record in the past year. My research is trying to provide early predictions for opioid risk in patients using patients’ electronic health records (EHR), including opioid use disorder and opioid overdose. Through this, we are able to predict the risk of each individual patient at a clinical visit, so that we can evaluate the potential risk for better intervention to prevent such risks. This is an ongoing project.

AI: With respect to your research surrounding the opioid epidemic, how does artificial intelligence play a role in this?

FW: The technology we use for the predictive model which is based in a machine learning and deep learning approach. We are using a nationwide EHR database which consists of 89 million patients. In order to appropriately analyze these patients through AI, our predictive model uses state-of-the-art methods which  combine graph neural networks and sequence based architectures. Through this, we have the best prediction performance compared to other methods. Our performance is significantly better than other approaches. Because our method is so effective, we are currently working on imposing a tool with our predictive models to be adopted in hospitals through collaboration with an EHR vendor.

AI: How does your current research impact our future with respect to the opioid epidemic?

FW: Right now, from the clinical side, manual-based approaches to assess risk of each individual patient are very limited. At Stony Brook, they have a risk assessment tool which is trying to extract past medical history and other factors, which is based on some prior knowledge. These factors are not concrete. Therefore, they only capture a very small population at risk. For us, the model is much more advanced, and we can take advantage of all of the information from a patient’s EHR history. We can even incorporate clinical notes. Our model is automated, which means that the clinician does not need to manually define rules for generating a risk report for each patient. In addition, our method allows us to generate a list of explanations for each patients’ risk factors. Ultimately, our model provides much more information for assisting clinicians to make decisions.

AI: Your research stretches beyond this. Please tell me about your work with database technologies.

FW: One major area for us is building big spatial data management systems. We are currently working on a collaborative project funded by the Human BioMolecular Atlas Program (HuBMAP). This program is an initiative to generate an atlas of the human body at extreme scale  - cellular and subcellular levels. Eventually we can extract and represent trillions of 3D cells from human bodies. To manage and process such 3D cells, it is not possible with traditional database technologies. So, we have developed a very highly scalable and efficient 3D data management technology. This technology allows us to quickly manage, query and visualize such 3D objects. We also apply this to 3D digital pathology. In 3D digital pathology we have a lot of spatial information such as nuclei, cells and vessels and we want to find the spatial relationships and patterns between such objects. Our technology allows for tens of millions of 3D objects per tissue volume to be processed, which is extremely challenging for traditional database systems. We have been working on spatial big data management systems for more than one decade, so we are really state-of-the-art for such systems.

AI: You also are involved in summer programs in Computer Science and Biomedical Informatics. Could you elaborate on this?

FW: I am leading a summer research program for high school students called the Computer Science and Informatics Summer Research Experience Program. The 6-week program has been running for four years with great success. This past summer, of 93 candidates nationwide, 21 students were offered a seat in the program and participated in such. In addition to the students, 14 faculty members also participated in this program, working with students on their research projects. The projects given to the students are well defined research driven projects. We also provide PhD mentors to supervise each individual student as they work on the research topics. Most of the research projects are related to AI. By the end of the summer, after all of the students’ hard work, we have a symposium on the last day. The last day is truly so impressive. This year the program existed online but normally it is in person. 

AI: What do you suggest to students looking to get involved in artificial intelligence?

FW: The best way to get involved in AI is to define problems you’d like to address where you have access to the respective real-world data and applications. If you have a problem with data, it’s very easy to get started. I currently work with high school students through graduate students in addressing problems and preparing these students in getting data; in particular, getting data from the Stony Brook Hospital which is a complicated process. For aspiring students in AI, being able to define concrete problems with available data is the best way to get started quickly.