MammographyTransfer LearningLuminal Subtypes

Classification of Luminal Subtypes in Full Mammogram Images Using Transfer Learning

arXiv / 2023 / preprint

Panambur AB, Madhu P, Maier A

preprint / 2023

Medical AI visual summary

MammographyTransfer LearningLuminal Subtypes

Visual placeholder for this publication. Add a dedicated figure in the Markdown frontmatter when a public method figure is available.

Scientific summary

This preprint evaluates whether full-field mammograms contain signal for luminal subtype classification using image-level supervision.

Abstract

This preprint studies luminal versus non-luminal subtype classification from full mammogram images using image-level labels.

Why it matters

Image-level subtype prediction is relevant for weakly supervised breast imaging research, especially when pixel-level tumor annotations are unavailable.

Contribution

It reports initial evidence for full-image transfer learning on luminal subtype classification using the public CMMD dataset.

Method overview

A ResNet-18 model is transferred from breast abnormality classification and fine-tuned for luminal versus non-luminal subtype prediction.

Key findings

The arXiv abstract reports mean AUC 0.6688 and mean F1 0.6693 on the test dataset.
The approach uses only image-level labels.
The reported improvement over baseline was statistically significant with p < 0.0001.

Citation

Panambur AB, Madhu P, Maier A. (2023). Classification of Luminal Subtypes in Full Mammogram Images Using Transfer Learning. arXiv.

TODO: Add BibTeX.