Title: Assessment of Deep Neural Networks from a Medical Imaging Perspective

Speaker: Dr. Mark Anastasio, UIUC - BME Department Chair

Abstract: It is widely accepted that the assessment and refinement of biomedical imaging technologies should be performed by objective, i.e., task-based, measures of image quality (IQ). However, the objective evaluation of deep learning-based image formation technologies remains largely lacking, despite the breakneck speed at which they are being developed. In this work, we objectively assess two types of deep neural networks that are being actively developed for medical imaging applications. We first report studies in which the performance of deep a learning-based image restoration method is objectively assessed. The performance of the ideal observer (IO) and common linear numerical observers are quantified, and detection efficiencies are computed to assess the potential impact of deep learning on signal detection performance in this application.  Next, we present procedures for assessing deep generative models (DGMs). While DGMs have been shown to produce medical images that look realistic, such images can contain difficult-to-detect errors that can compromise their use in a medical imaging application.  It also remains unclear which image statistics a DGM can reliably reproduce. We also review recent studies that reveal new and important insights regarding the relative capacity of generative adversarial models (GANs) and denoising diffusion probabilistic models (DDPMs) to learn spatial context that is relevant to medical imaging.