Cargando…

Self-Supervision for Medical Image Classification: State-of-the-Art Performance with ~100 Labeled Training Samples per Class

Is self-supervised deep learning (DL) for medical image analysis already a serious alternative to the de facto standard of end-to-end trained supervised DL? We tackle this question for medical image classification, with a particular focus on one of the currently most limiting factor of the field: th...

Descripción completa

Detalles Bibliográficos
Autores principales: Nielsen, Maximilian, Wenderoth, Laura, Sentker, Thilo, Werner, René
Formato: Online Artículo Texto
Lenguaje:English
Publicado: MDPI 2023
Materias:
Acceso en línea:https://www.ncbi.nlm.nih.gov/pmc/articles/PMC10451977/
https://www.ncbi.nlm.nih.gov/pubmed/37627780
http://dx.doi.org/10.3390/bioengineering10080895
_version_ 1785095552896073728
author Nielsen, Maximilian
Wenderoth, Laura
Sentker, Thilo
Werner, René
author_facet Nielsen, Maximilian
Wenderoth, Laura
Sentker, Thilo
Werner, René
author_sort Nielsen, Maximilian
collection PubMed
description Is self-supervised deep learning (DL) for medical image analysis already a serious alternative to the de facto standard of end-to-end trained supervised DL? We tackle this question for medical image classification, with a particular focus on one of the currently most limiting factor of the field: the (non-)availability of labeled data. Based on three common medical imaging modalities (bone marrow microscopy, gastrointestinal endoscopy, dermoscopy) and publicly available data sets, we analyze the performance of self-supervised DL within the self-distillation with no labels (DINO) framework. After learning an image representation without use of image labels, conventional machine learning classifiers are applied. The classifiers are fit using a systematically varied number of labeled data (1–1000 samples per class). Exploiting the learned image representation, we achieve state-of-the-art classification performance for all three imaging modalities and data sets with only a fraction of between 1% and 10% of the available labeled data and about 100 labeled samples per class.
format Online
Article
Text
id pubmed-10451977
institution National Center for Biotechnology Information
language English
publishDate 2023
publisher MDPI
record_format MEDLINE/PubMed
spelling pubmed-104519772023-08-26 Self-Supervision for Medical Image Classification: State-of-the-Art Performance with ~100 Labeled Training Samples per Class Nielsen, Maximilian Wenderoth, Laura Sentker, Thilo Werner, René Bioengineering (Basel) Article Is self-supervised deep learning (DL) for medical image analysis already a serious alternative to the de facto standard of end-to-end trained supervised DL? We tackle this question for medical image classification, with a particular focus on one of the currently most limiting factor of the field: the (non-)availability of labeled data. Based on three common medical imaging modalities (bone marrow microscopy, gastrointestinal endoscopy, dermoscopy) and publicly available data sets, we analyze the performance of self-supervised DL within the self-distillation with no labels (DINO) framework. After learning an image representation without use of image labels, conventional machine learning classifiers are applied. The classifiers are fit using a systematically varied number of labeled data (1–1000 samples per class). Exploiting the learned image representation, we achieve state-of-the-art classification performance for all three imaging modalities and data sets with only a fraction of between 1% and 10% of the available labeled data and about 100 labeled samples per class. MDPI 2023-07-28 /pmc/articles/PMC10451977/ /pubmed/37627780 http://dx.doi.org/10.3390/bioengineering10080895 Text en © 2023 by the authors. https://creativecommons.org/licenses/by/4.0/Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https://creativecommons.org/licenses/by/4.0/).
spellingShingle Article
Nielsen, Maximilian
Wenderoth, Laura
Sentker, Thilo
Werner, René
Self-Supervision for Medical Image Classification: State-of-the-Art Performance with ~100 Labeled Training Samples per Class
title Self-Supervision for Medical Image Classification: State-of-the-Art Performance with ~100 Labeled Training Samples per Class
title_full Self-Supervision for Medical Image Classification: State-of-the-Art Performance with ~100 Labeled Training Samples per Class
title_fullStr Self-Supervision for Medical Image Classification: State-of-the-Art Performance with ~100 Labeled Training Samples per Class
title_full_unstemmed Self-Supervision for Medical Image Classification: State-of-the-Art Performance with ~100 Labeled Training Samples per Class
title_short Self-Supervision for Medical Image Classification: State-of-the-Art Performance with ~100 Labeled Training Samples per Class
title_sort self-supervision for medical image classification: state-of-the-art performance with ~100 labeled training samples per class
topic Article
url https://www.ncbi.nlm.nih.gov/pmc/articles/PMC10451977/
https://www.ncbi.nlm.nih.gov/pubmed/37627780
http://dx.doi.org/10.3390/bioengineering10080895
work_keys_str_mv AT nielsenmaximilian selfsupervisionformedicalimageclassificationstateoftheartperformancewith100labeledtrainingsamplesperclass
AT wenderothlaura selfsupervisionformedicalimageclassificationstateoftheartperformancewith100labeledtrainingsamplesperclass
AT sentkerthilo selfsupervisionformedicalimageclassificationstateoftheartperformancewith100labeledtrainingsamplesperclass
AT wernerrene selfsupervisionformedicalimageclassificationstateoftheartperformancewith100labeledtrainingsamplesperclass