Drs. Ren Yuan (left) and Calum Mac Aulay (right) are recipients of the 2021/2022 Precision Health Catalyst Grant award for this project titled “An AI model to predict future lung cancers risk with low-dose screening CT”.
Read a summary of the project here.
“Our pilot project is to use AI tools to identify sub-visual changes in the “normal” lung before cancer develops and build a prediction model to predict the risk of future cancer in a “visually normal” lung within 1-2 years.”
Can you tell us about your interest in lung cancer and what are the challenges you are trying to address?
Lung cancer remains the leading cause of cancer deaths globally in both genders. Its five-year net survival is among the lowest, primarily due to most diagnoses being made when the cancer is in its incurable late stages. Two large sufficiently powered randomized trials showed a significant mortality reduction from lung cancer screening in high-risk ever-smokers using Low-Dose CT. Population based lung cancer screening has been implemented in BC.
Interval/incident cancer is a rising challenge for lung cancer screening programs. These are new cancers that occurs in between scheduled routine annual or biennial screening or found on the next routine surveillance screen where the patient had a prior negative scan. Compared to prevalent cancers that are found on the baseline CT, interval/incident cancers are usually biologically more aggressive, often found at more advanced stage, hence have poorer prognosis. The majority of screening CT scans have no or very small low-risk lung nodule(s) that do not require early recall CT scan. Identifying Interval/incident cancer in these individuals for early recall CT and intervention can improve the patient outcome. In addition, people who are identified with lower risk can have the next surveillance chest CT scan 2 years later with implication for reduction in resource utilization and costs.
How will your precision health research project address these challenges?
State-of-the-Art imaging analysis using AI tools has been investigated in prevalent cancers; what has not been done is to predict future cancer risk in “visually normal” lung, where interval /incident cancer will develop later. Our pilot project is to use AI tools to identify sub-visual changes in the “normal” lung before cancer develops and build a prediction model to predict the risk of future cancer in a “visually normal” lung within 1-2 years.
Predicting individual future cancer risk in each participant has a significant impact on lung cancer screening. It will enable precise/ personalized screening intervals (e.g., short intervals for biologically aggressive cancer and longer intervals for low-risk individuals) to improve the screening outcome. This pilot project will lay the groundwork to prospectively evaluate the utility of this AI algorithm using the BC Lung Screening Program that has launched across BC in April of 2022 and for national and international collaborative studies with other research groups.
What results have you seen so far?
RY: With the funding support from the UBC Faculty of Medicine’s Precision Health Catalyst Grant, we were able to rapidly begin the development of AI-based models. With the collaborations with Dr. Stephen Lam, who has led provincial, national and international lung cancer screening trials, we have investigated the longitudinal imaging data to decide eligible cases and analyzed their CT scans. To date, we have developed two models (Radiomics -Machine Learning model and Deep learning CNN model). We have trained and tested these models with our data, and the performance of both models is encouraging. We are applying the best-performing AI model to both lungs on an overlapping patch-by-patch basis for each CT to generate a map containing the risk score from these patches to predict the future cancer risk in the lungs and for each subject. The preliminary data have been accepted for the 2023 World Conference on Lung Cancer for oral presentation.
CM: An interesting observation from both of the two models (Radiomics -Machine Learning model and Deep learning CNN model) is that the changes (features) associated with the development of a future cancer are also changes that differentiate early small cancers from normal lung parenchyma. This observation increases our confidence that the changes we are measuring are a consequence of the underlying neoplastic process.
From your perspective, what do you think is exciting about the future direction of precision health?
RY: Precision Health is an individualized approach to maximizing health at the patient-, population-, and policy level. It involves tailoring medical care to individual patients based on their unique characteristics, and radiology plays a crucial role in achieving this goal, especially in oncology, where medical imaging is actively involved throughout each patient’s journey.
In the era of AI that bridges medical imaging and precision medicine, AI-powered tools can help in early disease detection, treatment planning, and risk assessment. There are exciting areas for radiologists in precision cancer care, such as: identifying subtle imaging changes that might indicate preclinical stages of diseases, enabling early detection and prevention; moving beyond qualitative assessments and incorporating quantitative measures for precise diagnosis and treatment monitoring; transforming radiology into digital-biomarker to assist accurate predictions; risk prediction to allows for early interventions and personalized preventive strategies, and many more.
Overall, the future direction of precision health in radiology provides radiologists with opportunities to contribute more directly and actively to patient care through innovative technologies, interdisciplinary collaboration, and a focus on early detection and tailored treatment strategies.
CM: The integration of multiple axis of information (patient clinical information, immune exposure history, CT data, spatial biology, gene expression and gene structural changes) covering the multiple scales neoplastic development (nano scale: base pair modification, gene structure changes, micro scale: nuclear organization, organelle and cell-cell spatial arrangements and macro scale: various imaging methodologies) will transform Precision Health. This will be accomplished through enhancing our understanding of the multiple neoplastic pathways of cancer, determining which pathway(s) the cancer is exploiting and the specific lesion’s position within these pathways. Most exciting, enabling the clinician to select specific targeted treatment strategies tailored to a lesion’s pathway(s) maximizing treatment benefit for the patient.
About Dr. Ren Yuan
Dr. Ren Yuan completed her medical school and radiology residency in PR China before moving to Canada. She completed a Ph.D. degree in quantitative imaging analysis of lung structure in smokers at the University of British Columbia, followed by clinical radiology fellowships at UBC and U. McMaster.
Dr. Ren Yuan is an oncological radiologist and research lead at Diagnostic Imaging, BC Cancer Vancouver Center, and clinical associate professor of Radiology, UBC. She is a member and clinician scientist with the Lung Cancer Program, Integrative Oncology at the BC Cancer Research Institution. She co-directs the Radiomics Core of the philanthropically funded PRO-Lung project at BC Cancer. She is a recipient of the 2022 Michael Smith Health Research BC Health Professional Investigator Award.
Ren’s current research focuses on clinical applications and implementation of artificial intelligence in lung cancer imaging. Other academic interests include investigating state-of-the-art imaging tools to assist precise and personalized oncology.
About Dr. Calum MacAulay
Dr. Calum MacAulay obtained a BSc in Engineering-Physics, a MSc Physics from Dalhousie University and his PhD in Physics from University of British Columbia. He is currently a Distinguished Scientist and is the head of the Department of Integrative Oncology at the BC Cancer Research Institute as well as an Associate Professor in Pathology and Laboratory Medicine and Associated Member of the Physics and Astronomy departments at UBC. His research has concentrated on the early detection and treatment of cancer using quantitative imaging tools in microscopy, photon – tissue interactions and understanding the genetic and molecular events driving the neoplastic process. The teams he has worked with have had a strong drive to translate their work into clinical tools and processes. Their work has led to the clinical introduction of clinical tools using tissue autofluorescence for the detection and assessment of lung and oral cancers as well as automated image cytometry systems for cervical screening and oral cancer screening.
Currently his research is focused on using machine learning applied to Lung Cancer CT screening, the in vivo optical biopsy of early cervical, fallopian, lung and oral cancers using multimodal optical probes, digital pathology tools and recent advances in machine learning coupled with spatial biology to evaluate the biological aggressiveness of neoplastic tissue.