Abstract

AI is leveraged commonly nowadays, and many industries have been benefited a lot by it, including healthcare. Memorial Sloan Kettering Cancer Center is one of the most prestigious cancer centers in the USA and its services generate a huge amount of medical image, which could be more than a small country has. As a member of the Radiology Informatics team here, we have many opportunities to work on a huge set of medical images. So, I'd like to share some stories about the challenges we faced when we plugged AI solutions into our clinical environment. I will go through some real examples from the research on deep learning models for medical image classification, auto-segmentation and then move to how we integrate the model into our clinical workflow. The talk will focus on the challenge of integration a bit more rather than how we design and train the model because while deep learning model especially convolutional neural network and its variations have been widely introduced in many research correlated to medical image, people seldom talk about the missing components when it's brought to a real-world clinical environment. As part of the conclusion, I will also provide some personal thinking about the future direction of AI in medical imaging.

All faculty and students are welcome to join.