Self-supervised learning has emerged as one of the best tools for representational learning without using human-annotated data. In this article, we design a self-supervised framework that can effectively classify different knee injuries from magnetic resonance scans. Apart from handling this multilabel classification task, we also highlighted the limitations of the existing geometrical transformation prediction-based pretext tasks, which will help us understand the latent behaviors of the self-supervised learning algorithms.

In medical image analysis, the cost of acquiring high-quality data and annotation by experts is a barrier in many medical applications. Most of the techniques used are based on a supervised learning framework and require a large amount of annotated data to achieve satisfactory performance. As an alternative, in this article, we propose a self-supervised learning approach for learning the spatial anatomical representations from the frames of magnetic resonance (MR) video clips for the diagnosis of knee medical conditions. The pretext model learns meaningful context-invariant spatial representations. The downstream task in our article is a class-imbalanced multilabel classification. Different experiments show that the features learned by the pretext model provide competitive performance in the downstream task. Moreover, the efficiency and reliability of the proposed pretext model in learning representations of minority classes without applying any strategy toward imbalance in the dataset can be seen from the results. To the best of our knowledge, this work is the first of its kind in showing the effectiveness and reliability of self-supervised learning algorithms in imbalanced multilabel classification tasks on MR scans.