We have a call to action in healthcare – we need to drive value.  Artificial intelligence (AI), if deployed correctly, can help accomplish this lofty mission.  In this discussion we will review the following lessons learned in deploying radiology AI at scale:  4 unexpected benefits of implementing AI emergent finding triage; the importance of investing in AI radiologist education; how “most” AI needs to be incorporated into the radiologist workflow; why a platform is required to deploy AI at scale and what a modern platform looks like; how to use AI to add value to your data; and, as Dr. Curt Langlotz famously said, why rads (practices) who use AI will replace those who don’t (a depiction of what the role of the radiologist might look like in a tech enabled future).

Cancer patients are imaged with multiple imaging modalities as part of routine cancer care. However, the rich information available from the images are not fully exploited to better manage patient care through earlier intervention as well as more precise targeted treatments. In this talk, I will present some of the new AI methodologies we have been developing to track tumor response as well as from routinely acquired imaging applied to image-guided radiation treatments using CT/cone-beam CT as well as MRI-guided precision treatments. I will also present some demonstration studies of how AI based automated segmentation and tumor as well as healthy tissue change assessment can be used to early detect treatment toxicities to enable clinicians to better manage cancer care. Finally, I will show how these developed methods have been put to routine clinical care for automating radiotherapy treatment planning at MSK.

Central nervous system (CNS) tumors come with vastly heterogeneous histologic, molecular, and radiographic landscapes, rendering their precise characterization challenging. The rapidly growing fields of biophysical modeling and radiomics have shown promise in better characterizing the molecular, spatial, and temporal heterogeneity of tumors. Integrative analysis of CNS tumors, including clinically acquired multi-parametric magnetic resonance imaging (mpMRI), assists in identifying macroscopic quantifiable tumor patterns of invasion and proliferation, potentially leading to improved (a) detection/segmentation of tumor subregions and (b) computer-aided diagnostic/prognostic/predictive modeling. This talk will touch upon example studies on this space, as well as an overview of the largest to-date real-world federated learning study to detect brain tumor boundaries.

Developing and deploying computational tools for neuro-oncology applications includes a sequence of complex steps to identify appropriate images, assess image quality, annotate, process and other prepare and manipulate data for analysis. We have implemented services and tools on the open source XNAT informatics platform to automate much of this workflow to improve both its efficiency and effectiveness. Dr. Marcus will discuss this automated workflow and its implementation in a number of data sets and applications at Washington University.

Medical artificial intelligence is rapidly moving into clinics, particularly in imaging-based specialties such as radiology. This transition is producing many new challenges, as the regulatory environment has struggled to keep up and AI training for healthcare workers is virtually non-existent. Dr Oakden-Rayner will provide a clinical safety perspective on medical AI, discuss a range of identified risks and potential harms, and discuss possible solutions to mitigate these risks as this exciting field continues to develop.

We propose and demonstrate a novel approach to training image classification models based on large collections of images with limited labels. We take advantage of availability of radiology reports to construct joint multimodal embedding that serves as a basis for classification. We demonstrate the advantages of this approach in application to assessment of pulmonary edema severity in congestive heart failure that motivated the development of the method.

Knee injuries and pain are prevalent across all ages, with varying causes from “anterior knee pain” in runners to osteoarthritis-related pain. Osteoarthritis pain is a particular problem because structural outcomes assessed on medical images often disagree with symptoms. Most studies trying to understand knee health and pain use simple biomarkers such as mean cartilage thickness. My talk will present an automated pipeline for quantifying the whole knee using statistical shape modeling. I will present a conventional statistical shape model as well as a novel approach that uses generative neural implicit representations. Both modeling approaches allow unsupervised identification of salient anatomic features. I will demonstrate how these features can be used to predict existing radiographic outcomes, patient demographics, and knee pain.

Distributed deep learning is an emerging research paradigm for enabling collaboratively training deep learning models without sharing patient data. 

In this talk, we will first investigate the use distributed deep learning to build medical imaging classification models in a real-world collaborative setting. 

We then present several strategies to tackle the data heterogeneity challenge and the lack of quality labeled data challenge in distributed deep learning.