BYOLMed3D: Self-Supervised Representation Learning of Medical Videos using Gradient Accumulation Assisted 3D BYOL Framework

TL;DR

This paper introduces BYOLMed3D, a self-supervised learning framework for medical videos that uses gradient accumulation to handle large batch sizes, outperforming existing methods in ACL tear injury detection.

Contribution

It presents one of the first applications of a 3D BYOL self-supervised model with gradient accumulation in medical video analysis, improving downstream task performance.

Findings

Outperforms existing self-supervised methods in ACL injury detection

Demonstrates robustness to data imbalance in medical videos

Shows superior results compared to Kinetics-400 pre-trained ResNet3D-18

Abstract

Applications on Medical Image Analysis suffer from acute shortage of large volume of data properly annotated by medical experts. Supervised Learning algorithms require a large volumes of balanced data to learn robust representations. Often supervised learning algorithms require various techniques to deal with imbalanced data. Self-supervised learning algorithms on the other hand are robust to imbalance in the data and are capable of learning robust representations. In this work, we train a 3D BYOL self-supervised model using gradient accumulation technique to deal with the large number of samples in a batch generally required in a self-supervised algorithm. To the best of our knowledge, this work is one of the first of its kind in this domain. We compare the results obtained through our experiments in the downstream task of ACL Tear Injury detection with the contemporary self-supervised pre-training methods and also with ResNet3D-18 initialized with the Kinetics-400 pre-trained weights. From the downstream task experiments, it is evident that the proposed framework outperforms the existing baselines.