Toronto team builds collaborative network to tackle AI in digital pathology
By
Jenni Bozec
In the ever-evolving world of diagnostic medicine, Dr. Lingxin Zhang sees both challenge and opportunity in the rapid shift toward digital pathology and artificial intelligence (AI). As a pathologist at Mount Sinai Hospital and Assistant Professor in the Department of Laboratory Medicine and Pathobiology (LMP) at the University of Toronto, Zhang is on a mission to help her colleagues adapt and thrive in this new landscape.
“Artificial Intelligence in digital pathology (AI in DP) isn’t merely about adopting new tools – it’s about cultivating strategic preparedness,” Zhang says, referencing LMP’s 2023-2028 Strategic Plan objective to integrate AI into diagnostic workflows.
The widespread adoption of digital pathology has sparked a surge in AI algorithm development. “AI has the potential to streamline diagnostics, increase accuracy, and ultimately make our work more reliable and objective. But only if it’s built with pathologists in mind.” Zhang firmly believes the implementation of AI in digital pathology should be pathologist-led, while addressing the operational needs of pathologists across academic and community centres. “For instance, AI may assist community pathologists with decision making and determine when ancillary studies or expert consultation is required, while AI-enabled automation can help subspecialty pathologists with repetitive quantitative tasks – from mitotic counting to Ki-67 scoring,” (methods used to assess the proliferative activity of cells in tissue samples).
“The nature of pathologists’ work positions us as both end-users of digital pathology systems and perpetual clinical problem generators.” It is vital, Zhang says, to have pathologists involved in the development of AI systems from the start to ensure trust and buy-in.
However, most pathologists don't have a background of computer science. They need to work with computer scientists who have AI expertise to help each other understand the various fields. That’s precisely the driving force behind a new Community of Interest (COI) she’s helping to establish at the University of Toronto. Alongside colleagues Dr. Fang-I Lu, Dr. Susan Done, Dr. Ioannis Prassas and Dr. April Khademi, Zhang hopes to bring together clinicians, scientists, and AI experts to tackle the most pressing challenges in the field.
Creating this space for dialogue is one of the aims of the COI, alongside discussing issues such as regulation, data privacy and ethics, integration with digital pathology systems and interactions with commercial software. They also aim to explore the change management involved. “It’s not just about us as doctors. This affects everyone - histology technicians, admin staff, trainees. We want tools that are easy to use, accessible, and truly helpful.”
“There’s a lot of brilliant work happening across hospitals in the Toronto university network,” Zhang says. “But we’re often working in silos. We want to bring these efforts together, discuss how we can share resources, and speak with a collective voice.”
The community of interest held its latest meeting on May 28, with speakers and fellow members of the Temerty Centre for Artificial intelligence Research and Education in Medicine (T-CAIREM)Dr. Anne Martel (a Professor of Medical Biophysics at the University of Toronto) and Dr. April Khademi (an Associate Professor in Biomedical, Electrical and Computer Engineering at Toronto Metropolitan University). Martel discussed her work with Pathologist, Dr. Sharon Nofech-Mozes at Sunnybrook Health Sciences Centre on creating a model for measuring residual breast cancer after neoadjuvant therapy to help predict survival rates. Khademi shared her career path shift from industry to academia for AI research, as well as her collaborative work with U of T pathologists on an award-winning Ki-67 algorithm to enhance inter-observer agreement.
The collaboration across hospitals in the U of T network is important for access to local data in training AI models. “What works in New York or Japan may not apply here in the Greater Toronto Area,” Zhang explains. “Our disease spectrum, our workflows - even the way we practice - are different. AI needs to be trained on local data to truly be useful and for us to be able to serve our local populations.”
Zhang is also realistic about the pressure from commercial AI vendors. She argues that academic centres like U of T must act quickly, or risk being left behind. “Our window of opportunity remains open, but if we don’t develop our own tools now, we’ll end up having to purchase whatever’s available - and it may not fit our needs.”
The challenges involved in bringing true digitization to the field, supported by well-designed AI tools, are broad and multi-faceted but Zhang hopes the COI will be one way pathologists and researchers can work together to find solutions.
“We want to make this a fun and engaging process. We’re not here to provide all the answers. We’re here to bring people together and hopefully we will find some of these solutions together.”
