Research Project Overviews

A Pan-Canadian Data Collection and Analysis Platform for Patient Radiation Protection and Safety

We are currently developing a decision support system (DSS) that provides real-time risk assessment of radiation exposure from medical imaging relative to a patient’s medical history. Our system will allow patients, radiologists, and medical researchers to evaluate if the benefits of performing an imaging study outweighs the potential cancer risks from low dose radiation, which has increased in terms of frequency and dose over the last decade. Health professionals can then determine if the patient should proceed with conducting the imaging or resort to a lower-dose or non-ionizing modality.

Our decision support system architecture has two main requirements. First, the currently most successful deep learning models used for diagnostic prediction, called transformers, lack a mechanism to analyze the temporal characteristics of electronic health records. To address this gap, we developed a temporally-embedded transformer to predict patients' primary diagnoses by analyzing their medical histories, including the elapsed time between visits. We evaluated our proposed model on a real-world medical dataset we curated. Our proposed model successfully predicted patients’ primary diagnosis in their final visit with improved predictive performance compared to an existing model in the literature. Second, the DSS should allow the secure communication and sharing of data between various stakeholders (e.g., patients, health professionals, medical researchers) while preserving the privacy of data subjects. In this project, we designed and evaluated a blockchain-based architecture to digitize trust in collaborative health research environments. Our proposed architecture provides human-in-the-loop feedback between patients, researchers, and artificial intelligence systems to improve predictive analytics and provide personalized care. Our architecture supports three functionalities: 1) Provenance management of research data, 2) Privacy management of data contributors, and 3) Distributed and verifiable trust among participants. An interactive application that monitors the amount of accumulated medical radiation patients received (called DoseDash) has also been designed.

Future work will focus on evaluating the predictive diagnostic power of our proposed model compared to the existing cancer risk models in the literature.

Principal Investigators

Dr. David A. Koff, MD FRCPC, Professor, Department of Radiology, McMaster University

Dr. Thomas E. Doyle, PhD PEng, Associate Professor, Department of Electrical and Computer Engineering, McMaster University, School of Biomedical Engineering, McMaster University, Vector Institute

Dr. Reza Samavi, PhD PEng, Department of Electrical, Computer, and Biomedical Engineering, Ryerson University, Vector Institute

Researchers

Omar Boursalie, PhD

Andrew Sutton, BASc, MASc (Computer Science)

Michelle Ferderbar, BMRSc, MSc (eHealth)

Project Manager

Jane A. Castelli

Industry Partner

RealTime Medical

Peer-reviewed Publications

• Boursalie O., Samavi R, and Doyle TE. (2021) Decoder Transformer for Temporally-Embedded Health Outcome Predictions. The 20th IEEE International Conference on Machine Learning Applications (ICMLA-21). (in-press)
• Boursalie O., Samavi R., and Doyle TE. (2021) Evaluation Metrics for Deep Learning Imputation Models. The 5th International Workshop on Health Intelligence (W3PHIAI-21) co-located with the 35th AAAI Conference on AI. (in-press)
• Boursalie O., Samavi R., Doyle TE., and Koff DA. (2020) Using Medical Imaging EffectiveDose in Deep Learning Models: Estimation and Evaluation. IEEE Transactions on Radiation and Plasma Medical Sciences (TRPMS). https://dx.doi.org/10.1109/TRPMS.2020.3029038
• Boursalie O., Samavi R., Doyle TE., and Koff DA. (2020) Deep Learning Model for Cancer Risk from Low Dose Medical Imaging Radiation. European Congress of Radiology (ECR). Poster Presentation. https://dx.doi.org/10.26044/esi2020/ESI-10315
• Ferderbar ML., Doyle TE, Samavi R, and Koff DA. (2019). An environmental scan of the national and provincial diagnostic reference levels in Canada for common adult computed tomography scans. Canadian Association of Radiologists' Journal. https://doi.org/10.1016/j.carj.2018.07.005
• Sutton, A, and Samavi R. (2019). Integrity Proofs for RDF Graphs. Open Journal of Semantic Web (OJSW). http://nbn-resolving.de/urn:nbn:de:101:1-2018102818300947746192
• Sutton A, Samavi R, Doyle TE, and Koff DA. (2018). Digitized Trust in Human-in-the-loop Health Research. In 16th Annual Conference on Privacy, Security and Trust (PST-18). https://doi.org/10.1109/PST.2018.8514168
• Sutton A, and Samavi R. (2018). Timestamp-based integrity proofs for linked data. In Proceedings of the International Workshop on Semantic Big Data (SBD-18). https://doi.org/10.1145/3208352.3208353
• Sutton A, and Samavi R. (2018). Tamper-Proof Privacy Auditing for Artificial Intelligence Systems. In Proceedings of the Twenty-Seventh International Joint Conference on Artificial Intelligence (IJCAI-18). https://www.ijcai.org/Proceedings/2018/0756.pdf
• Sutton A and Samavi R. (2017). Blockchain Enabled Privacy Audit Logs. International Semantic Web Conference (ISWC-17). Best Research Paper and Best Student Research Paper award nominee. http://iswc2017.semanticweb.org/paper-141/
• Boursalie O., Samavi R., Doyle T. (2017). Machine Learning and Mobile Health Monitoring Platforms: A Case Study on Research and Implementation Challenges. Journal of Health Informatics Research. https://doi.org/10.1007/s41666-018-0021-1