In the Era of Deep Learning, Why Reconstruct an Image at All?

Abstract

In the pursuit of precision medicine, the value of integrating a wide variety of sources of data and using quantitative approaches has been emphasized. Imaging has had a rapidly growing and important role in clinical medical practice for diagnosis, measure of treatment response, assessing toxicity, and, more recently, prediction of outcomes to guide treatment decisions. Although the past few decades have seen an evolution in quantitative imaging in radiology and radiation oncology, the radiology report today remains, for the most part, a human interpretation of visualizations of an imaging examination for the purpose of communicating the observations from a radiologist to the clinical team and thus remains largely qualitative. Some have described the radiology report as an extreme form of data compression, because this process converts megabytes of data into a few bytes [ 1 Harvey H. Separating the art of medicine from artificial intelligence. https://towardsdatascience.com/separating-the-art-of-medicine-from-artificial-intelligence-6582f86ea244Date accessed: October 7, 2020 Google Scholar ].