Synthetic data augmentation for CT-based emphysema subtype classification: A comparative evaluation of generative and classical approaches

Nicholas Dietrich, David McShannon · 2026

Read Synthetic data augmentation for CT-based emphysema subtype classification: A comparative evaluation of generative and classical approaches by Nicholas Dietrich, David McShannon (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/plos-pone-0355850.txt.

Description

Nicholas Dietrich and David McShannon test a popular fix for scarce medical images in this 2026 study: making synthetic training data. Their task was classifying emphysema subtypes in lung CT scans, where only a small set of real image patches was available. They compared three modern generative models, which learn to invent new scans, against simple geometric tricks like rotating and flipping real images.

None of the methods, alone or combined, improved the classifier's accuracy in a statistically reliable way over training on real images only. The best single result came from one generative model at an equal mix of real and synthetic data, but the gain was not significant. How realistic the fake images looked did not predict how much they helped. The authors say generative augmentation in tiny, texture-driven datasets needs better methods before it can be trusted.

Catalog record

CREATORNicholas Dietrich, David McShannon
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE38214
LICENSECC_BY
RIGHTSCC BY
SOURCEPLOS

Source: https://doi.org/10.1371/journal.pone.0355850

Browse the full library · About the Tower of Babel