Hidden Stratification Causes Clinically Meaningful Failures in Machine Learning for Medical Imaging

Proceedings of the ACM Conference on Health, Inference, and Learning
Luke Oakden-RaynerChristopher Ré

Abstract

Machine learning models for medical image analysis often suffer from poor performance on important subsets of a population that are not identified during training or testing. For example, overall performance of a cancer detection model may be high, but the model may still consistently miss a rare but aggressive cancer subtype. We refer to this problem as hidden stratification, and observe that it results from incompletely describing the meaningful variation in a dataset. While hidden stratification can substantially reduce the clinical efficacy of machine learning models, its effects remain difficult to measure. In this work, we assess the utility of several possible techniques for measuring hidden stratification effects, and characterize these effects both via synthetic experiments on the CIFAR-100 benchmark dataset and on multiple real-world medical imaging datasets. Using these measurement techniques, we find evidence that hidden stratification can occur in unidentified imaging subsets with low prevalence, low label quality, subtle distinguishing features, or spurious correlates, and that it can result in relative performance differences of over 20% on clinically important subsets. Finally, we discuss the clinical implicatio...Continue Reading

Citations

Sep 11, 2020·Nature Medicine·Samantha Cruz RiveraUNKNOWN SPIRIT-AI and CONSORT-AI Consensus Group
Sep 11, 2020·Nature Medicine·Xiaoxuan LiuUNKNOWN SPIRIT-AI and CONSORT-AI Working Group
Sep 13, 2020·BMC Medical Informatics and Decision Making·Federico CabitzaLuca Maria Sconfienza
Sep 11, 2020·BMJ : British Medical Journal·Xiaoxuan LiuUNKNOWN SPIRIT-AI and CONSORT-AI Working Group
Jun 17, 2020·Health Information Science and Systems·Nicholas ChedidRichard Andrew Taylor
Oct 22, 2020·NPJ Digital Medicine·Sujay KakarmathJoseph Kvedar
Nov 28, 2020·American Journal of Medical Genetics. Part a·Melissa D McCraddenLauren Chad
Jan 16, 2021·Intensive Care Medicine·Errol ColakMarzyeh Ghassemi
Feb 17, 2021·BMC Medical Imaging·Markus J AnkenbrandLaura M Schreiber
Mar 18, 2021·Scientific Reports·Abu Mohammed RaisuddinAleksei Tiulpin
May 1, 2021·Cancer Cell·Benjamin H KannHugo J W L Aerts
Apr 30, 2021·Trauma Surgery & Acute Care Open·Jeff ChoiChien-Hung Liao
Jun 26, 2021·Journal of Medical Imaging and Radiation Oncology·Catherine M JonesBen Hachey
Aug 17, 2021·Annual Review of Biomedical Data Science·Irene Y ChenMarzyeh Ghassemi
Sep 30, 2021·GigaScience·Jérôme DockèsJean-Baptiste Poline
Nov 21, 2021·Scientific Reports·Ekin YagisStefano Diciotti

❮ Previous
Next ❯

Related Concepts

Trending Feeds

COVID-19

Coronaviruses encompass a large family of viruses that cause the common cold as well as more serious diseases, such as the ongoing outbreak of coronavirus disease 2019 (COVID-19; formally known as 2019-nCoV). Coronaviruses can spread from animals to humans; symptoms include fever, cough, shortness of breath, and breathing difficulties; in more severe cases, infection can lead to death. This feed covers recent research on COVID-19.

Blastomycosis

Blastomycosis fungal infections spread through inhaling Blastomyces dermatitidis spores. Discover the latest research on blastomycosis fungal infections here.

Nuclear Pore Complex in ALS/FTD

Alterations in nucleocytoplasmic transport, controlled by the nuclear pore complex, may be involved in the pathomechanism underlying multiple neurodegenerative diseases including Amyotrophic Lateral Sclerosis and Frontotemporal Dementia. Here is the latest research on the nuclear pore complex in ALS and FTD.

Applications of Molecular Barcoding

The concept of molecular barcoding is that each original DNA or RNA molecule is attached to a unique sequence barcode. Sequence reads having different barcodes represent different original molecules, while sequence reads having the same barcode are results of PCR duplication from one original molecule. Discover the latest research on molecular barcoding here.

Chronic Fatigue Syndrome

Chronic fatigue syndrome is a disease characterized by unexplained disabling fatigue; the pathology of which is incompletely understood. Discover the latest research on chronic fatigue syndrome here.

Evolution of Pluripotency

Pluripotency refers to the ability of a cell to develop into three primary germ cell layers of the embryo. This feed focuses on the mechanisms that underlie the evolution of pluripotency. Here is the latest research.

Position Effect Variegation

Position Effect Variagation occurs when a gene is inactivated due to its positioning near heterochromatic regions within a chromosome. Discover the latest research on Position Effect Variagation here.

STING Receptor Agonists

Stimulator of IFN genes (STING) are a group of transmembrane proteins that are involved in the induction of type I interferon that is important in the innate immune response. The stimulation of STING has been an active area of research in the treatment of cancer and infectious diseases. Here is the latest research on STING receptor agonists.

Microbicide

Microbicides are products that can be applied to vaginal or rectal mucosal surfaces with the goal of preventing, or at least significantly reducing, the transmission of sexually transmitted infections. Here is the latest research on microbicides.