A Self supervised learning framework for imbalanced medical imaging datasets

Yash Kumar Sharma; Charan Ramtej Kodi; Vineet Padmanabhan

cs.CVApr 2, 2026

A Self supervised learning framework for imbalanced medical imaging datasets

Authors: Yash Kumar Sharma, Charan Ramtej Kodi, Vineet Padmanabhan

Organizations: Artificial Intelligence lab, School of Computer & Information Sciences, University of Hyderabad, Hyderabad 500046, India

Abstract

Two problems often plague medical imaging analysis: 1) Non-availability of large quantities of labeled training data, and 2) Dealing with imbalanced data, i.e., abundant data are available for frequent classes, whereas data are highly limited for the rare class. Self supervised learning (SSL) methods have been proposed to deal with the first problem to a certain extent, but the issue of investigating the robustness of SSL to imbalanced data has rarely been addressed in the domain of medical image classification. In this work, we make the following contributions: 1) The MIMV method proposed by us in an earlier work is extended with a new augmentation strategy to construct asymmetric multi-image, multi-view (AMIMV) pairs to address both data scarcity and dataset imbalance in medical image classification. 2) We carry out a data analysis to evaluate the robustness of AMIMV under varying degrees of class imbalance in medical imaging . 3) We evaluate eight representative SSL methods in 11 medical imaging datasets (MedMNIST) under long-tailed distributions and limited supervision. Our experimental results on the MedMNIST dataset show an improvement of 4.25% on retinaMNIST, 1.88% on tissueMNIST, and 3.1% on DermaMNIST.

Figures & tables

Panel of Figure 1 — Figure 1: Comparison between AMIMV-SSL (asymmetric multi-image view self-supervised learning) and single multi-view learning.

Dataset	IR	CV	NE	GI	RCR	FB/PI/I
PathMNIST	1.63	0.16	0.99	0.08	8.76	FB
BloodMNIST	2.74	0.41	0.96	0.21	7.09	FB
OCTMNIST	5.93	0.75	0.83	0.35	7.95	FB
BreastMNIST	2.71	0.65	0.84	0.23	26.92	PI
PneumoniaMNIST	2.87	0.68	0.82	0.24	25.78	PI
OrganAMNIST	4.54	0.47	0.95	0.24	3.92	PI

Table 1: Class imbalance analysis of MedMNIST 2D datasets using relative (Imbalance Ratio, Coefficient of Variation, Normalized Entropy, Gini Index) and absolute (Rare Class Ratio) measures. Datasets are categorized as Fairly Balanced (FB), Partially Imbalanced (PI) and Imbalanced (I), considering both distribution skewness and minority class sufficiency.

Dataset	MoCoV3	SimCLR	DINO	ReSSL	BYOL	VICReg	NNCLR	Barlow Twins	Ours
PathMNIST	92.50	92.92	92.03	91.99	93.36	92.32	92.72	92.43	82.59
BloodMNIST	95.72	96.52	82.85	95.49	93.72	93.25	95.05	64.44	91.75
OCTMNIST	79.96	68.82	60.70	78.56	74.94	75.20	79.16	77.28	72.00
BreastMNIST	85.26	82.44	70.38	72.44	73.21	73.46	73.08	79.62	84.61
PneumoniaMNIST	87.34	88.81	87.24	87.21	90.99	91.03	88.24	88.08	89.90
OrganAMNIST	92.63	89.87	91.06	91.89	86.58	89.67	92.56	91.21	93.57

Table 2: Mean accuracy scores of random initialization for self-supervised training using ResNet-50 across different methods and datasets.

Dataset	MoCoV3	SimCLR	DINO	ReSSL	BYOL	VICReg	NNCLR	Barlow Twins	Ours
PathMNIST	99.27	99.45	99.28	99.23	99.44	99.25	99.26	99.38	97.96
BloodMNIST	99.71	99.80	97.53	99.69	99.47	99.47	99.74	90.40	99.28
OCTMNIST	96.81	92.17	90.01	96.07	95.67	96.60	97.11	97.05	96.59
BreastMNIST	80.93	85.56	62.23	51.65	53.18	58.48	63.59	67.44	88.37
PneumoniaMNIST	98.39	94.36	90.29	94.23	97.45	96.50	94.27	92.38	98.19
OrganAMNIST	99.57	97.92	90.49	99.60	98.96	99.21	99.67	99.47	99.73

Table 3: Area under curve(AUC) scores of random initialization for self-supervised training using ResNet-50 across different methods and datasets.

