Fréchet radiomic distance (FRD): A versatile metric for comparing medical imaging datasets.
Authors
Affiliations (19)
Affiliations (19)
- organization=Department of Electrical and Computer Engineering, Duke University, city=NC, country=USA. Electronic address: [email protected].
- organization=Departament de Matemàtiques i Informàtica, Universitat de Barcelona, city=Barcelona, country=Spain; organization=Institute of Machine Learning in Biomedical Imaging, Helmholtz Center Munich, city=Munich, country=Germany; organization=School of Computation, Information and Technology, Technical University of Munich, city=Munich, country=Germany. Electronic address: [email protected].
- organization=Departament de Matemàtiques i Informàtica, Universitat de Barcelona, city=Barcelona, country=Spain.
- organization=Department of Electrical and Computer Engineering, Duke University, city=NC, country=USA. Electronic address: [email protected].
- organization=Department of Electrical and Computer Engineering, Duke University, city=NC, country=USA. Electronic address: [email protected].
- organization=Department of Electrical and Computer Engineering, Duke University, city=NC, country=USA. Electronic address: [email protected].
- organization=Department of Electrical and Computer Engineering, Duke University, city=NC, country=USA. Electronic address: [email protected].
- organization=Department of Biomedical Engineering, Yale University, city=CT, country=USA. Electronic address: [email protected].
- organization=Departament de Matemàtiques i Informàtica, Universitat de Barcelona, city=Barcelona, country=Spain. Electronic address: [email protected].
- organization=Departament de Matemàtiques i Informàtica, Universitat de Barcelona, city=Barcelona, country=Spain. Electronic address: [email protected].
- organization=Institute of Machine Learning in Biomedical Imaging, Helmholtz Center Munich, city=Munich, country=Germany; organization=School of Computation, Information and Technology, Technical University of Munich, city=Munich, country=Germany. Electronic address: [email protected].
- organization=Department of Radiology, Weill Cornell Medical College, city=NY, country=USA. Electronic address: [email protected].
- organization=Department of Radiology, Duke University, city=NC, country=USA. Electronic address: [email protected].
- organization=Department of Radiology & Biomedical Imaging, Yale University, city=CT, country=USA. Electronic address: [email protected].
- organization=Department of Biomedical Engineering, Yale University, city=CT, country=USA; organization=Department of Radiology & Biomedical Imaging, Yale University, city=CT, country=USA; organization=Department of Electrical Engineering, Yale University, city=CT, country=USA. Electronic address: [email protected].
- organization=Institute of Machine Learning in Biomedical Imaging, Helmholtz Center Munich, city=Munich, country=Germany; organization=School of Computation, Information and Technology, Technical University of Munich, city=Munich, country=Germany; organization=School of Biomedical Engineering & Imaging Sciences, King's College London, city=London, country=UK. Electronic address: [email protected].
- organization=Departament de Matemàtiques i Informàtica, Universitat de Barcelona, city=Barcelona, country=Spain; organization=Computer Vision Center, Universitat Autònoma de Barcelona, city=Bellaterra, country=Spain. Electronic address: [email protected].
- organization=Departament de Matemàtiques i Informàtica, Universitat de Barcelona, city=Barcelona, country=Spain; organization=Institució Catalana de Recerca i Estudis Avançats (ICREA), city=Barcelona, country=Spain. Electronic address: [email protected].
- organization=Department of Electrical and Computer Engineering, Duke University, city=NC, country=USA; organization=Department of Radiology, Duke University, city=NC, country=USA; organization=Department of Biostatistics & Bioinformatics, Duke University, city=NC, country=USA; organization=Department of Computer Science, Duke University, city=NC, country=USA. Electronic address: [email protected].
Abstract
Determining whether two sets of images belong to the same or different distributions or domains is a crucial task in modern medical image analysis and deep learning; for example, to evaluate the output quality of image generative models. Currently, metrics used for this task either rely on the (potentially biased) choice of some downstream task, such as segmentation, or adopt task-independent perceptual metrics (e.g., Fréchet Inception Distance/FID) from natural imaging, which we show insufficiently capture anatomical features. To this end, we introduce a new perceptual metric tailored for medical images, FRD (Fréchet Radiomic Distance), which utilizes standardized, clinically meaningful, and interpretable image features. We show that FRD is superior to other image distribution metrics for a range of medical imaging applications, including out-of-domain (OOD) detection, the evaluation of image-to-image translation (by correlating more with downstream task performance as well as anatomical consistency and realism), and the evaluation of unconditional image generation. Moreover, FRD offers additional benefits such as stability and computational efficiency at low sample sizes, sensitivity to image corruptions and adversarial attacks, feature interpretability, and correlation with radiologist-perceived image quality. Additionally, we address key gaps in the literature by presenting an extensive framework for the multifaceted evaluation of image similarity metrics in medical imaging-including the first large-scale comparative study of generative models for medical image translation-and release an accessible codebase to facilitate future research. Our results are supported by thorough experiments spanning a variety of datasets, modalities, and downstream tasks, highlighting the broad potential of FRD for medical image analysis.