McGill researcher leads the development of the first multimodal deep learning model to optimize care for individual cancer patients
Sarcomas are rare cancers that develop in the bones and soft tissues, such as fat, muscles, and connective tissues. With over 70 different subtypes of soft tissue sarcoma (STS), each with unique characteristics and survival rates, finding universally effective treatments has been challenging.
A research team led by a scientist at the Research Institute of the McGill University Health Centre (RI-MUHC) has made a significant breakthrough by harnessing artificial intelligence (AI) and deep learning. They have developed a model that can predict overall survival and risk of metastases —the spread of cancer to other parts of the body— for STS patients better than any previous model. The new model, recently published the journal NPJ Precision Oncology, can potentially lead to more personalized treatment plans and better outcomes for patients.
Anthony Bozzo, MD. M.Sc. FRCSC, a Junior Scientist in the Cancer Research Program of the RI-MUHC and an orthopedic oncologist at the Montreal General Hospital site of the McGill University Health Centre, spearheaded the development of this new multimodal neural network (MMNN) model. Working with colleagues at the Memorial Sloan Kettering Cancer Center, he created the innovative model that combines a patient’s clinical data—such as age, tumour size, tumour grade, and tumour subtype—with data from 3D magnetic resonance imaging (MRI) scans to predict overall survival and the risk of cancer spreading to other parts of the body.
Traditional prediction models have relied on just a few clinical variables, making them too simplistic to provide meaningful guidance for individual treatment. In contrast, the new MMNN model incorporates MRI scans, which provide rich, personalized data about each patient’s tumour. By including data from these images, the researchers have advanced from using single-source data (unimodal) to combining multiple types of data (multimodal), significantly enhancing the model’s accuracy.
The team’s research showed that this new approach is much more accurate than any existing models. This could pave the way for more personalized care, allowing doctors to tailor treatments specifically to each patient’s unique disease profile.
“This research began during my fellowship at Sloan Kettering,” says Dr. Bozzo. “I’m excited to continue the work here at the McGill University Health Centre, where Dr. Robert Turcotte has established a leading sarcoma unit. Together with my MUHC colleagues, Dr. Ahmed Aoude, Dr. James Tsui, and Dr. Natalia Gorelik, we are advancing the use of AI to better support our sarcoma patients.”
The next step for the researchers is to externally validate the MMNN model using larger data sets. They plan to use a federated learning process, in which multiple institutions collaborate to train the model while keeping their data decentralized to protect patient privacy. The team will begin by expanding the model with data from the McGill University Health Centre and will actively recruit other international health centres to join the effort.
“By improving the accuracy of predictions, this new model could help healthcare providers better manage sarcoma, ultimately leading to improved outcomes for patients,” adds Dr. Bozzo. “Our work is a promising step towards more individualized and effective treatments for sarcoma patients, and it showcases the potential of AI in transforming cancer care.”
The researchers gratefully acknowledge funding support from the Montreal General Hospital Foundation and Cedars Cancer Foundation. Dr. Bozzo thanks his research mentors Dr. Michelle Ghert and Dr. John Healey for their guidance.
About the publication
Bozzo, A., Hollingsworth, A., Chatterjee, S. et al. A multimodal neural network with gradient blending improves predictions of survival and metastasis in sarcoma. npj Precis. Onc. 8, 188 (2024). https://doi.org/10.1038/s41698-024-00695-7