Panel of Figure 2 — Figure 2: Per-class accuracy of AMIMV-SSL on MedMNIST datasets.

Panel of Figure 3 — Figure 3: t-SNE visualizations showing encoder representations that are learned by AMIMV-SSL across eleven MedMNIST datasets. Each subplot represents a distinct dataset, with colors representing ground truth class labels. The tight and distinctly separated clusters indicate strong and discriminative representation learning across various medical imaging modalities and imbalanced class distributions.

Panel of Figure 4 — Figure 4: Confusion matrices of AMIMV-SSL across eleven MedMNIST datasets, including BloodMNIST, BreastMNIST, DermaMNIST, OCTMNIST, PathMNIST, PneumoniaMNIST, RetinaMNIST, TissueMNIST, OrganAMNIST, and OrganCMNIST

Explore similar work

Task-Aligned Self-Supervised Learning for Medical Image Analysis: A Task-Oriented Review with Practical Design Guidelines
May 18, 2026Chathura Wimalasiri, Kishor Nandakishor, Marimuthu PalaniswamiContrastive Learning Self-Supervised Learning
Department of Electrical and Electronic Engineering, University of Melbourne, Parkville, VIC 3052, Australia
Domain-Specific Self-Supervised Representation Learning for Retinal Fundus Classification
Aug 27, 2026Bekzat Nurlanbekova, Fung Fung TingMedical Image Representation Learning Self-Supervised Learning
School of Information Technology Monash University Malaysia, Malaysia
Chaos-SSL: An Attention-Based Self-Supervised Learning Framework with Chaotic Transformation for Medical Image Classification
May 26, 2026Joao Batista FlorindoContrastive Learning Data Augmentation
Institute of Mathematics, Statistics and Scientific Computing, University of Campinas, Rua Sergio Buarque de Holanda, 651, Campinas, Brazil

May 18, 2026J/K move · Enter open · S save

Task-Aligned Self-Supervised Learning for Medical Image Analysis: A Task-Oriented Review with Practical Design Guidelines

Chathura Wimalasiri, Kishor Nandakishor, Marimuthu Palaniswami

Department of Electrical and Electronic Engineering, University of Melbourne, Parkville, VIC 3052, Australia

Contrastive Learning Self-Supervised Learning Medical Image Analysis Medical Image Representation Learning

Open paper AskPDF arXiv

Self-supervised learning (SSL) is increasingly used in medical image analysis to reduce dependence on costly expert annotations by learning transferable representations from unlabeled data. However, SSL performance depends not only on model architecture, but also on whether the pretext task preserves information required by the downstream clinical objective. This review presents a task-oriented synthesis of SSL methods for medical imaging, focusing on how pretext-task design interacts with imaging modality, label availability, and downstream performance. We analyze 75 studies published from 2017 to 2025 and organize them into four paradigms: contrastive learning, non-contrastive and predictive learning, generative and reconstruction-based learning, and hybrid learning. Rather than cataloging methods chronologically, we examine how these paradigms support classification, segmentation, detection, reconstruction, and regression. The evidence suggests that no SSL strategy is universally optimal. Contrastive objectives generally encourage global discriminative representations and are well aligned with classification, but may underrepresent subtle or localized pathology. Spatial prediction, masked modeling, and reconstruction-based objectives better preserve anatomical structure and are often more suitable for segmentation and dense prediction. Hybrid methods can provide balanced representations, although they increase training complexity. Across modalities, SSL is most beneficial in low-label and few-shot regimes, but its effectiveness depends on modality-aware augmentation, pathology-preserving corruption, and clinically meaningful evaluation. We conclude with practical design guidelines and identify open challenges, including pathology-aware pretext tasks, resource-efficient training for high-dimensional data, and standardized evaluation protocols.

Abstract

Explore similar work

Task-Aligned Self-Supervised Learning for Medical Image Analysis: A Task-Oriented Review with Practical Design Guidelines

Domain-Specific Self-Supervised Representation Learning for Retinal Fundus Classification

Chaos-SSL: An Attention-Based Self-Supervised Learning Framework with Chaotic Transformation for Medical Image Classification