Pioneering medical image processing with machine learning and Dr. April Khademi

Machine learning has been quite the stir in the latest developments with technology. Dr. April Khademi’s transformative research with machine learning has had a profound impact on medical image processing and has helped shape the way clinicians practice medicine. An associate professor in electrical, computer and biomedical engineering at Toronto Metropolitan University (TMU), her novel research focuses on the integration of AI with medical imaging, image processing and biomedical signals. Dr. Khademi is also a Canadian Research Chair in AI for medical imaging with applications in radiology, pathology and neurology. Her innovative work has led to advances against dementia, vascular disease, breast and brain cancer through image processing of both grayscale images and RGB images from cells and tissue microstructures. In addition, her team actively implements AI into the workflow of clinicians to complement their work in areas such as biomarker systems, diagnostics, and early disease detection.

The start of Dr. Khademi’s journey in biomedical engineering started in her late years during her undergraduate degree. Falling in love with digital signal processing, Dr. Khademi had a fascination with the application of mathematics being used to analyze 2D images. Combined with her love for image processing, Dr. Khademi started writing code for software that would segment medical images. During her undergraduate degree, she worked as an office manager, and when she went to present this to her supervisors, they were mortified by this!

Regardless, Dr. Khademi believed in the potential for medical image processing and sought a Ph.D in this specialization. Later on, Dr. Khademi went on to design novel segmentation tools in neurology through the use of only 25 images as her dataset (side note, this was approximately only about 10 years ago)! Current datasets are in the thousands of images and are sourced from multiple imaging centers. Dr. Khademi then went on to work in the industry such as GE healthcare, working with clinicians and R&D. However, Dr. Khademi’s passion for teaching remained with her. She soon taught at Guelph as a professor in medical imaging analysis before landing her current position at TMU.

For her current research, their lab focuses on diagnostics and assignment of therapies for radiologists and pathologists. Current diagnostics from images can be subjective, and the lab’s goal is to develop AI tools that would reduce the uncertainty and standardize the analysis of medical imaging. Khademi’s lab not only places a huge emphasis on the quality of the tools, but also its implementation in the healthcare setting. With an arsenal of AI tools in breast cancer and neurology, one of Khademi’s latest tools has been tested worldwide with nearly 100 pathologists. The results indicated that this made the pathologists more accurate, efficient, and agree more with their colleagues.

When the CUBEC team asked her about the implications of the ethics of AI with medical data, Dr. Khademi had the following to say: “It’s dependent on the setup of the study. Both the doctor and the patient technically own the data, and it’s a current issue debated in the legal community. There are advantages with machine learning. The patient's data is hard to retrieve from weights of the machine learning model. There’s a responsibility to protect the data, get patient consent, and encrypt data, but it’s still a debated topic.”

Dr. Khademi noted that within the last decade, the large progress on access to open repositories and opportunities has improved the accuracy of AI. However, she also highlighted that there were several challenges integrating machine learning in the clinical setting. Whereas in the United States, healthcare is privatized, and hospitals are more keen on trying new technologies. This is not the same in Canada, and the path to do so is not clear. Some of the concerns that Dr. Khademi laid out were the integration of models into the workflow in hospital database systems, decision making, quality of care and patient management. In other words, the unanswered question is: how will it change the way doctors practice medicine? From Dr. Khademi: “There needs to be a way to define how AI will be used by doctors and clinicians, and it’s ultimately for them to give the greenlight for these tools to be used and how it will affect patient care.”

Dr. Khademi had the following advice for students interested in the AI field: “Getting into AI in the medical field is something you need to love and be passionate about. Strong mathematical skills, computer programming, medical imaging, deep learning, and other skills are needed for this field. There is also the need to understand the clinical aspect, because otherwise your innovation dies out on a shelf. Understanding clinical impact is an important toolset that you require to be in this field. “

‍