A dermal backflow atlas of lymphatic dysfunction in breast cancer-related lymphedema with machine learning-based stage prediction.

Authors

Lalida Sutejitsiri, Hiroo Suami, Poul M F Nielsen, Anthony R J Phillips, Hayley M Reynolds

Published in

Breast cancer research and treatment. Volume 219. Issue 3. Oct 07, 2026. Epub Oct 07, 2026.

Abstract

Breast cancer-related lymphedema (BCRL) is a chronic condition caused by impaired lymphatic drainage following cancer treatment, manifesting as fluid accumulation, adipogenesis, and lymph reflux to the skin (dermal backflow). The affected regions are diverse and therefore, improved understanding of the spatial distribution of lymphatic dysfunction may facilitate earlier detection and disease monitoring. This study aimed to develop a spatial atlas of dermal backflow in BCRL and apply machine learning (ML) models to predict disease severity.
Data from 282 female BCRL patients (289 affected limbs) were obtained from the Australian Lymphedema Education, Research and Treatment (ALERT) Centre Databank, including indocyanine green (ICG) lymphography studies and clinical information. A spatial atlas of dermal backflow prevalence in the upper extremities was constructed by co-registering and segmenting clinician-annotated ICG diagrams. Five ML models were developed to classify MD Anderson Cancer Center (MDACC) lymphedema stage, incorporating dermal backflow percentage and clinical features.
The atlas revealed consistent spatial patterns of dermal backflow, with the highest prevalence in the wrist, forearm, and the olecranon region. Subgroup-specific heat maps stratified by MDACC stage, L-Dex ratio, limb volume difference, drainage regions, and hand dermal backflow further demonstrated distinct spatial variations with disease severity. The inclusion of dermal backflow percentage improved ML prediction of MDACC stage, with the best performance achieved by multinomial logistic regression and random forest models.
Dermal backflow prevalence revealed non-uniform, region-specific patterns in BCRL patients, which can be visualized using spatial atlases. Integrating quantitative imaging features into ML models improves BCRL staging and supports objective, data-driven disease assessment.

PMID:
42842077
Bibliographic data and abstract were imported from PubMed on 08 Oct 2026.

Read full publication at:
Please sign in to see all details.

Stats

1-terrible, 9-excellent. How would you rate this publication? Sign in in to submit your rating.

Post a comment

You need to be signed in to post comments. You can sign in here.

Comments

There are no comments yet.