Self-Supervision for Medical Image Classification
Standard
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.
In: BIOENGINEERING-BASEL, Vol. 10, No. 8, 895, 28.07.2023.Research output: SCORING: Contribution to journal › SCORING: Journal article › Research › peer-review
Harvard
APA
Vancouver
Bibtex
}
RIS
TY - JOUR
T1 - Self-Supervision for Medical Image Classification
T2 - State-of-the-Art Performance with ~100 Labeled Training Samples per Class
AU - Nielsen, Maximilian
AU - Wenderoth, Laura
AU - Sentker, Thilo
AU - Werner, René
PY - 2023/7/28
Y1 - 2023/7/28
N2 - 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.
AB - 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.
U2 - 10.3390/bioengineering10080895
DO - 10.3390/bioengineering10080895
M3 - SCORING: Journal article
C2 - 37627780
VL - 10
JO - BIOENGINEERING-BASEL
JF - BIOENGINEERING-BASEL
SN - 2306-5354
IS - 8
M1 - 895
ER -